The school administration is forcing their children to attend special classes during government holidays and that they are punishing their children if they complain about it. TENKASI: Parents of students studying in an Alangulam-based private school,

Parents say private school in Alangulam conducting special classes for students on govt holidays

LIV golfers will not receive Official World Golf Ranking points, and it seems unlikely the majors will bend their criteria on the fly to give LIV a path to participate. Plus, more notes from Bob ...

LIV Golfers Want a Path to Major Championships, but History Is Not on Their Side

This subscription will allow existing subscribers of The World to access all of our online content, including the E-Editions area. NOTE: To claim your access to the site, you will need to enter ...

Public courses new trend for major championship

In alignment with the National Education Policy 2020, the State Education Department on Thursday unveiled a series of changes in the board examinations for classes 10 and 12, effective from the academic year 2023–24. These decisions aim to enhance the ...

Major changes in board exams for classes 10 and 12

We couldn't find Craft Events in St Columb Major at the moment. Subscribe weekly email newsletter for St Columb Major.

Crafts Events In St Columb Major

In response to the Official World Golf Ranking’s decision to deny LIV Golf world ranking points, Bryson DeChambeau has proposed a system to keep LIV members in the major championships.

Bryson DeChambeau Proposes Major Qualifying System for LIV Golfers

In an interview with PTI, Pradhan said the HECI will have three major roles which are regulation ... and accreditation of programmes and courses. We had set up a committee under Dr Radhakrishnan ...

Higher Education bill in Parliament soon; medical, law colleges won't be under its ambit: Edu Min

In an interview with PTI, Pradhan said the HECI will have three major roles which are regulation ... accreditation of colleges, and accreditation of programmes and courses. We had set up a committee under Dr Radhakrishnan for revamping the NAAC (National ...

Medical, law colleges not be under HECI, funding autonomy to be with administrative ministry: Pradhan

HECI bill to be introduced in Parliament soon; medical, law colleges not to be brought under it: Pradhan

The Design Major aims to nurture global-minded designers ... boundaries and gives students the freedom to proactively select classes that align with their interests and needs.

Design Major at Okinawa Prefectural University of Art

Major Kalshi Classes Pvt

kalshi

At a certain point, it becomes unclear how many notifications are demanding to be clicked on your phone. As you wake, your groggy brain attempts to recall what you tweeted as you dozed off last night. What did you say to earn this ephemeral moment of attention and glory? Last year, I experienced such a moment. I work for Manifold, a prediction markets platform, which hosts questions such as “Will TikTok be forced to sell?” or “When will GPT-5 be released?”. Users will then bet on these questions to earn play money and work their way up the leaderboards. This prices the market and creates accurate probabilities about the future. One night, while scrolling Manifold for content to share on our socials, I stumbled onto a market predicting the discovery of room-temperature superconductors. It gave a 17% chance that a recently published study claiming to create such a superconductor would be replicated. Before sharing it, I searched for other news articles and related tweets to engage with. To my surprise, I found… nothing. We had beaten the news to the story. Over the next week, our market was widely shared and received millions of views. But something interesting was happening: many users were signing up and returning each morning to check the odds, yet most had little interest in trading. On X, Paul Graham wrote, “A couple times a day I check the prediction market on LK-99. It has doubled since its low on July 29. Still only 24%, but it's moving up, not down.” This revealed two crucial insights. First, we had focused too much on proving the accuracy of prediction markets when people were drawn to them for the fast, aggregated data they provided. Second, we realised that most users were less interested in trading and more engaged as viewers, simply wanting to observe market trends. This required a fundamental shift in our product design and marketing strategies. From Academic Experiments to Consumer Product Prediction markets have traditionally been confined to academic experiments and niche applications, often brushed aside as speculative or impractical. However, our journey with Manifold has revealed untapped potential in this space if you can nail the product-market fit. In the last few years, several startups have taken different approaches to making prediction markets more accessible, valuable, and mainstream. Polymarket is the leading crypto platform and offers real-money markets on topics that other platforms, especially those in the U.S., can’t touch—like political elections. This strategy has attracted recent investment of $70 million from big names like Peter Thiel and Vitalik Buterin, with Nate Silver even coming on board as an advisor. With the election cycle heating up, their trading volume increased to over $100 million per month. Even Trump showed interest and regularly shared their odds on his socials, perhaps because Polymarket’s odds favoured him more than the polls. On the other hand, Kalshi is playing the long game by focusing on regulatory compliance and appealing to institutions. Unlike Polymarket, Kalshi allows US customers and has gone through the rigorous process of securing approval from the Commodity Futures Trading Commission (CFTC) to operate legal betting markets on specific economic indicators and events. This approach aligns with their goal of becoming the go-to platform for traditional financial institutions. But Kalshi isn’t stopping there—they’re actively lobbying and even suing the CFTC to expand their offerings to include more popular markets, like elections. Their recent partnership with SIG as a market maker to provide liquidity is a big step toward making Kalshi a trusted, compliant platform for large-scale traders and investors. Meanwhile, Manifold is taking a different route by focusing on accessibility and user engagement. By opting for a play-money model, Manifold sidesteps the regulatory headaches that come with real-money platforms. We are the only site where users can create their own markets allowing us to scale the range of questions being asked compared to other platforms. This user-generated content model ensures that Manifold is always the fastest to respond to breaking news and the questions that come with it. While our markets currently operate on play money, we’re planning to introduce a sweepstakes model that will let users win real money on certain markets. Because of the gamification and reputation at stake, users are still incentivised to trade accurately allowing our markets to be just as good as our competitors. As these three platforms—Polymarket, Kalshi, and Manifold—demonstrate, the path to product-market fit in prediction markets is not a one-size-fits-all journey. That said, all 3 platforms have had to overcome some shared challenges that come with prediction markets. Identifying the challenges with your product’s marketability Every new product faces the challenge of overcoming initial skepticism, especially when it's something unfamiliar. For example, most people have no connotation with the terms “prediction market” or “forecasting platform”. This presents a rare opportunity to create their initial impression but comes with the risk of them comparing it to a gambling platform with a negative connotation. This makes prediction markets arduous to share. Manifold users who tried to share links without framing them properly were barraged with downvotes on Reddit. But this isn’t the fault of the sharers, but rather the product itself and what it communicates about itself when viewed at first glance. There’s an intuition people have that an idea or product can only grow as much as people are willing to share it. I would argue that this isn’t the bottleneck. You have to look one step further. A product’s growth is limited by how easy it is to share. You could have a cult-like following obsessed with an idea and bombarding anyone who dares to listen. Yet, if it is too complex and cannot be presented in a relatable way, it’s not going to grow. And even if you can convey an accurate picture of what a prediction market is, it still may not be obvious to someone what value it presents to them. Luckily, prediction markets are now in a position where enough credible people and news outlets understand them and are willing to leverage their reputation to share them. Presenting them as a way to “follow the odds on something you care about” rather than “play this game/make money betting about something you care about” has helped a lot too. Narrowing your focus on Product-market-fit As we developed our product, we found ourselves constantly pulled in different directions, often sacrificing key use cases. Initially, we over-optimised for traders, believing that product-market fit (PMF) depended on teaching as many people as possible how to trade and making it fun for them. However, this limited approach overlooked a significant segment of users who were more interested in the information prediction markets provided than in trading itself. The superconductor rumours and the ongoing election cycle offered strong examples of this. Many users cared about the probabilities but had little interest in trading. Also, they didn’t need the markets to be perfectly accurate; they just wanted a quick gauge of how events, like Biden potentially dropping out, could impact the race. The speed and convenience of prediction markets filled this gap far better than slow polls could. By making the shift to focus on highlighting our speed and readability, it becomes harder for skeptical potential users to dismiss us. Previously, they could choose to believe we didn’t have enough proof of accuracy and thus weren’t worth anyone’s time. Or that the only reason to interact with prediction markets is if you were a gambler, which made sense as our marketing and copy were solely focused on pushing people to make trades. However, it’s much harder to deny that prediction markets create the fastest, aggregated data that is convenient to read. Manifold’s data still suggests that prediction markets are incredibly accurate and we still care about our traders. But, by focusing on the value we provide in other areas, we aren’t so reliant on winning the accuracy argument or trying to convert everyone to enjoy trading. What’s next for prediction markets? We are entering exciting times for prediction markets. This election cycle will be their first test with significant volume and mainstream attention to see how they perform compared to more traditional forecasting methods. Even more exciting will be seeing how many news outlets and people begin including prediction markets as a data point when forming their own opinions. We are also continuously discovering more advantages prediction markets have that align with what users care about. For example, news sites heavily optimise for reporting breaking news as soon as possible. When Biden dropped out Manifold sent an email 20 minutes before news outlets such as the NYT did to alert our users. Finally, questions remain about prediction markets’ longevity in the US, especially as the CFTC continues its legal battle with Kalshi. How this will play out could have significant implications for the future of the industry. At a certain point, it becomes unclear how many notifications are demanding to be clicked on your phone. As you wake, your groggy brain attempts to recall what you tweeted as you dozed off last night. What did you say to earn this ephemeral moment of attention and glory? Last year, I experienced such a moment. I work for Manifold , a prediction markets platform, which hosts questions such as “Will TikTok be forced to sell?” or “When will GPT-5 be released?”. Users will then bet on these questions to earn play money and work their way up the leaderboards. This prices the market and creates accurate probabilities about the future. Manifold One night, while scrolling Manifold for content to share on our socials, I stumbled onto a market predicting the discovery of room-temperature superconductors. It gave a 17% chance that a recently published study claiming to create such a superconductor would be replicated. Before sharing it, I searched for other news articles and related tweets to engage with. To my surprise, I found… nothing. We had beaten the news to the story. Over the next week, our market was widely shared and received millions of views. But something interesting was happening: many users were signing up and returning each morning to check the odds, yet most had little interest in trading. On X, Paul Graham wrote , “A couple times a day I check the prediction market on LK-99. It has doubled since its low on July 29. Still only 24%, but it's moving up, not down.” On X, Paul Graham wrote This revealed two crucial insights. First, we had focused too much on proving the accuracy of prediction markets when people were drawn to them for the fast, aggregated data they provided. Second, we realised that most users were less interested in trading and more engaged as viewers, simply wanting to observe market trends. This required a fundamental shift in our product design and marketing strategies. From Academic Experiments to Consumer Product Prediction markets have traditionally been confined to academic experiments and niche applications, often brushed aside as speculative or impractical. However, our journey with Manifold has revealed untapped potential in this space if you can nail the product-market fit. In the last few years, several startups have taken different approaches to making prediction markets more accessible, valuable, and mainstream. Polymarket is the leading crypto platform and offers real-money markets on topics that other platforms, especially those in the U.S., can’t touch—like political elections. This strategy has attracted recent investment of $70 million from big names like Peter Thiel and Vitalik Buterin, with Nate Silver even coming on board as an advisor . With the election cycle heating up, their trading volume increased to over $100 million per month . Even Trump showed interest and regularly shared their odds on his socials, perhaps because Polymarket’s odds favoured him more than the polls. Polymarket investment of $70 million Nate Silver even coming on board as an advisor $100 million per month On the other hand, Kalshi is playing the long game by focusing on regulatory compliance and appealing to institutions. Unlike Polymarket, Kalshi allows US customers and has gone through the rigorous process of securing approval from the Commodity Futures Trading Commission (CFTC) to operate legal betting markets on specific economic indicators and events. This approach aligns with their goal of becoming the go-to platform for traditional financial institutions. But Kalshi isn’t stopping there—they’re actively lobbying and even suing the CFTC to expand their offerings to include more popular markets, like elections. Their recent partnership with SIG as a market maker to provide liquidity is a big step toward making Kalshi a trusted, compliant platform for large-scale traders and investors. Kalshi suing the CFTC partnership with SIG Meanwhile, Manifold is taking a different route by focusing on accessibility and user engagement. By opting for a play-money model, Manifold sidesteps the regulatory headaches that come with real-money platforms. We are the only site where users can create their own markets allowing us to scale the range of questions being asked compared to other platforms. This user-generated content model ensures that Manifold is always the fastest to respond to breaking news and the questions that come with it. While our markets currently operate on play money, we’re planning to introduce a sweepstakes model that will let users win real money on certain markets. Because of the gamification and reputation at stake, users are still incentivised to trade accurately allowing our markets to be just as good as our competitors. sweepstakes model As these three platforms—Polymarket, Kalshi, and Manifold—demonstrate, the path to product-market fit in prediction markets is not a one-size-fits-all journey. That said, all 3 platforms have had to overcome some shared challenges that come with prediction markets. Identifying the challenges with your product’s marketability Every new product faces the challenge of overcoming initial skepticism, especially when it's something unfamiliar. For example, most people have no connotation with the terms “prediction market” or “forecasting platform”. This presents a rare opportunity to create their initial impression but comes with the risk of them comparing it to a gambling platform with a negative connotation. This makes prediction markets arduous to share. Manifold users who tried to share links without framing them properly were barraged with downvotes on Reddit. But this isn’t the fault of the sharers, but rather the product itself and what it communicates about itself when viewed at first glance. There’s an intuition people have that an idea or product can only grow as much as people are willing to share it. I would argue that this isn’t the bottleneck. You have to look one step further. A product’s growth is limited by how easy it is to share. You could have a cult-like following obsessed with an idea and bombarding anyone who dares to listen. Yet, if it is too complex and cannot be presented in a relatable way, it’s not going to grow. And even if you can convey an accurate picture of what a prediction market is, it still may not be obvious to someone what value it presents to them. Luckily, prediction markets are now in a position where enough credible people and news outlets understand them and are willing to leverage their reputation to share them. Presenting them as a way to “follow the odds on something you care about” rather than “play this game/make money betting about something you care about” has helped a lot too. Narrowing your focus on Product-market-fit As we developed our product, we found ourselves constantly pulled in different directions, often sacrificing key use cases. Initially, we over-optimised for traders, believing that product-market fit (PMF) depended on teaching as many people as possible how to trade and making it fun for them. However, this limited approach overlooked a significant segment of users who were more interested in the information prediction markets provided than in trading itself. The superconductor rumours and the ongoing election cycle offered strong examples of this. Many users cared about the probabilities but had little interest in trading. Also, they didn’t need the markets to be perfectly accurate; they just wanted a quick gauge of how events, like Biden potentially dropping out, could impact the race. The speed and convenience of prediction markets filled this gap far better than slow polls could. By making the shift to focus on highlighting our speed and readability, it becomes harder for skeptical potential users to dismiss us. Previously, they could choose to believe we didn’t have enough proof of accuracy and thus weren’t worth anyone’s time. Or that the only reason to interact with prediction markets is if you were a gambler, which made sense as our marketing and copy were solely focused on pushing people to make trades. However, it’s much harder to deny that prediction markets create the fastest, aggregated data that is convenient to read. Manifold’s data still suggests that prediction markets are incredibly accurate and we still care about our traders. But, by focusing on the value we provide in other areas, we aren’t so reliant on winning the accuracy argument or trying to convert everyone to enjoy trading. Manifold’s data What’s next for prediction markets? We are entering exciting times for prediction markets. This election cycle will be their first test with significant volume and mainstream attention to see how they perform compared to more traditional forecasting methods. Even more exciting will be seeing how many news outlets and people begin including prediction markets as a data point when forming their own opinions. We are also continuously discovering more advantages prediction markets have that align with what users care about. For example, news sites heavily optimise for reporting breaking news as soon as possible. When Biden dropped out Manifold sent an email 20 minutes before news outlets such as the NYT did to alert our users. Manifold sent an email 20 minutes before news outlets Finally, questions remain about prediction markets’ longevity in the US, especially as the CFTC continues its legal battle with Kalshi. How this will play out could have significant implications for the future of the industry.

Could Prediction Markets Offer Better Odds at Who Wins the US Elections?

¿Podrían los mercados de predicción ofrecer mejores probabilidades sobre quién ganará las elecciones en Estados Unidos?

Os mercados de predición poderían ofrecer mellores probabilidades de quen gañe as eleccións estadounidenses?

予測市場は、米国の選挙で誰が勝利するかについてより良いオッズを提供できるか？

შეუძლია თუ არა პროგნოზირების ბაზრებს უკეთესი შანსები შესთავაზოს ვინ მოიგებს აშშ-ს არჩევნებს?

Урьдчилан таамаглах зах зээл нь АНУ-ын сонгуульд хэн ялах талаар илүү сайн магадлалыг санал болгож чадах уу?

Kunnen voorspellingsmarkten betere kansen bieden op wie de Amerikaanse verkiezingen wint?

¿Predicción qhatukuna aswan allin probabilidadkunata qonmanchu Pikunachus EE.UU.

Mafekitari Mashambadziro Anogona Kupa Zvirinani Odd kune Ndiani Anokunda Sarudzo dzeUS?

Suuqyada Saadaasha Miyay Bixiyaan Khatar Wanaagsan Cidda Ku Guuleysata Doorashooyinka Mareykanka?

ዕዳጋታት ትንበያ ኣብ ምርጫ ኣሜሪካ መን ይዕወት ዝሓሸ ዕድል ከቕርቡ ይኽእሉ ድዮም?

ABŞ-nyň saýlawlarynda kim ýeňiş gazansa, çaklama bazarlary has gowy teklip hödürläp bilermi?

预测市场能否为美国大选的获胜提供更好的预测机会？

If you haven’t been living under a rock (or been on this platform at all), you may (read: will) be aware of the following events: Elon Musk bought Twitter for $44bn. He then renamed it X, a throwback to the X.com domain he has owned since the 90’s and the precursor to PayPal Now that that’s covered, some thoughts on why he did this, where it’s headed and how every user might get involved. Why Twitter? The party line in the lead-up to and ever since the Twitter takeover has been that Twitter is the world’s “digital town square”. Prior to acquisition, some stated concerns that the purchasing party had with the platform were its restrictions on free speech. In an ideal world, the town square is a place for open communication, serendipitous interaction and open commerce. Under the prior regime, Twitter displayed restricted characteristics of the first two features of the town square. My bet in writing this is that the vision for X will try and comprehensively and unapologetically service all 3. Twitter has a number of characteristics that make it uniquely ripe for a platform mission of this scale. It is a unique source of real-time news - no other platforms (with possibly the exception of TikTok) are as uniquely geared to delivering a stream of takes as soon as action happens in the real world. As far as social platforms go, it is a unique source of understanding people’s unconscious thoughts - there is less filter (and substantially more shitposting) on Twitter than other platforms. These two characteristics bear a lot of similarities to WeChat, a platform for which Musk has expressed in the past. Similar to Twitter, WeChat’s origins were in an unfiltered* feed of people’s messages benefitting from ‘gotta-have-this’ network effects. admiration Through this stream of low-effort, low-barrier user generated content, WeChat was able to easily become a place for the breaking of news in real time. These elements created an extremely valuable network of data. The logical next step from here was to introduce the launch of ‘mini-apps’ - platform-native applications built by third-party developers to utilise WeChat’s network to offering augmented possibilities within the platform to compete with, well, every other tech company. X could be a vision of WeChat built for a more independent, more global audience. As such, while the platform structure may remain similar, the apps may look very different. This, in my opinion, is what so many fail to understand about western ‘replications’ of these super-apps being ill-fated (excuse the meme below). What’s been done so far? Since Elon’s takeover, a few defining changes have been made to the platform that may be indicative of longer-term strategic moves. Creators can charge whatever fee they want for their ‘true fans’ to access exclusive content. Creator Subscriptions. A decentralised form of moderation that allows users to provide additional context to potentially misleading Xeets. This Community Notes. Twitter extended the long-form tweet limit to 10,000 characters. It also expanded its video capabilities, as best exemplified by Tucker Carlson’s relocation of to the platform. Extended Content Limits. his content Payouts to creators based on their contribution to platform ad revenue, denominated as a % of ad revenue derived from ads placed in the replies to their Xeets. Ad-sharing agreements. This is the big one. Reducing the prominence of scams and bots is the obvious first-order effect. However, it also could serve a broader economic purpose. The first generation of consumer-ready LLMs benefitted from having free reign over the internet to train their models over whatever data they wanted. This has become the subject of major copyright cases. X’s API restrictions force those looking to scrape the platform for data to pay for the contributions X’s rich dataset makes to their LLMs (or similar), creating a fairer economic model for all involved. The logical next (creator-friendly) step would be, as they have done with ad sharing, to payout network participants for what percentage of their data is being paid for as part of these API charges. API Restrictions & Rate Limits. What’s next? I don’t hold out hope for X to become a fully decentralised protocol in the manner of the Dorsey-era Bluesky initiative. However, given Elon’s ambition starting this foray into social, I do expect it to become platformised. This (fairly aggressive) assumption is based off of two things: 1) Elon’s long-stated ambition to build a super-app of some variety, and 2) the re-application of the X.com domain to open up pursuit of old goals. ‘X-as-a-Platform’ in this sense isn’t referring to you just being able to Xeet whatever you want, whenever you want on the platform (as has always been the case). Instead, I refer to either of these platform definitions of choice: A venue for creating and facilitating interactions between different groups (e.g. buyers and sellers, but not limited to this) A stage from which applications can be built. Based on what we’ve seen and heard so far, a lot of the plumbing will be done by Twitter themselves. Already we have heard of them allude to in-house features that will “allow people to conduct their entire financial world” from within the platform. It has arguably made the greatest leap of any mass-scale social platform in terms of rewarding creators via the ad revenue sharing system introduced this month. As talented as the remaining engineers at Twitter may be, the scope for their delivery of novel products & tools is no match for the widespread creativity and productive powers of As such, I believe that to build the ‘financial world’ of Musk’s imagination, a real platform strategy is in order. Where is the low hanging fruit to be built for this ecosystem? What might be built for the X economy? Novel transaction rails for novel interactions The idea of “conduct(ing) your entire financial world” from within X is a headline grabbing concept. To enable this first and foremost requires rails for processing these financial transactions. Cryptocurrencies (coin-agnostic) offer a perfect manner for doing this in a censorship-resistant, self-custodial and programmable fashion for users on the platform. Payments for creator subscriptions or Reddit-style rewards do not have to be done via some meaningless platform-native token, but rather in the self-custodial currency of choice that users can hold and transact via X. Rather than having to deal with the muddle of SWIFT codes, bank account numbers or even wallet addresses, users will be able to direct transactions simply via account handles. Feature Markets Got a request for feature on Twitter? Say, for example, I want to be able to create debate markets within comments sections, whereby comments are sorted by Agree/Disagree sentiment to a statement tweet. The most liked comments on either side of the argument are presented side-by-side at the top. I want the winner of this debate to be able to extract a reward. Even if it is not a central priority for Twitter to develop internally or rollout across the platform, why shouldn’t the developer be able to deploy these kinds of features in a sandbox environment to a group of opt-in users that are interested in seeing these kinds of i) experimental features or ii) balanced idea markets? After deploying from a developer suite (which could be a paid service based on its potential economic upside), developers stand to win rewards based on any of i) revenues from any features that they create with some kind of financial element or ii) the portion of total Twitter revenue derived from engagement with these features (including ads). makes this idea sound a bit more novel than it really is - the reality is that they are extremely analogical to marketplaces on platforms like WordPress, Shopify or even Google extensions. For whatever reason, such functionality has been limited to date in the world of social. Outside of social, it is the backbone of gargantuan markets: while maybe a bit broader in scope as a platform, Apple’s App Store contributed $86.8bn in platform revenues in 2022. Feature markets IP & Data Licensing Mechanisms Beyond the ad revenue sharing scheme that is in place at the moment, there should be other ways for creators to monetise their content (particularly for creators whose content is ripe for being repurposed elsewhere). Further, in order for developers to benefit from the potential programmability of X’s data, there need to be responsible, legally viable frameworks in place for use of customer data. Arweave’s Universal Data License ( ) represents a good proxy of these kinds of systems in that it allows creators to dictate how any given piece of content can be used or remixed whilst earning rewards when it is used for external purposes. UDL Custom algorithms For all the work that X’s algorithm puts into keeping you delighted, it is inevitable that everyone gets annoyed with their stream. If I have a day where I want to look at golf content, I don’t want to be tagged as wanting to view that for the rest of my life. If I’m using Twitter for professional purposes but got easily distracted by entertainment news every now and then, I don’t want these 5-minute forays to destroy Twitter as a professional tool for the rest of my career. Sure, features like Lists etc all serve to give the user input into what enters their feed, but this then has the effect of limiting the amazing serendipitous post that pops up in your feed that makes Twitter such a great platform. Enter custom algos. Developers can play around and deploy different thematic algorithms for users to test. In one example, developers may able to create a proxy of what any given influencer’s feed may look like. I, for one, would be very curious to see what Elon’s doomscrolling experience looks like compared to mine. At the cross-platform level, this would likely be a read-only function. It is hard to imagine X giving up the reigns to what kinds of content get rewarded or distributed at scale. However, once advanced community tools are built-in, why not allow developers to play with different reward mechanisms for different communities? The average Crypto Twitter user likely gets value out of very different styles of Tweets/replies than, say, someone on Movie Twitter. Once these communities can be effectively delineated, a lot of design space opens up for novel reward mechanisms. To date, Twitter’s own pre-Elon development has been the first and only mover in developing these kinds of ‘algorithm marketplaces’. This will be an interesting space to watch. How big this market could be is dependent on your views towards what portion of the $231bn annual social media market is determined by algorithms versus other factors such as marketing or community. Bluesky Cogsec Apps Ever since the events of 2016, misinformation and informational integrity have been core themes in the social space. How can we better protect people against this kind of misinformation? Do people even want to be protected or have balanced opinions? Creating an open market for cogsec applications that act as a form of ‘cybersecurity’ against misinformation would be a boon to any platform’s claims of being a bastion of free speech. These tools can adopt a variety of different features to de-bias certain pieces of information. Much of this is very speculative, and would require user opt-in to avoid being entirely Orwellian: content filters to repackage tweets with less emotional appeal whilst maintaining integrity of the original message or reputational slashing for accounts that are repetitively tagged as producing misleading content by Community Notes contributors are two examples. Community Notes is actually a step in the right direction as far as cogsec goes on the platform - the most obvious extension here is to allow (and possibly even encourage) users to moderate i) Community Notes themselves and ii) any measures taken by X as a platform to censor certain content on the platform. Beyond this, the idea of rewards for reputable and reliable user contributions to Community Notes seems a likely extension of this first-phase of platform revenue sharing mechanisms. Commerce Infrastructure For most, buying useless shit is a key tenet of the operation of their financial world. As such, ‘arm-the-rebels’ style commerce tools are likely to creep into the X platform (apologies, Shopify). I believe that, first and foremost, we are not far away from centralised features that allow people to sell directly from their profile or embed checkouts within tweets. Better Community Tools Community building on X today is mostly a function of its content algorithm. I like startup content, I fall in with startup people on Twitter. How can this be made more proactive, and how can these small subcultures on the platform create their own mini-apps without having to secede from the platform to places like Discords or Slacks? To the above, I think it is likely that X will enter the work communications space for remote teams. With this comes the opportunity to rebuild the kind of plugins and additional apps seen on Slack, Discord and their ilk from the ground up. Prediction markets & other voting mechanisms In these early days of Elon-era Twitter/X, we have seen many company-level decisions be informed by simple Yes/No polls from his handle. How much further can these kind of trends go across a broader range of potential applications? Prediction markets have been a polarising topic since the first large scale, skin-in-the-game markets were launched around the time of Ethereum (Augur being the first mover in this case, followed by successors such as Polymarket, and to name a few). Manifold Markets Kalshi If these products failed to achieve breakout growth, what makes a prediction market on X any different? The first factor is . The aforementioned existing prediction market platforms are used by small, esoteric audiences seeking gains from supposed informational advantages. X.com has an inbuilt audience of ~330mm monthly active users who are on the platform because they have opinions (or want to find new ones). What better way to prove your value in the market of opinions than creating financially staked markets from it? distribution The second factor is X’s now proven ability to . Similar to the ad revenue sharing scheme, prediction markets features on Twitter could work by allowing users to create their own markets based on assertions they make in their Xeets. If users make sufficiently provocative or challenging assertions, one would assume that the prediction markets for these assertions are particularly hot and might allow the original author to take a share of the market for their provoking thoughts. share revenue with its creators This creates a financial incentive for creators to publish thought-provoking mini-content on the platform, and allows ‘reply guys’ to also get in on the action by providing opposing/supporting evidence to influence their side of the prediction market. In this model, these prediction markets can be organised more organically and allow for second-order pieces of information supporting or opposing the cause to play an influential role in market outcomes. Novel platform analytics One of the things that web3 has done best (and has been best for crypto twitter) is the depth of analytics openly available to anyone in the space. Data from block explorers can be repackaged and contorted into easily readable formats for anyone interested to easier digest exactly what is happening on chain. Nothing like this has achieved success at scale for social. Sure, anyone with the requisite level of skill can develop their own sentiment analysis programs or data visualisation techniques based on what is happening, but platforms don’t make it easy for the everyday user to play around with and explore different elements of platform activity. Why not introduce -style analytics platforms for granular data on social platform activity? At the topline, there would be full transparency around platform activity - for example, how likes/RTs/replies are distributed among power users of the platform or how bot activity is trending over time. Where it gets very interesting is if these platforms can then open-source sentiment analysis tools. In this world, we could see analysts developing publicly available dashboards displaying changes in tone from different account categories over time, common triggers of trends and much much more. Dune

The Economy of Elon Musk's X

ইলন মাস্কের এক্স এর অর্থনীতি

Die Wirtschaft von Elon Musks X

La economía de la X de Elon Musk

L'économie du X d'Elon Musk

एलोन मस्क की एक्स की अर्थव्यवस्था

イーロン・マスクのXの経済

Elon Musk의 X의 경제

A economia do X de Elon Musk

Elon Musk'un X'inin Ekonomisi

Nền kinh tế của Elon Musk's X

埃隆·马斯克 X 的经济

Tealfeed is a knowledge-sharing platform that connects and empowers creators, keeping news informative. It is a superb platform, that some of us use at Hackernoon. In this AMA, we were joined by Tealfeed Co-founder Sourabh Kaushik, we discuss everything Tealfeed, including future ambitions for the platform. You can check out Tealfeed slogging here. This Slogging thread by Jack Boreham, Sourabh Kaushik, Mónica Freitas, Limarc Ambalina and Sara Pinto occurred in slogging's official #amas channel, and has been edited for readability. Jack Boreham Dec 23, 2021, 10:01 AM Hey @channel, please join me in welcoming our next AMA guest, , the founder of Tealfeed. Tealfeed is a knowledge-sharing platform that empowers creators, connects people and keeps user feeds informative. It is a great platform that I personally use myself, and I know many of you also use it. Please feel free to ask Sourabh anything about: What Tealfeed is How he came up with the idea of Tealfeed and its name Informative content on the internet Upcoming plans for the platform How people can join the platform and who is accepted How Tealfeed is monetised What his ambitions are for the platform Other questions regarding Tealfeed Jack Boreham Dec 23, 2021, 10:04 AM Hi , thank you for joining us, a pleasure to have you with us! To start, can you explain a bit about yourself and what your platform Tealfeed is? Sourabh Kaushik Dec 23, 2021, 10:10 AM Through Tealfeed, we are building a knowledge-sharing platform that keeps people’s feeds informative and helps them grow in their personal and professional lives. At the same time, we are building Tealfeed to empower creators by providing them a platform where they could build their own audience, monetize their followers, get paid projects from brands based on their niche, all at one single platform. Sourabh Kaushik Dec 23, 2021, 10:11 AM I and Rohit Garg started working on our idea while we were still in college completing our B.tech education. We initially started with the idea of building a close-knit community where it was easier for people to share their experiences and learnings in the startup field. As more and more people started joining the platform and as we started talking to more of our creators, we realized the need to accommodate more topics on the platform and that gave birth to Tealfeed. So, we launched this platform this year itself in the month of May and currently, we have 7000+ wonderful creators on the platform creating content in the field of business, startups, design, tech, self care, psychology, writing, and what not. Really grateful to our creators for joining us in our vision. Jack Boreham Dec 23, 2021, 10:26 AM Your platform really does that! Having been a member for about a month, I love how it captures audiences for creators! How did you manage to get creator's onboard and market the platform for the average Joe who doesn't create content? Jack Boreham Dec 23, 2021, 10:29 AM How does the monetisation work? In my opinion, the creator often finds it difficult to earn money through their work! It's a tough dog eats dog world. Sourabh Kaushik Dec 23, 2021, 10:56 AM When we started it out, we wanted to seed the platform with creators who are aligned with our vision to make a positive difference through their content. So, Me and Rohit started finding creators who are actively writing articles via their websites/online platforms and reached out to them for a one-to-one discussion. The one-on-one discussions with creators about their goals, aspirations, needs, requirements, and motivation helped us not only in onboarding them as our first set of creators but also helped us understand their pain points. Even today, we try talking to as many creators as we can and understanding their expectations from the platform. This has helped us in deciding some of our upcoming features, for example, we would be soon launching a chat feature to help creators and users build meaningful relationships. Next year, we also plan on introducing more forms of content on the platform including videos, podcasts, short-form posts (currently Tealfeed is an article-only platform). We believe that if we focus more on understanding our creators and users, and if comes at the expense of having a not-so-exponential growth initially, that's completely okay because happy people are the ones who are going to become the advocates for us tomorrow. Sourabh Kaushik Dec 23, 2021, 11:01 AM Speaking about monetisation, July 2022 is when we are planning to introduce it. As you rightly pointed out, creators earning money online is a complicated space and we believe that happens because a single model doesn't work for all. So, here are three things we would be introducing on the platform. Sourabh Kaushik Dec 23, 2021, 11:14 AM Revenue sharing with creators (on boosted/sponsored content) Individual subscription (creators would have an option to put a paywall in front of their content which could be accessed via their paying subscribers) Offering them paid content projects through brands I would explain why we chose these three models as a part of our platform. There are creators who already have their own set of followers, then there are ones who are just starting out or may not have an existing set of followers, and then there are ones who focus more on creating content for brands. All these creators have different requirements, and a single model might not be the best possible way to cater to all of them. Creators can choose to utilise any of these models (or all of them) to start earning on the platform. Our end goal is to build a truly sustainable knowledge-sharing community led by creators. People see creators doing what they love for a living and aspire to follow that path. We are on a mission to turn their dreams into an ambitious career, empowering them with content monetisation tools and a platform to foster lasting communities. Mónica Freitas Dec 23, 2021, 11:14 AM Hi, great to have you here. What would you say are the main themes discussed on Tealfeed? What makes this platform stand out from the crowd? Sourabh Kaushik Dec 23, 2021, 11:36 AM Thank you for the question Mónica Freitas. The most active topics on Tealfeed currently include Technology, Business, Design, Gaming, Blockchain, Personal Growth, Self-care, Psychology. We would always keep the platform and knowledge free on the platform (individual creators can apply paywall in front of some of their content though). How a user onboarding process works on the platform is, whenever someone signs up on the platform, they are being asked about their interests, and their feed is curated based on it. As we keep conversing with our users and creators, we have realised that many of them would want their feed to be about the content from people they follow. So, we actually provide two feeds: Home feed- based on a combination of user’s interests, people they follow, and more such parameters Following feed- based on content from people a user follows Any user is free to seamlessly switch between the two feeds. People in a society behave according to the norms of the society. People tend to do things that provide them rewards and shy away from doing things that might punish them. An online platform is no different. Through Tealfeed, we intend to democratize knowledge and build an online community where not only creators have incentives to create good content but users too have motivation (would be rolling out features in 2022 to reward healthy content consumption habits) to consume content that would, in turn, boost their own personal and professional growth. Limarc Ambalina Dec 23, 2021, 12:08 PM Hey  I've been in contact with your team before so I was happy to see you become a Slogging guest thanks for joining us. Since there are lots of players in this market, how do you differentiate yourselves from other blogging platforms for professionals? Newcomers might ask, how is this different from the mediums, 's and even the tumblrs and blogspots of the world? http://dev.to Jack Boreham Dec 23, 2021, 12:35 PM I agree! I think a client-facing approach works. Having 1-1 discussions is an excellent way to spread the word and get genuine writers on board.  But getting writers on board and having a community engaging with one another are two separate things. That monetisation plan sounds amazing and will help out many creator's. What are the challenges of introducing so many models? How will you overcome them? Sara Pinto Dec 23, 2021, 12:51 PM Hey , glad to have you here! I see there's a lot planned for the platform, that's exciting! Can anyone join Tealfeed and have access to the information there? How's the process to join in? Sourabh Kaushik Dec 23, 2021, 2:31 PM Hi Limarc Ambalina, have always enjoyed your writeups on Tealfeed. Thank you to your wonderful team for inviting me here. The platforms you mentioned are some really wonderful ones catering to the needs of its community in their own unique way. Through Tealfeed, we envision providing creators with a single platform for their multiple needs. You want to host different forms of content, you can do it on Tealfeed. You want to build an online profile that could help you get projects, you can do it here. Want to monetize your content, you can do that too. Want to communicate with your followers and other creators, that would be possible on Tealfeed too. The concept of Creators’ economy has been booming like never before. And, the first step towards helping our creators in creating content that can create a difference in their lives and others, is to simplify the process for them. The reason we have been talking more about creators is that the content has to come first so that we can make the feeds of our users better. We are drowning in information but starving for knowledge. And, if we have to change that, it has to start by empowering creators by making the process easier for them. And, that’s the vision and plans that we share with our creators and users of why they should give Tealfeed a chance. Sourabh Kaushik Dec 23, 2021, 2:44 PM Jack Boreham, when we talk about creators on a global level doing different kinds of jobs and having different goals and expectations from content creation, one of the major challenges is to understand their expectations and understand the patterns. For example, in region A, the idea of freelancing and taking up paid projects might have not been that common as compared to region B. And monetary expectations can also vary depending on the experience and the median scale in a particular region. To take all of these factors into consideration and develop a model that works well for people across different regions is a challenge we are working hard on to ace. And, these regular interactions with users and creators from different parts of the world have helped us gain a deeper understanding of their processes, expectations, motivations, incentives, and a lot of many more such important things. Mónica Freitas Dec 23, 2021, 2:49 PM That's awesome! How did you come up with the name Tealfeed? Sourabh Kaushik Dec 23, 2021, 2:50 PM Thanks for the question Sara Pinto. Yes, Tealfeed is completely free and accessible to all. You can simply visit and click on Get Started to sign up via Google, Facebook, LinkedIn, or Email. You would be asked to choose your interests from a variety of topics and would be provided with recommendations of creators based on your interests. And, that's all you need to do to get started to Tealfeed. If you want to post an article, you can click on "Create" after you sign up and start writing and publishing on Tealfeed. Would like to invite you Sara on the platform to check it out! http://www.tealfeed.com 💚 1 Sourabh Kaushik Dec 23, 2021, 3:03 PM Mónica Freitas, to the founders, a startup is like a baby, and to name a baby is one of the most special moments. So, thanks for asking about it. The colour Teal is a combination of blue and green, and stands as a symbol of open communication and clarity of thoughts. In Tibet, it’s symbolic of the infinity of the sea and sky, while it is the color of truth and faith for Egyptians. So, in our case, we wanted to provide people with a feed that’s an infinite source of knowledge and where people could place their trust in the information they are consuming. And so, we decided to name our platform Tealfeed. Also, the colour Teal gives us a unique brand positioning and brings some uniqueness and freshness. 🔥 1 Mónica Freitas Dec 23, 2021, 3:31 PM That's a beautiful analogy! Love it!  Also, do you know who your primary audience is in terms of demographics? Jack Boreham Dec 23, 2021, 5:35 PM I can imagine it must be tough! But I love the ambition. I personally cannot wait for your platform to start monetisation. Have you thought about developing an app for Tealfeed? Would it help with monetisation as increased accessibility could benefit creators? Limarc Ambalina Dec 23, 2021, 5:43 PM Thanks for your answer one thing I've just personally been wondering is what was your decisioning behind rebranding from Cofounders Town to Tealfeed? You know I've been following you for a while since I know you from those days 😄 Sourabh Kaushik Dec 24, 2021, 11:01 AM Mónica Freitas, we fall more in love with the name Tealfeed when we see people around us relating to it too, glad that you too can relate to it. A large number of our users and creators are from countries like India, USA, UK, Canada, and there are a few more countries which are climbing up the charts in the last few weeks. We have been working on getting the concept of our platform to people from other regions too, and the response from people has been amazing. We have been also witnessing more people in the age group of 18-35 using our platform as compared to the other age groups. 🔥 1 Sourabh Kaushik Dec 24, 2021, 11:17 AM Jack Boreham, yes we are planning to launch the app somewhere mid-year in 2022 itself. An app provides more control and accessibility especially when we talk about building meaningful relationships with one's audience. A full-fledged monetization (pilot-testing would start earlier) and the launch of the app are two important developments that we are planning almost simultaneously. Sourabh Kaushik Dec 24, 2021, 11:25 AM Limarc Ambalina, the ones who have been associated with the platform even before it was named Tealfeed are the ones who have been instrumental and pivotal in making the platform into what it is today, so thanks for your constant support. As I mentioned earlier, we started with building a community where people could share their experiences and learnings about startups and businesses. But, as people from different personal and professional backgrounds started using the platform and started sharing their knowledge from various domains, we started feeling the need to accommodate more topics and give the platform a name that could give everyone a feeling of belongingness and inclusivity. And, that's when we thought of naming the platform Tealfeed! Mónica Freitas Dec 24, 2021, 12:11 PM thank you for your answers. It's really interesting to get these insights. Jack Boreham Dec 24, 2021, 12:43 PM So we have monetization, any future plans/ambitions past that? 😄 Sourabh Kaushik Dec 25, 2021, 7:30 AM Jack Boreham, Ultimately the goal is to build a platform that creates value in people's lives through content and provides the amateur and professional creators the tools that could help them avoid unnecessary hassles around content creation. The idea is to make the journey of becoming a content creator very easy and accessible for anyone and everyone who would want to start sharing their knowledge. People look at successful individuals and aspire to become like them, people dream to become successful one day. We would consider ourselves very privileged if we get to become a part of this journey and help people in achieving their personal and professional goals. This was the idea we started with, this is the idea that drives us today, and this would be the idea that would be the deciding factor in all the features we would be launching on the platform in the future. Jack Boreham Dec 25, 2021, 8:55 AM Thank you, . Super insightful; we can't wait to see the platform grow. That's a wrap on the AMA, any closing thoughts? How can people join/find Tealfeed😄 Sourabh Kaushik Dec 25, 2021, 9:01 AM The platform is available at . It's free for everyone to join, gain knowledge, and start creating content. You need to simply visit the platform and click on Get Started to signup via Google/Email/Linkedin/Facebook. Select your favorite topics, follow the top creators, and you are all set. Looking forward to the suggestions and feedbacks from the people of this wonderful community. http://www.tealfeed.com ❤ 1 Sourabh Kaushik Dec 25, 2021, 9:04 AM Thank you Jack Boreham, Limarc Ambalina, Mónica Freitas, Sara Pinto, and the entire HN team for having me here. It's been a wonderful experience talking about the platform here. A very Merry Christmas to everyone. Stay safe, take care:) ❤ 2

Empowering Creators with Tealfeed Co-founder Sourabh Kaushik

Authors:
(1) Halil Yigit Oksuz, Control Systems Group at Technische Universitat Berlin, Germany and Exzellenzcluster Science of Intelligence, Technische Universitat Berlin, Marchstr. 23, 10587, Berlin, Germany;
(2) Fabio Molinari, Control Systems Group at Technische Universitat Berlin, Germany;
(3) Henning Sprekeler, Exzellenzcluster Science of Intelligence, Technische Universit¨at Berlin, Marchstr. 23, 10587, Berlin, Germany and Modelling Cognitive Processes Group at Technische Universit¨at Berlin, Germany;
(4) Jorg Raisch, Control Systems Group at Technische Universitat Berlin, Germany and Exzellenzcluster Science of Intelligence, Technische Universitat Berlin, Marchstr. 23, 10587, Berlin, Germany. Table of Links Abstract and Introduction Problem Setup Federated fair over-the-air learning (FedAir) Algorithm Convergence Properties Numerical Example Conclusion and References VI. CONCLUSION In this paper, we have introduced the FedFAir algorithm which uses Over-the-Air Computation to carry out efficient decentralized learning while providing fairness and improved performance. We have shown that the FedFAir algorithm converges to an optimal solution of the minimax problem. Furthermore, we have also illustrated our theoretical findings with a numerical example. Future research will include the development of resilient federated learning algorithms when there are malicious agents in the system. REFERENCES [1] B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient learning of deep networks from decentralized data,” in Artificial intelligence and statistics. PMLR, 2017, pp. 1273–1282. [2] V. Smith, C.-K. Chiang, M. Sanjabi, and A. S. Talwalkar, “Federated multi-task learning,” Advances in neural information processing systems, vol. 30, 2017. [3] K. Bonawitz, H. Eichner, W. Grieskamp, D. Huba, A. Ingerman, V. Ivanov, C. Kiddon, J. Konecny, S. Mazzocchi, B. McMahan et al., “Towards federated learning at scale: System design,” Proceedings of Machine Learning and Systems, vol. 1, pp. 374–388, 2019. [4] Q. Yang, Y. Liu, T. Chen, and Y. Tong, “Federated machine learning: Concept and applications,” ACM Transactions on Intelligent Systems and Technology (TIST), vol. 10, no. 2, pp. 1–19, 2019. [5] A. Nedic, “Distributed gradient methods for convex machine learning problems in networks: Distributed optimization,” IEEE Signal Processing Magazine, vol. 37, no. 3, pp. 92–101, 2020. [6] V. P. Chellapandi, A. Upadhyay, A. Hashemi, and S. H. Zak, “On ˙ the convergence of decentralized federated learning under imperfect information sharing,” IEEE Control Systems Letters, 2023. [7] T. Omori and K. Kashima, “Combinatorial optimization approach to client scheduling for federated learning,” IEEE Control Systems Letters, 2023. [8] T. Sery, N. Shlezinger, K. Cohen, and Y. C. Eldar, “Over-the-air federated learning from heterogeneous data,” IEEE Transactions on Signal Processing, vol. 69, pp. 3796–3811, 2021. [9] T. Gafni, N. Shlezinger, K. Cohen, Y. C. Eldar, and H. V. Poor, “Federated learning: A signal processing perspective,” IEEE Signal Processing Magazine, vol. 39, no. 3, pp. 14–41, 2022. [10] M. Ye, X. Fang, B. Du, P. C. Yuen, and D. Tao, “Heterogeneous federated learning: State-of-the-art and research challenges,” ACM Computing Surveys, vol. 56, no. 3, pp. 1–44, 2023. [11] J. Koneˇcn `y, H. B. McMahan, F. X. Yu, P. Richt´arik, A. T. Suresh, and D. Bacon, “Federated learning: Strategies for improving communication efficiency,” arXiv preprint arXiv:1610.05492, 2016. [12] T. Li, A. K. Sahu, A. Talwalkar, and V. Smith, “Federated learning: Challenges, methods, and future directions,” IEEE Signal Processing Magazine, vol. 37, no. 3, pp. 50–60, 2020. [13] P. Kairouz, H. B. McMahan, B. Avent, A. Bellet, M. Bennis, A. N. Bhagoji, K. Bonawitz, Z. Charles, G. Cormode, R. Cummings et al., “Advances and open problems in federated learning,” Foundations and Trends® in Machine Learning, vol. 14, no. 1–2, pp. 1–210, 2021. [14] H. Y. Oksuz, F. Molinari, H. Sprekeler, and J. Raisch, “Federated learning in wireless networks via over-the-air computations,” in 2023 62nd IEEE Conference on Decision and Control (CDC), 2023, pp. 4379–4386. [15] F. Molinari, N. Agrawal, S. Sta´nczak, and J. Raisch, “Max-consensus over fading wireless channels,” IEEE Transactions on Control of Network Systems, vol. 8, no. 2, pp. 791–802, 2021. [16] M. Frey, I. Bjelakovi´c, and S. Sta´nczak, “Over-the-air computation in correlated channels,” IEEE Transactions on Signal Processing, vol. 69, pp. 5739–5755, 2021. [17] K. Yang, T. Jiang, Y. Shi, and Z. Ding, “Federated learning via overthe-air computation,” IEEE Transactions on Wireless Communications, vol. 19, no. 3, pp. 2022–2035, 2020. [18] F. Molinari and J. Raisch, “Exploiting wireless interference for distributively solving linear equations,” IFAC-PapersOnLine, vol. 53, no. 2, pp. 2999–3006, 2020. [19] D. Bertsekas, A. Nedic, and A. Ozdaglar, Convex analysis and optimization. Athena Scientific, 2003, vol. 1. [20] M. Mohri, G. Sivek, and A. T. Suresh, “Agnostic federated learning,” in International Conference on Machine Learning. PMLR, 2019, pp. 4615–4625. [21] Z. Hu, K. Shaloudegi, G. Zhang, and Y. Yu, “Federated learning meets multi-objective optimization,” IEEE Transactions on Network Science and Engineering, vol. 9, no. 4, pp. 2039–2051, 2022. [22] D. P. Bertsekas, “Necessary and sufficient conditions for a penalty method to be exact,” Mathematical programming, vol. 9, no. 1, pp. 87–99, 1975. [23] K. Srivastava, A. Nedi´c, and D. Stipanovi´c, “Distributed min-max optimization in networks,” in 2011 17th International Conference on Digital Signal Processing (DSP). IEEE, 2011, pp. 1–8. [24] R. Ahlswede, “Multi-way communication channels,” in Second International Symposium on Information Theory: Tsahkadsor, Armenia, USSR, Sept. 2-8, 1971, 1973. [25] A. Giridhar and P. Kumar, “Toward a theory of in-network computation in wireless sensor networks,” IEEE Communications Magazine, vol. 44, no. 4, pp. 98–107, apr 2006. [26] F. Molinari, N. Agrawal, S. Sta´nczak, and J. Raisch, “Over-the-air max-consensus in clustered networks adopting half-duplex communication technology,” IEEE Transactions on Control of Network Systems, 2022. [27] W. Rudin et al., Principles of mathematical analysis. McGraw-hill New York, 1976, vol. 3. [28] D. Tse and P. Viswanath, Fundamentals of Wireless Communication. Cambridge University Press, 2012. [29] A. F. Molisch, Wireless communications. John Wiley & Sons, 2012. [30] B. Sklar, “Rayleigh fading channels in mobile digital communication systems. i. characterization,” IEEE Communications magazine, vol. 35, no. 7, pp. 90–100, 1997. [31] R. G. Bartle and D. R. Sherbert, Introduction to real analysis. Wiley New York, 2000. [32] B. T. Polyak, Introduction to optimization. optimization software. Inc., Publications Division, New York, 1987. [33] Y. I. Alber, A. N. Iusem, and M. V. Solodov, “On the projected subgradient method for nonsmooth convex optimization in a hilbert space,” Mathematical Programming, vol. 81, pp. 23–35, 1998. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Halil Yigit Oksuz, Control Systems Group at Technische Universitat Berlin, Germany and Exzellenzcluster Science of Intelligence, Technische Universitat Berlin, Marchstr. 23, 10587, Berlin, Germany; (2) Fabio Molinari, Control Systems Group at Technische Universitat Berlin, Germany; (3) Henning Sprekeler, Exzellenzcluster Science of Intelligence, Technische Universit¨at Berlin, Marchstr. 23, 10587, Berlin, Germany and Modelling Cognitive Processes Group at Technische Universit¨at Berlin, Germany; (4) Jorg Raisch, Control Systems Group at Technische Universitat Berlin, Germany and Exzellenzcluster Science of Intelligence, Technische Universitat Berlin, Marchstr. 23, 10587, Berlin, Germany. Authors: Authors: (1) Halil Yigit Oksuz, Control Systems Group at Technische Universitat Berlin, Germany and Exzellenzcluster Science of Intelligence, Technische Universitat Berlin, Marchstr. 23, 10587, Berlin, Germany; (2) Fabio Molinari, Control Systems Group at Technische Universitat Berlin, Germany; (3) Henning Sprekeler, Exzellenzcluster Science of Intelligence, Technische Universit¨at Berlin, Marchstr. 23, 10587, Berlin, Germany and Modelling Cognitive Processes Group at Technische Universit¨at Berlin, Germany; (4) Jorg Raisch, Control Systems Group at Technische Universitat Berlin, Germany and Exzellenzcluster Science of Intelligence, Technische Universitat Berlin, Marchstr. 23, 10587, Berlin, Germany. Table of Links Abstract and Introduction Abstract and Introduction Problem Setup Problem Setup Federated fair over-the-air learning (FedAir) Algorithm Federated fair over-the-air learning (FedAir) Algorithm Convergence Properties Convergence Properties Numerical Example Numerical Example Conclusion and References Conclusion and References VI. CONCLUSION In this paper, we have introduced the FedFAir algorithm which uses Over-the-Air Computation to carry out efficient decentralized learning while providing fairness and improved performance. We have shown that the FedFAir algorithm converges to an optimal solution of the minimax problem. Furthermore, we have also illustrated our theoretical findings with a numerical example. Future research will include the development of resilient federated learning algorithms when there are malicious agents in the system. REFERENCES [1] B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient learning of deep networks from decentralized data,” in Artificial intelligence and statistics. PMLR, 2017, pp. 1273–1282. [2] V. Smith, C.-K. Chiang, M. Sanjabi, and A. S. Talwalkar, “Federated multi-task learning,” Advances in neural information processing systems, vol. 30, 2017. [3] K. Bonawitz, H. Eichner, W. Grieskamp, D. Huba, A. Ingerman, V. Ivanov, C. Kiddon, J. Konecny, S. Mazzocchi, B. McMahan et al., “Towards federated learning at scale: System design,” Proceedings of Machine Learning and Systems, vol. 1, pp. 374–388, 2019. [4] Q. Yang, Y. Liu, T. Chen, and Y. Tong, “Federated machine learning: Concept and applications,” ACM Transactions on Intelligent Systems and Technology (TIST), vol. 10, no. 2, pp. 1–19, 2019. [5] A. Nedic, “Distributed gradient methods for convex machine learning problems in networks: Distributed optimization,” IEEE Signal Processing Magazine, vol. 37, no. 3, pp. 92–101, 2020. [6] V. P. Chellapandi, A. Upadhyay, A. Hashemi, and S. H. Zak, “On ˙ the convergence of decentralized federated learning under imperfect information sharing,” IEEE Control Systems Letters, 2023. [7] T. Omori and K. Kashima, “Combinatorial optimization approach to client scheduling for federated learning,” IEEE Control Systems Letters, 2023. [8] T. Sery, N. Shlezinger, K. Cohen, and Y. C. Eldar, “Over-the-air federated learning from heterogeneous data,” IEEE Transactions on Signal Processing, vol. 69, pp. 3796–3811, 2021. [9] T. Gafni, N. Shlezinger, K. Cohen, Y. C. Eldar, and H. V. Poor, “Federated learning: A signal processing perspective,” IEEE Signal Processing Magazine, vol. 39, no. 3, pp. 14–41, 2022. [10] M. Ye, X. Fang, B. Du, P. C. Yuen, and D. Tao, “Heterogeneous federated learning: State-of-the-art and research challenges,” ACM Computing Surveys, vol. 56, no. 3, pp. 1–44, 2023. [11] J. Koneˇcn `y, H. B. McMahan, F. X. Yu, P. Richt´arik, A. T. Suresh, and D. Bacon, “Federated learning: Strategies for improving communication efficiency,” arXiv preprint arXiv:1610.05492, 2016. [12] T. Li, A. K. Sahu, A. Talwalkar, and V. Smith, “Federated learning: Challenges, methods, and future directions,” IEEE Signal Processing Magazine, vol. 37, no. 3, pp. 50–60, 2020. [13] P. Kairouz, H. B. McMahan, B. Avent, A. Bellet, M. Bennis, A. N. Bhagoji, K. Bonawitz, Z. Charles, G. Cormode, R. Cummings et al., “Advances and open problems in federated learning,” Foundations and Trends® in Machine Learning, vol. 14, no. 1–2, pp. 1–210, 2021. [14] H. Y. Oksuz, F. Molinari, H. Sprekeler, and J. Raisch, “Federated learning in wireless networks via over-the-air computations,” in 2023 62nd IEEE Conference on Decision and Control (CDC), 2023, pp. 4379–4386. [15] F. Molinari, N. Agrawal, S. Sta´nczak, and J. Raisch, “Max-consensus over fading wireless channels,” IEEE Transactions on Control of Network Systems, vol. 8, no. 2, pp. 791–802, 2021. [16] M. Frey, I. Bjelakovi´c, and S. Sta´nczak, “Over-the-air computation in correlated channels,” IEEE Transactions on Signal Processing, vol. 69, pp. 5739–5755, 2021. [17] K. Yang, T. Jiang, Y. Shi, and Z. Ding, “Federated learning via overthe-air computation,” IEEE Transactions on Wireless Communications, vol. 19, no. 3, pp. 2022–2035, 2020. [18] F. Molinari and J. Raisch, “Exploiting wireless interference for distributively solving linear equations,” IFAC-PapersOnLine, vol. 53, no. 2, pp. 2999–3006, 2020. [19] D. Bertsekas, A. Nedic, and A. Ozdaglar, Convex analysis and optimization. Athena Scientific, 2003, vol. 1. [20] M. Mohri, G. Sivek, and A. T. Suresh, “Agnostic federated learning,” in International Conference on Machine Learning. PMLR, 2019, pp. 4615–4625. [21] Z. Hu, K. Shaloudegi, G. Zhang, and Y. Yu, “Federated learning meets multi-objective optimization,” IEEE Transactions on Network Science and Engineering, vol. 9, no. 4, pp. 2039–2051, 2022. [22] D. P. Bertsekas, “Necessary and sufficient conditions for a penalty method to be exact,” Mathematical programming, vol. 9, no. 1, pp. 87–99, 1975. [23] K. Srivastava, A. Nedi´c, and D. Stipanovi´c, “Distributed min-max optimization in networks,” in 2011 17th International Conference on Digital Signal Processing (DSP). IEEE, 2011, pp. 1–8. [24] R. Ahlswede, “Multi-way communication channels,” in Second International Symposium on Information Theory: Tsahkadsor, Armenia, USSR, Sept. 2-8, 1971, 1973. [25] A. Giridhar and P. Kumar, “Toward a theory of in-network computation in wireless sensor networks,” IEEE Communications Magazine, vol. 44, no. 4, pp. 98–107, apr 2006. [26] F. Molinari, N. Agrawal, S. Sta´nczak, and J. Raisch, “Over-the-air max-consensus in clustered networks adopting half-duplex communication technology,” IEEE Transactions on Control of Network Systems, 2022. [27] W. Rudin et al., Principles of mathematical analysis. McGraw-hill New York, 1976, vol. 3. [28] D. Tse and P. Viswanath, Fundamentals of Wireless Communication. Cambridge University Press, 2012. [29] A. F. Molisch, Wireless communications. John Wiley & Sons, 2012. [30] B. Sklar, “Rayleigh fading channels in mobile digital communication systems. i. characterization,” IEEE Communications magazine, vol. 35, no. 7, pp. 90–100, 1997. [31] R. G. Bartle and D. R. Sherbert, Introduction to real analysis. Wiley New York, 2000. [32] B. T. Polyak, Introduction to optimization. optimization software. Inc., Publications Division, New York, 1987. [33] Y. I. Alber, A. N. Iusem, and M. V. Solodov, “On the projected subgradient method for nonsmooth convex optimization in a hilbert space,” Mathematical Programming, vol. 81, pp. 23–35, 1998. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Boosting Fairness and Robustness in Over-the-Air Federated Learning: Conclusion and References

Authors:
(1) Divyanshu Kumar, Enkrypt AI;
(2) Anurakt Kumar, Enkrypt AI;
(3) Sahil Agarwa, Enkrypt AI;
(4) Prashanth Harshangi, Enkrypt AI. Table of Links Abstract and 1 Introduction 2 Problem Formulation and Experiments 3 Experiment Set-up & Results 4 Conclusion and References A. Appendix 4 CONCLUSION Our work investigates the LLM’s safety against Jailbreak attempts. We have demonstrated how fine-tuned and quantized models are vulnerable to jailbreak attempts and stress the importance of using external guardrails to reduce this risk. Fine-tuning or quantizing model weights alters the risk profile of LLMs, potentially undermining the safety alignment established through RLHF. This could result from catastrophic forgetting, where LLMs lose memory of safety protocols, or the finetuning process shifting the model’s focus to new topics at the expense of existing safety measures. The lack of safety measures in these fine-tuned and quantized models is concerning, highlighting the need to incorporate safety protocols during the fine-tuning process. We propose using these tests as part of a CI/CD stress test before deploying the model. The effectiveness of guardrails in preventing jailbreaking highlights the importance of integrating them with safety practices in AI development. This approach not only enhances AI models but also establishes a new standard for responsible AI development. By ensuring that AI advancements prioritize innovation and safety, we promote ethical AI deployment, safeguarding against potential misuse and fostering a secure digital future. REFERENCES Zifan Wang Andy Zou. AdvBench Dataset, July 2023. URL https://github.com/ llm-attacks/llm-attacks/tree/main/data. Patrick Chao, Alexander Robey, Edgar Dobriban, Hamed Hassani, George J. Pappas, and Eric Wong. Jailbreaking Black Box Large Language Models in Twenty Queries. arXiv, October 2023. doi: 10.48550/arXiv.2310.08419. Edoardo Debenedetti, Giorgio Severi, Nicholas Carlini, Christopher A. Choquette-Choo, Matthew Jagielski, Milad Nasr, Eric Wallace, and Florian Tramer. Privacy Side Channels in Machine ` Learning Systems. arXiv, September 2023. doi: 10.48550/arXiv.2309.05610. Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, and Luke Zettlemoyer. QLoRA: Efficient Finetuning of Quantized LLMs. arXiv, May 2023. doi: 10.48550/arXiv.2305.14314. Micah Gorsline, James Smith, and Cory Merkel. On the Adversarial Robustness of Quantized Neural Networks. arXiv, May 2021. doi: 10.1145/3453688.3461755. Kai Greshake, Sahar Abdelnabi, Shailesh Mishra, Christoph Endres, Thorsten Holz, and Mario Fritz. Not what you’ve signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection. arXiv, February 2023. doi: 10.48550/arXiv.2302.12173. Bing He, Mustaque Ahamad, and Srijan Kumar. PETGEN: Personalized Text Generation Attack on Deep Sequence Embedding-based Classification Models. In KDD ’21: Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining, pp. 575–584. Association for Computing Machinery, New York, NY, USA, August 2021. ISBN 978-1-45038332-5. doi: 10.1145/3447548.3467390. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-Rank Adaptation of Large Language Models. arXiv, June 2021. doi: 10.48550/arXiv.2106.09685. Shuhei Kashiwamura, Ayaka Sakata, and Masaaki Imaizumi. Effect of Weight Quantization on Learning Models by Typical Case Analysis. arXiv, January 2024. doi: 10.48550/arXiv.2401. 17269. Siwon Kim, Sangdoo Yun, Hwaran Lee, Martin Gubri, Sungroh Yoon, and Seong Joon Oh. ProPILE: Probing Privacy Leakage in Large Language Models. arXiv, July 2023. doi: 10.48550/arXiv. 2307.01881. Aounon Kumar, Chirag Agarwal, Suraj Srinivas, Aaron Jiaxun Li, Soheil Feizi, and Himabindu Lakkaraju. Certifying LLM Safety against Adversarial Prompting. arXiv, September 2023. doi: 10.48550/arXiv.2309.02705. Thai Le, Suhang Wang, and Dongwon Lee. MALCOM: Generating Malicious Comments to Attack Neural Fake News Detection Models. IEEE Computer Society, November 2020. ISBN 978-1- 7281-8316-9. doi: 10.1109/ICDM50108.2020.00037. Yi Liu, Gelei Deng, Zhengzi Xu, Yuekang Li, Yaowen Zheng, Ying Zhang, Lida Zhao, Tianwei Zhang, and Yang Liu. Jailbreaking ChatGPT via Prompt Engineering: An Empirical Study. arXiv, May 2023. doi: 10.48550/arXiv.2305.13860. Anay Mehrotra, Manolis Zampetakis, Paul Kassianik, Blaine Nelson, Hyrum Anderson, Yaron Singer, and Amin Karbasi. Tree of Attacks: Jailbreaking Black-Box LLMs Automatically. arXiv, December 2023. doi: 10.48550/arXiv.2312.02119. Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Christiano, Jan Leike, and Ryan Lowe. Training language models to follow instructions with human feedback. arXiv, March 2022. doi: 10.48550/arXiv.2203.02155. Xiangyu Qi, Yi Zeng, Tinghao Xie, Pin-Yu Chen, Ruoxi Jia, Prateek Mittal, and Peter Henderson. Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! arXiv, October 2023. doi: 10.48550/arXiv.2310.03693. Traian Rebedea, Razvan Dinu, Makesh Narsimhan Sreedhar, Christopher Parisien, and Jonathan Cohen. NeMo Guardrails: A Toolkit for Controllable and Safe LLM Applications with Programmable Rails. ACL Anthology, pp. 431–445, December 2023. doi: 10.18653/v1/2023. emnlp-demo.40. Alexander Wei, Nika Haghtalab, and Jacob Steinhardt. Jailbroken: How Does LLM Safety Training Fail? arXiv, July 2023. doi: 10.48550/arXiv.2307.02483. Martin Weyssow, Xin Zhou, Kisub Kim, David Lo, and Houari Sahraoui. Exploring ParameterEfficient Fine-Tuning Techniques for Code Generation with Large Language Models. arXiv, August 2023. doi: 10.48550/arXiv.2308.10462. Yisong Xiao, Aishan Liu, Tianyuan Zhang, Haotong Qin, Jinyang Guo, and Xianglong Liu. RobustMQ: Benchmarking Robustness of Quantized Models. arXiv, August 2023. doi: 10.48550/ arXiv.2308.02350. Andy Zhou, Bo Li, and Haohan Wang. Robust Prompt Optimization for Defending Language Models Against Jailbreaking Attacks. arXiv, January 2024. doi: 10.48550/arXiv.2401.17263. Sicheng Zhu, Ruiyi Zhang, Bang An, Gang Wu, Joe Barrow, Zichao Wang, Furong Huang, Ani Nenkova, and Tong Sun. AutoDAN: Interpretable Gradient-Based Adversarial Attacks on Large Language Models. arXiv, October 2023. doi: 10.48550/arXiv.2310.15140. Andy Zou, Zifan Wang, Nicholas Carlini, Milad Nasr, J. Zico Kolter, and Matt Fredrikson. Universal and Transferable Adversarial Attacks on Aligned Language Models. arXiv, July 2023. doi: 10.48550/arXiv.2307.15043. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Divyanshu Kumar, Enkrypt AI; (2) Anurakt Kumar, Enkrypt AI; (3) Sahil Agarwa, Enkrypt AI; (4) Prashanth Harshangi, Enkrypt AI. Authors: Authors: (1) Divyanshu Kumar, Enkrypt AI; (2) Anurakt Kumar, Enkrypt AI; (3) Sahil Agarwa, Enkrypt AI; (4) Prashanth Harshangi, Enkrypt AI. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Problem Formulation and Experiments 2 Problem Formulation and Experiments 3 Experiment Set-up & Results 3 Experiment Set-up & Results 4 Conclusion and References 4 Conclusion and References A. Appendix A. Appendix 4 CONCLUSION Our work investigates the LLM’s safety against Jailbreak attempts. We have demonstrated how fine-tuned and quantized models are vulnerable to jailbreak attempts and stress the importance of using external guardrails to reduce this risk. Fine-tuning or quantizing model weights alters the risk profile of LLMs, potentially undermining the safety alignment established through RLHF. This could result from catastrophic forgetting, where LLMs lose memory of safety protocols, or the finetuning process shifting the model’s focus to new topics at the expense of existing safety measures. The lack of safety measures in these fine-tuned and quantized models is concerning, highlighting the need to incorporate safety protocols during the fine-tuning process. We propose using these tests as part of a CI/CD stress test before deploying the model. The effectiveness of guardrails in preventing jailbreaking highlights the importance of integrating them with safety practices in AI development. This approach not only enhances AI models but also establishes a new standard for responsible AI development. By ensuring that AI advancements prioritize innovation and safety, we promote ethical AI deployment, safeguarding against potential misuse and fostering a secure digital future. REFERENCES Zifan Wang Andy Zou. AdvBench Dataset, July 2023. URL https://github.com/ llm-attacks/llm-attacks/tree/main/data. Patrick Chao, Alexander Robey, Edgar Dobriban, Hamed Hassani, George J. Pappas, and Eric Wong. Jailbreaking Black Box Large Language Models in Twenty Queries. arXiv, October 2023. doi: 10.48550/arXiv.2310.08419. Edoardo Debenedetti, Giorgio Severi, Nicholas Carlini, Christopher A. Choquette-Choo, Matthew Jagielski, Milad Nasr, Eric Wallace, and Florian Tramer. Privacy Side Channels in Machine ` Learning Systems. arXiv, September 2023. doi: 10.48550/arXiv.2309.05610. Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, and Luke Zettlemoyer. QLoRA: Efficient Finetuning of Quantized LLMs. arXiv, May 2023. doi: 10.48550/arXiv.2305.14314. Micah Gorsline, James Smith, and Cory Merkel. On the Adversarial Robustness of Quantized Neural Networks. arXiv, May 2021. doi: 10.1145/3453688.3461755. Kai Greshake, Sahar Abdelnabi, Shailesh Mishra, Christoph Endres, Thorsten Holz, and Mario Fritz. Not what you’ve signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection. arXiv, February 2023. doi: 10.48550/arXiv.2302.12173. Bing He, Mustaque Ahamad, and Srijan Kumar. PETGEN: Personalized Text Generation Attack on Deep Sequence Embedding-based Classification Models. In KDD ’21: Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining, pp. 575–584. Association for Computing Machinery, New York, NY, USA, August 2021. ISBN 978-1-45038332-5. doi: 10.1145/3447548.3467390. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-Rank Adaptation of Large Language Models. arXiv, June 2021. doi: 10.48550/arXiv.2106.09685. Shuhei Kashiwamura, Ayaka Sakata, and Masaaki Imaizumi. Effect of Weight Quantization on Learning Models by Typical Case Analysis. arXiv, January 2024. doi: 10.48550/arXiv.2401. 17269. Siwon Kim, Sangdoo Yun, Hwaran Lee, Martin Gubri, Sungroh Yoon, and Seong Joon Oh. ProPILE: Probing Privacy Leakage in Large Language Models. arXiv, July 2023. doi: 10.48550/arXiv. 2307.01881. Aounon Kumar, Chirag Agarwal, Suraj Srinivas, Aaron Jiaxun Li, Soheil Feizi, and Himabindu Lakkaraju. Certifying LLM Safety against Adversarial Prompting. arXiv, September 2023. doi: 10.48550/arXiv.2309.02705. Thai Le, Suhang Wang, and Dongwon Lee. MALCOM: Generating Malicious Comments to Attack Neural Fake News Detection Models. IEEE Computer Society, November 2020. ISBN 978-1- 7281-8316-9. doi: 10.1109/ICDM50108.2020.00037. Yi Liu, Gelei Deng, Zhengzi Xu, Yuekang Li, Yaowen Zheng, Ying Zhang, Lida Zhao, Tianwei Zhang, and Yang Liu. Jailbreaking ChatGPT via Prompt Engineering: An Empirical Study. arXiv, May 2023. doi: 10.48550/arXiv.2305.13860. Anay Mehrotra, Manolis Zampetakis, Paul Kassianik, Blaine Nelson, Hyrum Anderson, Yaron Singer, and Amin Karbasi. Tree of Attacks: Jailbreaking Black-Box LLMs Automatically. arXiv, December 2023. doi: 10.48550/arXiv.2312.02119. Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Christiano, Jan Leike, and Ryan Lowe. Training language models to follow instructions with human feedback. arXiv, March 2022. doi: 10.48550/arXiv.2203.02155. Xiangyu Qi, Yi Zeng, Tinghao Xie, Pin-Yu Chen, Ruoxi Jia, Prateek Mittal, and Peter Henderson. Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! arXiv, October 2023. doi: 10.48550/arXiv.2310.03693. Traian Rebedea, Razvan Dinu, Makesh Narsimhan Sreedhar, Christopher Parisien, and Jonathan Cohen. NeMo Guardrails: A Toolkit for Controllable and Safe LLM Applications with Programmable Rails. ACL Anthology, pp. 431–445, December 2023. doi: 10.18653/v1/2023. emnlp-demo.40. Alexander Wei, Nika Haghtalab, and Jacob Steinhardt. Jailbroken: How Does LLM Safety Training Fail? arXiv, July 2023. doi: 10.48550/arXiv.2307.02483. Martin Weyssow, Xin Zhou, Kisub Kim, David Lo, and Houari Sahraoui. Exploring ParameterEfficient Fine-Tuning Techniques for Code Generation with Large Language Models. arXiv, August 2023. doi: 10.48550/arXiv.2308.10462. Yisong Xiao, Aishan Liu, Tianyuan Zhang, Haotong Qin, Jinyang Guo, and Xianglong Liu. RobustMQ: Benchmarking Robustness of Quantized Models. arXiv, August 2023. doi: 10.48550/ arXiv.2308.02350. Andy Zhou, Bo Li, and Haohan Wang. Robust Prompt Optimization for Defending Language Models Against Jailbreaking Attacks. arXiv, January 2024. doi: 10.48550/arXiv.2401.17263. Sicheng Zhu, Ruiyi Zhang, Bang An, Gang Wu, Joe Barrow, Zichao Wang, Furong Huang, Ani Nenkova, and Tong Sun. AutoDAN: Interpretable Gradient-Based Adversarial Attacks on Large Language Models. arXiv, October 2023. doi: 10.48550/arXiv.2310.15140. Andy Zou, Zifan Wang, Nicholas Carlini, Milad Nasr, J. Zico Kolter, and Matt Fredrikson. Universal and Transferable Adversarial Attacks on Aligned Language Models. arXiv, July 2023. doi: 10.48550/arXiv.2307.15043. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

Increased LLM Vulnerabilities from Fine-tuning and Quantization: Conclusion and References

Authors:
(1) Divyanshu Kumar, Enkrypt AI;
(2) Anurakt Kumar, Enkrypt AI;
(3) Sahil Agarwa, Enkrypt AI;
(4) Prashanth Harshangi, Enkrypt AI. Table of Links Abstract and 1 Introduction 2 Problem Formulation and Experiments 3 Experiment Set-up & Results 4 Conclusion and References A. Appendix 3 EXPERIMENT SET-UP & RESULTS The following section outlines the testing environments utilized for various LLMs. The LLMs are tested under three different downstream tasks: (1) fine-tuning, (2) quantization, and (3) guardrails. They are chosen to cover most of the LLM practical use cases and applications in the industry and academia. For TAP configuration, as mentioned before, we use GPT-3.5-turbo as the attack model, and GPT-4-turbo as the judge model. We employ Anyscale endpoints, the OpenAI API, and HuggingFace for our target model. Further information regarding the model and its sources is available in appendix A.1. The details about different settings are mentioned below : • Fine-tuning: Fine-tuning LLMs on different tasks increases their effectiveness in completing the tasks as it incorporates the specialized domain knowledge needed; these can include SQL code generation, chat, and more. We compare the jailbreaking vulnerability of foundational models compared to their fine tune versions. This helps us understand the role of fine-tuning in increasing or decreasing the vulnerability of LLMs and the strategies to mitigate this risk Weyssow et al. (2023). We use foundation models such as Llama2, Mistral, and MPT-7B and their fine-tuned versions such as CodeLlama, SQLCoder, Dolphin, and Intel Neural Chat. The details of the models and versions are specified in the appendix. From the table 1, we empirically conclude that fine-tuned models lose their safety alignment and are jailbroken quite easily compared to the foundational models. • Quantization: Many models require huge computational resources during training, finetuning, and even during inference Hu et al. (2021). Quantization is one of the most popular ways to reduce the computational burden, but it comes at the cost of the numerical precision of model parameters. The quantized models we evaluate below use GPT-Generated Unified Format (GGUF) for quantization, which involves scaling down model weights (stored in 16-bit floating point numbers) to save computational resources at the cost of numerical precision of the model parameters. Kashiwamura et al. (2024) Gorsline et al. (2021) Xiao et al.(2023) Hu et al. (2021) Dettmers et al. (2023). The table 2 demonstrates that quantization of the model renders it susceptible to vulnerabilities. • Guardrails: Guardrails act as a line of defence against LLM attacks. Their primary function is to meticulously filter out prompts that could potentially lead to harmful or malicious outcomes, preventing such prompts from reaching the LLM as an instruction Rebedea et al. (2023). This proactive approach not only promotes the safe utilization of LLMs but also facilitates their optimal performance, thereby maximizing their potential benefits in various domains Kumar et al. (2023) Wei et al. (2023) Zhou et al. (2024). We use our proprietary jailbreak attack detector derived from Deberta-V3 models, trained on harmful prompts generated to jailbreak LLMs. You can reach out to the authors to get more details on this model. From table 3, we observe that the introduction of guardrails as a pre-step has a significant effect and can mitigate jailbreaking attempts by a considerable margin. The results from tables 1, 2, and 3 conclusively show the vulnerability of LLMs post fine-tuning and quantization, and effectiveness of external guardrails in mitigating this vulnerability. The detailed experimental results can be found in appendix A.2. Additionally, this section provides a comprehensive analysis of the experiments conducted, offering insights into the various aspects of the research findings. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Divyanshu Kumar, Enkrypt AI; (2) Anurakt Kumar, Enkrypt AI; (3) Sahil Agarwa, Enkrypt AI; (4) Prashanth Harshangi, Enkrypt AI. Authors: Authors: (1) Divyanshu Kumar, Enkrypt AI; (2) Anurakt Kumar, Enkrypt AI; (3) Sahil Agarwa, Enkrypt AI; (4) Prashanth Harshangi, Enkrypt AI. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Problem Formulation and Experiments 2 Problem Formulation and Experiments 3 Experiment Set-up & Results 3 Experiment Set-up & Results 4 Conclusion and References 4 Conclusion and References A. Appendix A. Appendix 3 EXPERIMENT SET-UP & RESULTS The following section outlines the testing environments utilized for various LLMs. The LLMs are tested under three different downstream tasks: (1) fine-tuning, (2) quantization, and (3) guardrails. They are chosen to cover most of the LLM practical use cases and applications in the industry and academia. For TAP configuration, as mentioned before, we use GPT-3.5-turbo as the attack model, and GPT-4-turbo as the judge model. We employ Anyscale endpoints, the OpenAI API, and HuggingFace for our target model. Further information regarding the model and its sources is available in appendix A.1. The details about different settings are mentioned below : • Fine-tuning: Fine-tuning LLMs on different tasks increases their effectiveness in completing the tasks as it incorporates the specialized domain knowledge needed; these can include SQL code generation, chat, and more. We compare the jailbreaking vulnerability of foundational models compared to their fine tune versions. This helps us understand the role of fine-tuning in increasing or decreasing the vulnerability of LLMs and the strategies to mitigate this risk Weyssow et al. (2023). We use foundation models such as Llama2, Mistral, and MPT-7B and their fine-tuned versions such as CodeLlama, SQLCoder, Dolphin, and Intel Neural Chat. The details of the models and versions are specified in the appendix. From the table 1, we empirically conclude that fine-tuned models lose their safety alignment and are jailbroken quite easily compared to the foundational models. • Fine-tuning: • Quantization : Many models require huge computational resources during training, finetuning, and even during inference Hu et al. (2021). Quantization is one of the most popular ways to reduce the computational burden, but it comes at the cost of the numerical precision of model parameters. The quantized models we evaluate below use GPT-Generated Unified Format (GGUF) for quantization, which involves scaling down model weights (stored in 16-bit floating point numbers) to save computational resources at the cost of numerical precision of the model parameters. Kashiwamura et al. (2024) Gorsline et al. (2021) Xiao et al.(2023) Hu et al. (2021) Dettmers et al. (2023). The table 2 demonstrates that quantization of the model renders it susceptible to vulnerabilities. Quantization • Guardrails : Guardrails act as a line of defence against LLM attacks. Their primary function is to meticulously filter out prompts that could potentially lead to harmful or malicious outcomes, preventing such prompts from reaching the LLM as an instruction Rebedea et al. (2023). This proactive approach not only promotes the safe utilization of LLMs but also facilitates their optimal performance, thereby maximizing their potential benefits in various domains Kumar et al. (2023) Wei et al. (2023) Zhou et al. (2024). We use our proprietary jailbreak attack detector derived from Deberta-V3 models, trained on harmful prompts generated to jailbreak LLMs. You can reach out to the authors to get more details on this model. From table 3, we observe that the introduction of guardrails as a pre-step has a significant effect and can mitigate jailbreaking attempts by a considerable margin. Guardrails The results from tables 1, 2, and 3 conclusively show the vulnerability of LLMs post fine-tuning and quantization, and effectiveness of external guardrails in mitigating this vulnerability. The detailed experimental results can be found in appendix A.2. Additionally, this section provides a comprehensive analysis of the experiments conducted, offering insights into the various aspects of the research findings. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

Increased LLM Vulnerabilities from Fine-tuning and Quantization: Experiment Set-up & Results

Authors:
(1) Divyanshu Kumar, Enkrypt AI;
(2) Anurakt Kumar, Enkrypt AI;
(3) Sahil Agarwa, Enkrypt AI;
(4) Prashanth Harshangi, Enkrypt AI. Table of Links Abstract and 1 Introduction 2 Problem Formulation and Experiments 3 Experiment Set-up & Results 4 Conclusion and References A. Appendix ABSTRACT Large Language Models (LLMs) have become very popular and have found use cases in many domains, such as chatbots, auto-task completion agents, and much more. However, LLMs are vulnerable to different types of attacks, such as jailbreaking, prompt injection attacks, and privacy leakage attacks. Foundational LLMs undergo adversarial and alignment training to learn not to generate malicious and toxic content. For specialized use cases, these foundational LLMs are subjected to fine-tuning or quantization for better performance and efficiency. We examine the impact of downstream tasks such as fine-tuning and quantization on LLM vulnerability. We test foundation models like Mistral, Llama, MosaicML, and their fine-tuned versions. Our research shows that fine-tuning and quantization reduces jailbreak resistance significantly, leading to increased LLM vulnerabilities. Finally, we demonstrate the utility of external guardrails in reducing LLM vulnerabilities. 1 INTRODUCTION Generative models are becoming more and more important as they are becoming capable of automating a lot of tasks, taking autonomous actions and decisions, and at the same time, becoming better at content generation and summarization tasks. As LLMs become more powerful, these capabilities are at risk of being misused by an adversary, which can lead to fake content generation, toxic, malicious, or hateful content generation, privacy leakages, copyrighted content generation, and much more Chao et al. (2023) Mehrotra et al. (2023) Zou et al. (2023) Greshake et al. (2023) Liu et al. (2023) Zhu et al. (2023) He et al. (2021) Le et al. (2020). To prevent LLMs from generating content that contradicts human values and to prevent their malicious misuse, they undergo a supervised fine-tuning phase after their pre-training, and they are further evaluated by humans and trained using reinforcement learning from human feedback (RLHF) Ouyang et al. (2022), to make them more aligned with human values. Further, special filters called guardrails are put in place as filters to prevent LLMs from getting toxic prompts as inputs and outputting toxic or copyrighted responses Rebedea et al. (2023) Kumar et al. (2023) Wei et al. (2023) Zhou et al. (2024). The complexity of human language makes it difficult for LLMs to completely understand what instructions are right and which are wrong in terms of human values. After going through the alignment training and after the implementation of guardrails, it becomes unlikely that the LLM will generate a toxic response. But these safety measures can easily be circumvented using adversarial attacks, and the LLM can be jailbroken to generate any content that the adversary wants, as shown in recent works Chao et al. (2023) Mehrotra et al. (2023) Zhu et al. (2023). Recent works such as the Prompt Automatic Iterative Refinement (PAIR) attacks Chao et al. (2023) and Tree-of-attacks pruning (TAP) Mehrotra et al. (2023) have shown the vulnerability of LLMs and how easy it is to jailbreak them into generating content for harmful tasks specified by the user. Similarly, a class of methods called privacy leakage attacks Debenedetti et al. (2023) are used to attack LLMs to extract their training data or personally identifiable information Kim et al. (2023), and prompt injection attacks can be used to make an LLM application perform tasks that are not requested by the user but are hidden in the third-party instruction which the LLM automatically executes. Figure 1 shows how an instruction can be hidden inside a summarization text and how the LLM will ignore the previous instruction to execute the malicious instruction. Qi et al. (2023) showed that it only takes a few examples to fine-tune an LLM into generating toxic responses by forgetting its safety training. Our work in this paper extends that notion and shows that both finetuning the LLM on any task (not necessarily toxic content generation) and quantization can affect its safety training. In this study, we use a subset of adversarial harmful prompts called AdvBench SubsetAndy Zou (2023). It contains 50 prompts asking for harmful information across 32 categories. It is a subset of prompts from the harmful behaviours dataset in the AdvBench benchmark selected to cover a diverse range of harmful prompts. The attacking algorithm used is tree-of-attacks pruning Mehrotra et al. (2023) as it has shown to have the best performance in jailbreaking and, more importantly, this algorithm fulfils three important goals (1) Black-box: the algorithm only needs black-box access to the model (2) Automatic: it does not need human intervention once started, and (3) Interpretable: the algorithm generates semantically meaningful prompts. The TAP algorithm is used with the tasks from the AdvBench subset to attack the target LLMs in different settings, and their response is used to evaluate whether or not they have been jailbroken. The rest of the paper is organized in the following manner. Section 2 talks about the experimental setup for jailbreaking in which the models are tested. It specifically describes the different modes of downstream process that an LLM has undergone e.g fine-tuning, quantization and tested for these modes. Section 3 describes the experiment set-up used and defines guardrails Rebedea et al. (2023), fine-tuning and quantization Kashiwamura et al. (2024) Gorsline et al. (2021) Xiao et al. (2023) Hu et al. (2021) Dettmers et al. (2023) settings used in the experimental context. We demonstrate the results in detail and show how model vulnerability is affected by downstream tasks for LLMs. Finally, section 4 concludes the study and talks about methods to reduce model vulnerability and ensure safe and reliable LLM development. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Divyanshu Kumar, Enkrypt AI; (2) Anurakt Kumar, Enkrypt AI; (3) Sahil Agarwa, Enkrypt AI; (4) Prashanth Harshangi, Enkrypt AI. Authors: Authors: (1) Divyanshu Kumar, Enkrypt AI; (2) Anurakt Kumar, Enkrypt AI; (3) Sahil Agarwa, Enkrypt AI; (4) Prashanth Harshangi, Enkrypt AI. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Problem Formulation and Experiments 2 Problem Formulation and Experiments 3 Experiment Set-up & Results 3 Experiment Set-up & Results 4 Conclusion and References 4 Conclusion and References A. Appendix A. Appendix ABSTRACT Large Language Models (LLMs) have become very popular and have found use cases in many domains, such as chatbots, auto-task completion agents, and much more. However, LLMs are vulnerable to different types of attacks, such as jailbreaking, prompt injection attacks, and privacy leakage attacks. Foundational LLMs undergo adversarial and alignment training to learn not to generate malicious and toxic content. For specialized use cases, these foundational LLMs are subjected to fine-tuning or quantization for better performance and efficiency. We examine the impact of downstream tasks such as fine-tuning and quantization on LLM vulnerability. We test foundation models like Mistral, Llama, MosaicML, and their fine-tuned versions. Our research shows that fine-tuning and quantization reduces jailbreak resistance significantly, leading to increased LLM vulnerabilities. Finally, we demonstrate the utility of external guardrails in reducing LLM vulnerabilities. 1 INTRODUCTION Generative models are becoming more and more important as they are becoming capable of automating a lot of tasks, taking autonomous actions and decisions, and at the same time, becoming better at content generation and summarization tasks. As LLMs become more powerful, these capabilities are at risk of being misused by an adversary, which can lead to fake content generation, toxic, malicious, or hateful content generation, privacy leakages, copyrighted content generation, and much more Chao et al. (2023) Mehrotra et al. (2023) Zou et al. (2023) Greshake et al. (2023) Liu et al. (2023) Zhu et al. (2023) He et al. (2021) Le et al. (2020). To prevent LLMs from generating content that contradicts human values and to prevent their malicious misuse, they undergo a supervised fine-tuning phase after their pre-training, and they are further evaluated by humans and trained using reinforcement learning from human feedback (RLHF) Ouyang et al. (2022), to make them more aligned with human values. Further, special filters called guardrails are put in place as filters to prevent LLMs from getting toxic prompts as inputs and outputting toxic or copyrighted responses Rebedea et al. (2023) Kumar et al. (2023) Wei et al. (2023) Zhou et al. (2024). The complexity of human language makes it difficult for LLMs to completely understand what instructions are right and which are wrong in terms of human values. After going through the alignment training and after the implementation of guardrails, it becomes unlikely that the LLM will generate a toxic response. But these safety measures can easily be circumvented using adversarial attacks, and the LLM can be jailbroken to generate any content that the adversary wants, as shown in recent works Chao et al. (2023) Mehrotra et al. (2023) Zhu et al. (2023). Recent works such as the Prompt Automatic Iterative Refinement (PAIR) attacks Chao et al. (2023) and Tree-of-attacks pruning (TAP) Mehrotra et al. (2023) have shown the vulnerability of LLMs and how easy it is to jailbreak them into generating content for harmful tasks specified by the user. Similarly, a class of methods called privacy leakage attacks Debenedetti et al. (2023) are used to attack LLMs to extract their training data or personally identifiable information Kim et al. (2023), and prompt injection attacks can be used to make an LLM application perform tasks that are not requested by the user but are hidden in the third-party instruction which the LLM automatically executes. Figure 1 shows how an instruction can be hidden inside a summarization text and how the LLM will ignore the previous instruction to execute the malicious instruction. Qi et al. (2023) showed that it only takes a few examples to fine-tune an LLM into generating toxic responses by forgetting its safety training. Our work in this paper extends that notion and shows that both finetuning the LLM on any task (not necessarily toxic content generation) and quantization can affect its safety training. In this study, we use a subset of adversarial harmful prompts called AdvBench SubsetAndy Zou (2023). It contains 50 prompts asking for harmful information across 32 categories. It is a subset of prompts from the harmful behaviours dataset in the AdvBench benchmark selected to cover a diverse range of harmful prompts. The attacking algorithm used is tree-of-attacks pruning Mehrotra et al. (2023) as it has shown to have the best performance in jailbreaking and, more importantly, this algorithm fulfils three important goals (1) Black-box: the algorithm only needs black-box access to the model (2) Automatic: it does not need human intervention once started, and (3) Interpretable: the algorithm generates semantically meaningful prompts. The TAP algorithm is used with the tasks from the AdvBench subset to attack the target LLMs in different settings, and their response is used to evaluate whether or not they have been jailbroken. The rest of the paper is organized in the following manner. Section 2 talks about the experimental setup for jailbreaking in which the models are tested. It specifically describes the different modes of downstream process that an LLM has undergone e.g fine-tuning, quantization and tested for these modes. Section 3 describes the experiment set-up used and defines guardrails Rebedea et al. (2023), fine-tuning and quantization Kashiwamura et al. (2024) Gorsline et al. (2021) Xiao et al. (2023) Hu et al. (2021) Dettmers et al. (2023) settings used in the experimental context. We demonstrate the results in detail and show how model vulnerability is affected by downstream tasks for LLMs. Finally, section 4 concludes the study and talks about methods to reduce model vulnerability and ensure safe and reliable LLM development. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

Increased LLM Vulnerabilities from Fine-tuning and Quantization: Abstract and Introduction

Sujeet Kashid is a Robotics Software Engineer, currently based in the San Francisco Bay Area working with Woven by Toyota. His expertise is integrating and improving the Human Machine Interface (HMI) systems of autonomous vehicles, and he completed his Masters in Science from University of Cincinnati with a specialization in Robotics and Intelligent Systems. Kashid’s research could help increase the safety of operation of Unmanned Aerial Vehicles (UAVs) in GPS-denied indoor areas, and provides a way for multiple robots to cooperate and distribute tasks. Kashid sat down with us and broke down the evolution of the autonomous vehicles, and more importantly, what the next steps are for the automobile industry. Imagine a world of driverless cars–it seems somewhat Orwellian, doesn’t it? Like a scene out of your favorite sci-fi movie, where cars zoom around at breakneck speeds, you never fear they will collide, or in the 1990 film, Total Recall, when we see a driverless taxi. With autonomous vehicles already deployed through major cities like San Francisco and Los Angeles, that sci-fi future is (actually) now. But the world of autonomous vehicles is not limited to this sci-fi vision, it has implications much more far reaching. To gain a better understanding of how autonomous vehicles will continue to evolve, I was lucky enough to sit down with robotics software engineer Sujeet Kashid, who currently works with Woven, which is leading next-generation cars at Toyota. Woven by Toyota is leading Toyota’s “transformation into a mobility company,” according to their website. They’re trying to innovate with one of the most staunch and traditional of all Japanese car brands, which has a vision for the future. If you’re planning on saving up to buy a Toyota car anytime soon, they’re working on incorporating autonomous driving and advanced driver assistance systems. Some of the things they’re working with include OS platform Arene, Woven City, and Toyota's new “test course” for mobility. He is situated at the crossroads of the industry, and gives us a clear picture of not only where we are today, but what the future holds. The Road to Automated Vehicles is Paved with Good Intentions According to Forbes Advisor, in 2022, a total of 42,514 fatal car accidents happened on roads across the country. Meanwhile, 1.6 million crashes resulted in injuries and an additional 4.2 million accidents caused property damages. The main cause of car crashes is drunk driving, lack of seatbelt use, speeding and distracted drivers. Having autonomous vehicles on the roads will go beyond just aiding human error, we’re entering a world where the vehicles themselves can communicate with each other. “In a world where autonomous vehicles can talk to each other, and work cooperatively, they will be able to quickly clear out a way for the emergency vehicles to move faster and uninterrupted to their destination,” said Kashid. Because let's face it: Autonomous vehicles have the potential to reduce road collisions. They also provide a variety of other societal and environmental benefits. But how much of this is realistic? “The intention of robotics is to help people and make the world a better place,” says Kashid, when talking about the safety effects of the robotics industry. The industry has made drastic efforts to help improve the wellbeing of individuals both on the roads and at work–it’s safety focused. “Most road accidents can be attributed to driver error,” says Kashid. “My hope is that autonomous vehicles become ubiquitous making the road a safer place.” In fact, the implications stretch further than the road, entering into workplace safety and time-saving measures. “My research provides a method to operate Unmanned Aerial Vehicles in indoor, GPS-denied environments safely,” says Kashid, diving into another future avenue of exploration. The ability for warehouse vehicles to operate independently will only serve to speed up operations and lower costs. Having autonomous vehicles communicate with each other can help guard our roads with safety. “They can communicate with each other to reach road help faster and save more lives,” he said. “In a world where autonomous vehicles can talk to each other and work cooperatively, this is totally possible.” Pedal To The Metal: The Speedbumps Autonomous Vehicles Face Putting autonomous vehicles on the road is no cheap task. Just look at Waymo, formerly known as the Google self-driving car project, who are trying to make it safe and easy for people to get around with driverless taxis. Waymo has raised over $5.5 billion to get their vehicles on the road in Phoenix alone. While autonomous vehicles are significantly safer than their manned alternatives, a couple of accidents have set the project back. Meanwhile, Tesla switches gears for their customers. Most advanced autonomous features available to many consumers, like Full Self-Driving with Tesla, are only Level 2 autonomous vehicles,” says Kashid. Toyota offers Level 2 autonomous vehicle systems as well, one that can control both steering and acceleration, but Level 2 means it still requires a driver. Right now the issue is twofold before according to the next level. “The next stage in autonomous vehicles is making the functions ubiquitous and affordable,” said Kashid. The potential speed bumps in the development of autonomous vehicles is simple. “I think it's money and time,” said Kashid. “The first cars invented were battery electric vehicles, but it needed huge technological advancements in batteries to make them hold more energy, last longer and make them affordable.” Autonomous cars need to become more affordable, too. “We need technological advancements to make autonomous vehicles affordable, and that research and development takes a lot of money now to enjoy its benefits in the near future,” he adds. Forecasting The Future: Creating Jobs While Innovating With Autonomous Vehicles The intention of robotics is to help people and make the world a better place, notes Kashid. But some tasks may get phased out as technology evolves, and people might lose jobs. “I believe the biggest area of need is that free and accessible skills education be provided to these work forces, so they can be moved to other kinds of jobs,” said Kashid. For example, an Uber driver can be taught to be an autonomous vehicle operator for testing and gathering data.” This might not only mitigate job losses in certain areas, but also help the growth of development of autonomous vehicles if that work force is taught how to help, thus achieving a symbiotic relationship,” he adds. Who doesn’t want to live in a safer world? In order for this vision to be played out there needs to be a massive effort made. This falls in large part on the automobile industry to create affordable autonomous vehicles, but also on the consumers to have faith in the product. Change is scary and often met by resistance. In this case, change should be met with open arms. The future is not far away and thanks to the efforts of those like Kashid, it might be here sooner than we think. Sujeet Kashid is a Robotics Software Engineer, currently based in the San Francisco Bay Area working with Woven by Toyota. His expertise is integrating and improving the Human Machine Interface (HMI) systems of autonomous vehicles, and he completed his Masters in Science from University of Cincinnati with a specialization in Robotics and Intelligent Systems. Kashid’s research could help increase the safety of operation of Unmanned Aerial Vehicles (UAVs) in GPS-denied indoor areas, and provides a way for multiple robots to cooperate and distribute tasks. Kashid sat down with us and broke down the evolution of the autonomous vehicles, and more importantly, what the next steps are for the automobile industry. Sujeet Kashid is a Robotics Software Engineer, currently based in the San Francisco Bay Area working with Woven by Toyota. His expertise is integrating and improving the Human Machine Interface (HMI) systems of autonomous vehicles, and he completed his Masters in Science from University of Cincinnati with a specialization in Robotics and Intelligent Systems. Kashid’s research could help increase the safety of operation of Unmanned Aerial Vehicles (UAVs) in GPS-denied indoor areas, and provides a way for multiple robots to cooperate and distribute tasks. Kashid sat down with us and broke down the evolution of the autonomous vehicles, and more importantly, what the next steps are for the automobile industry. Sujeet Kashid is a Robotics Software Engineer, currently based in the San Francisco Bay Area working with Woven by Toyota. His expertise is integrating and improving the Human Machine Interface (HMI) systems of autonomous vehicles, and he completed his Masters in Science from University of Cincinnati with a specialization in Robotics and Intelligent Systems. Kashid’s research could help increase the safety of operation of Unmanned Aerial Vehicles (UAVs) in GPS-denied indoor areas, and provides a way for multiple robots to cooperate and distribute tasks. Kashid sat down with us and broke down the evolution of the autonomous vehicles, and more importantly, what the next steps are for the automobile industry. Imagine a world of driverless cars–it seems somewhat Orwellian, doesn’t it? Like a scene out of your favorite sci-fi movie, where cars zoom around at breakneck speeds, you never fear they will collide, or in the 1990 film, Total Recall, when we see a driverless taxi. With autonomous vehicles already deployed through major cities like San Francisco and Los Angeles, that sci-fi future is (actually) now. Total Recall, But the world of autonomous vehicles is not limited to this sci-fi vision, it has implications much more far reaching. To gain a better understanding of how autonomous vehicles will continue to evolve, I was lucky enough to sit down with robotics software engineer Sujeet Kashid, who currently works with Woven, which is leading next-generation cars at Toyota. Woven by Toyota is leading Toyota’s “transformation into a mobility company,” according to their website. They’re trying to innovate with one of the most staunch and traditional of all Japanese car brands, which has a vision for the future. If you’re planning on saving up to buy a Toyota car anytime soon, they’re working on incorporating autonomous driving and advanced driver assistance systems. Some of the things they’re working with include OS platform Arene, Woven City, and Toyota's new “test course” for mobility. He is situated at the crossroads of the industry, and gives us a clear picture of not only where we are today, but what the future holds. The Road to Automated Vehicles is Paved with Good Intentions According to Forbes Advisor , in 2022, a total of 42,514 fatal car accidents happened on roads across the country. Meanwhile, 1.6 million crashes resulted in injuries and an additional 4.2 million accidents caused property damages. The main cause of car crashes is drunk driving, lack of seatbelt use, speeding and distracted drivers. Forbes Advisor Forbes Advisor Having autonomous vehicles on the roads will go beyond just aiding human error, we’re entering a world where the vehicles themselves can communicate with each other. “In a world where autonomous vehicles can talk to each other, and work cooperatively, they will be able to quickly clear out a way for the emergency vehicles to move faster and uninterrupted to their destination,” said Kashid. Because let's face it: Autonomous vehicles have the potential to reduce road collisions. They also provide a variety of other societal and environmental benefits. But how much of this is realistic? “The intention of robotics is to help people and make the world a better place,” says Kashid, when talking about the safety effects of the robotics industry. The industry has made drastic efforts to help improve the wellbeing of individuals both on the roads and at work–it’s safety focused. “Most road accidents can be attributed to driver error,” says Kashid. “My hope is that autonomous vehicles become ubiquitous making the road a safer place.” In fact, the implications stretch further than the road, entering into workplace safety and time-saving measures. “My research provides a method to operate Unmanned Aerial Vehicles in indoor, GPS-denied environments safely,” says Kashid, diving into another future avenue of exploration. The ability for warehouse vehicles to operate independently will only serve to speed up operations and lower costs. Having autonomous vehicles communicate with each other can help guard our roads with safety. “They can communicate with each other to reach road help faster and save more lives,” he said. “In a world where autonomous vehicles can talk to each other and work cooperatively, this is totally possible.” Pedal To The Metal: The Speedbumps Autonomous Vehicles Face Putting autonomous vehicles on the road is no cheap task. Just look at Waymo, formerly known as the Google self-driving car project, who are trying to make it safe and easy for people to get around with driverless taxis. Waymo has raised over $5.5 billion to get their vehicles on the road in Phoenix alone. While autonomous vehicles are significantly safer than their manned alternatives, a couple of accidents have set the project back. Meanwhile, Tesla switches gears for their customers. Most advanced autonomous features available to many consumers, like Full Self-Driving with Tesla, are only Level 2 autonomous vehicles,” says Kashid. Toyota offers Level 2 autonomous vehicle systems as well, one that can control both steering and acceleration, but Level 2 means it still requires a driver. Right now the issue is twofold before according to the next level. “The next stage in autonomous vehicles is making the functions ubiquitous and affordable,” said Kashid. The potential speed bumps in the development of autonomous vehicles is simple. “I think it's money and time,” said Kashid. “The first cars invented were battery electric vehicles, but it needed huge technological advancements in batteries to make them hold more energy, last longer and make them affordable.” Autonomous cars need to become more affordable, too. “We need technological advancements to make autonomous vehicles affordable, and that research and development takes a lot of money now to enjoy its benefits in the near future,” he adds. Forecasting The Future: Creating Jobs While Innovating With Autonomous Vehicles The intention of robotics is to help people and make the world a better place, notes Kashid. But some tasks may get phased out as technology evolves, and people might lose jobs. “I believe the biggest area of need is that free and accessible skills education be provided to these work forces, so they can be moved to other kinds of jobs,” said Kashid. For example, an Uber driver can be taught to be an autonomous vehicle operator for testing and gathering data.” This might not only mitigate job losses in certain areas, but also help the growth of development of autonomous vehicles if that work force is taught how to help, thus achieving a symbiotic relationship,” he adds. Who doesn’t want to live in a safer world? In order for this vision to be played out there needs to be a massive effort made. This falls in large part on the automobile industry to create affordable autonomous vehicles, but also on the consumers to have faith in the product. Change is scary and often met by resistance. In this case, change should be met with open arms. The future is not far away and thanks to the efforts of those like Kashid, it might be here sooner than we think.

The Next Step Of Autonomous Cars: Safety Concerns And Affordability, Says Expert

A Data-centric Approach to Class-specific Bias in Image Data Augmentation: Conclusion and Limitation

Nigeria is leveraging its burgeoning tech sector to accelerate economic growth. Vice President Kashim Shettima’s latest initiative underscores this shift, with projections indicating that investments in AI and blockchain could significantly boost Nigeria’s GDP by up to 2% annually. This proclamation came during the launch of the AI Expertise Blockchain and Technology Training and Outsourcing Initiative, held at the Yakubu Gowon NYSC Orientation Camp in Dutse. The initiative, a collaborative effort with tech firm Gluwa, is set to train 1,000 Nigerians annually in cutting-edge fields such as artificial intelligence and blockchain technology. Shettima emphasized that this program is integral to Nigeria's digital transformation strategy and is pivotal for its leadership ambitions in Africa’s tech landscape. In his address, Shettima called upon Nigerian youths to seize this opportunity, framing it as a critical step towards becoming integral players in the global workforce. “We are not merely catching up with the rest of the world; we are poised to overtake them,” he declared, reflecting his confidence in Nigeria's technological future. Additionally, the collaboration with Gluwa—an influential player in blockchain technology—aims to elevate financial inclusion through the Central Bank Digital Currency (CBDC), known as eNaira. This partnership with the Central Bank of Nigeria (CBN) seeks to extend financial services to over 200 million Nigerians, leveraging blockchain’s potential to create a more inclusive financial ecosystem. The Vice President also highlighted Jigawa State’s role as a frontrunner in embracing digital technology, labeling the initiative as a significant milestone in the nation's journey through the Fourth Industrial Revolution. Shettima reiterated the federal government’s commitment to establishing tech hubs nationwide to fuel long-term economic growth. Further demonstrating the government's dedication to fostering small and medium enterprises (SMEs), Shettima announced a grant of N150,000 for each participating business owner at the Expanded National MSME Clinic. This financial support, mandated by President Bola Tinubu, is part of a broader effort to ease business operations and encourage entrepreneurship across Nigeria. Governor Umar Namadi of Jigawa State also voiced strong support for the initiative, stressing its importance for local youth and the state’s long-standing relationship with ICT education. He emphasized that Jigawa's history with technology education aligns with the federal government’s Renewed Hope Agenda and the state’s own strategic goals for digital innovation. The event also saw the inauguration of several key projects, including a solar-powered irrigation farm, a new ICT & Digital Economy Agency, and a series of solar plants aimed at enhancing market infrastructure. These developments reflect a broader strategy to integrate sustainable practices and technological advancements into Nigeria’s economic framework. As Nigeria stands on the brink of a technological renaissance, the AI Expertise Blockchain and Technology Training and Outsourcing Initiative symbolizes a decisive move towards establishing the country as a global leader in technology and innovation. Don’t forget to like and share the story! Vested Interest Disclosure: This author is an independent contributor publishing via our business blogging program. HackerNoon has reviewed the report for quality, but the claims herein belong to the author. #DYOR. Nigeria is leveraging its burgeoning tech sector to accelerate economic growth. Vice President Kashim Shettima’s latest initiative underscores this shift, with projections indicating that investments in AI and blockchain could significantly boost Nigeria’s GDP by up to 2% annually. This proclamation came during the launch of the AI Expertise Blockchain and Technology Training and Outsourcing Initiative, held at the Yakubu Gowon NYSC Orientation Camp in Dutse. The initiative, a collaborative effort with tech firm Gluwa , is set to train 1,000 Nigerians annually in cutting-edge fields such as artificial intelligence and blockchain technology. Shettima emphasized that this program is integral to Nigeria's digital transformation strategy and is pivotal for its leadership ambitions in Africa’s tech landscape. Gluwa In his address, Shettima called upon Nigerian youths to seize this opportunity, framing it as a critical step towards becoming integral players in the global workforce. “We are not merely catching up with the rest of the world; we are poised to overtake them,” he declared, reflecting his confidence in Nigeria's technological future. Additionally, the collaboration with Gluwa—an influential player in blockchain technology—aims to elevate financial inclusion through the Central Bank Digital Currency (CBDC), known as eNaira. This partnership with the Central Bank of Nigeria (CBN) seeks to extend financial services to over 200 million Nigerians, leveraging blockchain’s potential to create a more inclusive financial ecosystem. The Vice President also highlighted Jigawa State’s role as a frontrunner in embracing digital technology, labeling the initiative as a significant milestone in the nation's journey through the Fourth Industrial Revolution. Shettima reiterated the federal government’s commitment to establishing tech hubs nationwide to fuel long-term economic growth. Further demonstrating the government's dedication to fostering small and medium enterprises (SMEs), Shettima announced a grant of N150,000 for each participating business owner at the Expanded National MSME Clinic. This financial support, mandated by President Bola Tinubu, is part of a broader effort to ease business operations and encourage entrepreneurship across Nigeria. Governor Umar Namadi of Jigawa State also voiced strong support for the initiative, stressing its importance for local youth and the state’s long-standing relationship with ICT education. He emphasized that Jigawa's history with technology education aligns with the federal government’s Renewed Hope Agenda and the state’s own strategic goals for digital innovation. The event also saw the inauguration of several key projects, including a solar-powered irrigation farm, a new ICT & Digital Economy Agency, and a series of solar plants aimed at enhancing market infrastructure. These developments reflect a broader strategy to integrate sustainable practices and technological advancements into Nigeria’s economic framework. As Nigeria stands on the brink of a technological renaissance, the AI Expertise Blockchain and Technology Training and Outsourcing Initiative symbolizes a decisive move towards establishing the country as a global leader in technology and innovation. Don’t forget to like and share the story! Vested Interest Disclosure: This author is an independent contributor publishing via our business blogging program. HackerNoon has reviewed the report for quality, but the claims herein belong to the author. #DYOR. Vested Interest Disclosure: This author is an independent contributor publishing via our business blogging program . HackerNoon has reviewed the report for quality, but the claims herein belong to the author. #DYOR. Vested Interest Disclosure: business blogging program business blogging program

Stories, News & Mentions about KALSHI

Kalshi’s Stories on HackerNoon

Kalshi's News & Mentions