Noma Dumezwani (The Little Mermaid) is set as a lead opposite Alexander Skarsgård, in Apple TV+&#x27;s sci-fi drama series Murderbot.

Noma Dumezweni Joins Alexander Skarsgård In Apple Sci-Fi Series 'Murderbot'

The first season of a sci-fi series flopped when it was released in 2022. But the show debuted its seconds season this month, which sent it to the number one spot on IMDB&#x27;s top sci-fi shows.

Sci-fi series tops IMDB's trending shows list after 'disappointing' 2022 debut - and season two beats Netflix's Stranger Things and Black Mirror | Daily Mail Online

David Benioff and DB Weiss masterminded one of the best-loved shows on TV – until fans revolted at the ending. Now, after signing an exclusive deal with Netflix and partnering up with fellow showrunner Alexander Woo, they’re back, with an adaptation of a mind-bending trilogy of Chinese ...

Inside 3 Body Problem, Netflix’s galaxy-brained sci-fi gamble from the creators of Game of Thrones

Here’s an in-depth refresher ahead of the release of ‘Dune: Part Two’

Dune recap: What happened in Part One of Denis Villeneuve’s sci-fi epic? | The Independent

SiFi

sifi

Are You Hungry, Bro? Let’s Hack Your Favorite Food Delivery App

هل أنت جائع يا أخي؟ دعنا نجرب تطبيق توصيل الطعام المفضل لديك

¿Tienes hambre, hermano? Vamos a hackear tu aplicación de entrega de comida favorita

Tu as faim, mon pote ? Hackons votre application de livraison de repas préférée

Ou grangou bro? Ann Hack App Livrezon Manje Pi renmen ou a

Éhes vagy, tesó? Hackeljük meg kedvenc ételkiadó alkalmazását

Lapar, Bro? Yuk Retas Aplikasi Pengiriman Makanan Favoritmu

Hai fame, fratello? Facciamo un hack della tua app di consegna di cibo preferita

お腹が空いていますか？お気に入りのフードデリバリーアプリをハックしてみましょう

Vai tu esi izsalcis, brāli? Uzlauzsim jūsu iecienītāko ēdienu piegādes lietotni

Adakah Anda Lapar, Bro? Mari Menggodam Apl Penghantaran Makanan Kegemaran Anda

A je i uritur, vëlla? Le të hakojmë aplikacionin tuaj të shpërndarjes së ushqimit të preferuar

Ulambile, Bro? Masisebenzise uhlelo lwakho lokusebenza oluthandayo lokulethwa kokudla

RISC-V made the news recently when revealed the BeagleV (Beagle Five) SBC (single board computer). The first affordable, highly available SBC that uses the RISC-V architecture. The board isn’t yet commercially available but you can p in April. For those who don’t know anything about RISC-V, let me explain. BeagleBoard.org https://beagleboard.org/beaglev when they shi What is RISC-V RISC-V stands for ‘reduced instruction set computer’. Every CPU in a computer uses instruction sets which let you speak to the CPU and tell it what to do. Reduced instruction set CPUs like RISC-V or are highly configurable for specific use cases but can be limited in their capabilities to do lots of things at once. ARM RISC-V came out of Berkley in 2010. It was the fifth version of an Open Source RISC architecture (hence RISC-V) and has since become the definitive RISC architecture today. About a year ago the RISC-V foundation became RISC-V international, a global non-profit organisation HQ-ed Geneva, Switzerland. It’s now a global community with over 750 members in more than 50 countries. A large contributing factor is that RISC-V is becoming mainstream with companies and universities all over the world getting in evolved. Folks can join as a big company, a small company or an individual contributor. I recommend you go and do all the and reading you like. to their website signing up Getting started with RISC-V RISC-V is easy to get started with if you know where to look. Members of the community have put together troves of documentation, tutorials, course materials and book for all levels. You can sign up for a course created in collaboration with entitled: ‘RVFPGA: Complete Course in Understanding Computer Architecture’. If you follow the same link you can also contribute course materials there too. the online training, ‘imagination university’ If you’re an organisation you can become . This is a technical program where groups of people can gain expertise in RISC_V and engage with the wider community at the same time. a RISC-V training partner If you’re more of a self server there are also lots of more general educational resources too. The community has put together a , the materials that go with them, various courses that use RISC-V in someway and an assortment of other helpful courses. list of interactive RISC-V courses More information and a few more resources can found under . I found it a little difficult to dig up so there it is in case you see it here. /community/learn on their website BeagleV RISC-V SBC The BeagleBoard.org foundation, Seeed Studio, and Chinese silicon vendor Starfive partnered to design and launch the BeagleV SBC. It’s powered by StarFive JH7100 dual-core SiFive U74 RISC-V processor with Vision DSP, NVDLA engine, and neural network engine for AI acceleration. If that doesn't make a whole lot of sense to you it means the board is capable. You’ll be able to use it like most other development boards out there and know that it is built to handle AI/ML processes. This really means two things. From a practical perspective, it means you can develop an all open AI/ML device. No proprietary code, no proprietary silicon, all open-source (unless you count buying the board I suppose). It’s an impressive board. But second, if successful, nicely proves the validity and ability of fully open-source tech in real-world products. It’ll be one for the open-source die-hards out there, everyone else too. and Large hardware companies will be more ready to adopt the RISC-V architecture, it will become more accessible to hobbyists and small businesses because ultimately it will be cheaper, the communities will grow providing more support and, perhaps most exciting of all, we could see an increase in new hardware/device companies that capitalise on all open-source AI. Contributing Like most modern-day open-source projects and communities, anyone can contribute and it's easy to do so. Gone are the day when you needed a CS degree to really contribute anything useful. If you are a developer and would like to contribute though you can , see what you can do. go to the GitHub, have a poke around If you’re not, first become an individual contributor (a type of free membership) and sign up to the working groups and mailing lists where your skills can come in handy. There’s docs groups, marketing groups, events groups, tech groups, and more. From what I’ve seen RISC-V as an org have done a great job in making it easy to get started. So if you’re interested in contributing to RISC-V the all open-source architecture, you can. Community Full disclosure, I am brand new to the RISC-V community. The first thing I was involved in was the first London Open Source RISC-V meet up of 2021 the other day. But I’ve been looking for an open hardware to get stuck into and think this is it. something They seem a good sort, the first 20 minutes of the virtual meet up was a discussion about biscuits. And then the three talks that followed were terribly impressive if occasionally a little too technical for me, but I’ll get there. If you want to get involved with the community the best thing to do is to sign up and start working your way through stuff. Join some open meetings, sign up to some groups, send messages to the mailing lists and generally spread the word. If you want to sign up as an individual or as an organisation just do it. Published under paywall here .

Getting Started With RISC-V Board

Before I scare everyone off, I do not mean to say that itself is doomed to fail. What I am referring to is that I believe the implementation of solely as a currency is destined to be far less successful than the world believes it to be. This includes the likes of Litecoin and Dash, and less so privacy based ones like Monero and Zcash. The reason I believe this lies in the value proposition of blockchain itself. To start, let’s breakdown what a blockchain really is. Bitcoin blockchain Data Structure > Consensus Layer> Incentive Model At its core, blockchain is nothing more than a novel structure for storing and interacting with data. The incentive layer is merely a mechanism designed to keep the consensus layer and data structure intact. Its potential to transform the world lies only marginally in its ability to store value, and far more in its ability to link together computers. Let’s look at the tradeoffs of bitcoin vs. a traditional currency. — Many proponents of bitcoin cite its deflationary component as a strong advantage, but it also has significant downsides. Bitcoin inflation cannot be controlled and as such is lower than other currencies. However, it still has inflation, it is simply lower. At some point this means that bitcoin will appreciate in value due to the growth of value created on the platform outpacing its inflation rate (assuming that bitcoin has already fully penetrated its use case, obviously bitcoin is a speculative asset today and has blown up despite virtually no mainstream payment use). This, however, has never been the purpose of an actual currency. Currency is a base layer of the economy, we don’t put our money in it expecting returns, we put our money in it so that it can be easily spent. If we want to gain a return on our assets, we put it in other financial instruments. Is bitcoin supposed to be the ultimate be all end all financial instrument that just creates unlimited wealth for everyone? Regardless maybe it is a slight value add that you gain money by keeping it in an appreciating version of cash, it simply is not enough of a disruptive force to gain widespread adoption because we have other versions of this that achieve the same goal already. This aspect of bitcoin also has significant drawbacks. In a world run on bitcoin, government bailouts cannot exist. The 2008 recession would have turned into a depression without the US government’s ability to essentially cover the losses of large financial firms by taking a little value out of each dollar in circulation. Increasing the money supply may not be a perfect system, but it does allow a certain amount of liquidity in crisis events that would not exist with Bitcoin. Even in a world where the US Government would have thought of using its own stash of Bitcoin to prop us AIG, it has significant disincentives to do this because unlike the US Dollar, it appreciates in value, thus making the losses not only immediate but also spread out over the future. Inflation — Proponents of bitcoin also say that the lack of central control empowers the individual to own his own money. No entity has the ability to seize control over her or his assets for any reason, which makes it much safer to own. First of all, this is a myth. As John Locke put it, we have entered a social contract with our government and endowed it with dominion over us in exchange for enforcement of individual rights. Aside from theory, what this means in practice is that governments can do whatever the hell they want to. In some respects they can’t actually take it from you without your ultimate consent, but they can apply virtually unlimited amounts of pressure to get you to do so for them. The limits of this pressure are exclusively bound by the brutality of the country you happen to be a part of, ergo ** **The lack of central authority also has significant downsides to the legal ramifications of money. In the event of a death where the will is written out to someone’s children, but nobody has the private key, what do you do? In the event of a Bernie Madoff scandal where Bernie instead took all of his thefts and put them into bitcoin, how do you recover that? If there is a lending platform on bitcoin that faces uncertainty, how will it handle a bank run without FDIC insurance? The lawlessness of bitcoin can be seen as a good thing, but it also has a hint of anarchy in that it almost seems like it despises the governments we’ve created. Obviously this makes sense in corrupt countries, but in countries where the rule of law is sound and successful the technology already exists to perform every function that bitcoin performs. I can make feeless transactions with Venmo, I can invest in stocks to earn a return, and why would I do any of this without the ultimate fallback of the US court system to protect my money? Lack of central control this is least effective in the places that need it the most. — Another theoretical benefit of bitcoin is its lack of a ‘single point of failure’ allowing it in theory to be more secure than holding your money in a traditional bank. This is a facade and it envisions some kind of apocalyptic scenario in which all bank data is erased or someone goes in and manually changes your dollar figures. First of all, most large banks are supported as SIFIs by the government, which have already proven to be too big to fail. Make whatever value assessments you need to about this, but also remember it is likely that the government would have the exact same responsibility regardless of the type of currency used in the economy. In terms of the manually changing dollar figures, I have no statistics on the matter so I can’t argue too strongly that it doesn’t happen. All I can say is I’ve never heard of anything like it and any bank participating in practices such as these would most likely lose their entire customer base. Single point of failure All this being said, the ‘disruptive potential’ of bitcoin is slim to none. Today, it is essentially operating on a wave of mass hype due to the ridiculous gains it has seen as an asset class, and has been struggling to perform the exact activities it was designed to perform. Once it can actually execute feeless transactions and contains a more complete scripting language, then maybe we can talk, but for now it is emphatically a worse economic system than the one we have today. So then the question becomes, if not currency, what is blockchain good for? My main thesis in this respect boils down to a couple arguments. — This means that the dollar values ascribed to the value of the token are exclusively valued to a particular asset which you need to purchase. In Ethereum, Ether is worth the purchase of computing power on the network, in Filecoin the asset is tied to the value of the network storage hardware, etc. Bitcoin trying to be the be all end all store of value is cool in principle but has no disruptive potential i.e. reason that people would want to use it. It is this circular logic that says as more people use bitcoin it is more likely to become a widely used currency and will appreciate in value because more people are using it, but if there is no reason to use it in the first place then all the value accumulated from people joining the network to get profit is a sham. What tying the incentive layer of a blockchain to an asset does is a) give a more clear picture of the valuation structure of a blockchain and thus decreases volatility of the asset and b) improves the performance of the protocol itself as developers are incentivized to make the protocol as production ready as possible so that the most possible people use it and it has the most disruptive potential. In bitcoin this has been a significant problem as we have seen core developers neglect proper performance of the network in order to profit off of their personal scaling solutions (lightning network / blockstream conflict of interest) instead of profiting from the increase in value due to the widespread use of the network. Were bitcoin tied to say the net cost savings of using the network over alternative money transfer mechanisms, the developers would have had no problem increasing the block size (obviously this ignores the decentralization aspect, but it would have been far less contentious). The incentive layer should be restricted to incentivizing correct participation in the network — When you have this network of computers all correctly incentivized to perform specific tasks, what this gives you is a computing platform in which nobody has pricing power. This is a beautiful thing, because all the tech companies today have pricing power and they use it mercilessly. The blockchain essentially drops all price of computing to commodity levels, but only for applications that you can program into a blockchain, which I suppose is theoretically all of them. The core innovation of blockchain is that previously when we created these peer-to-peer networks we had no way of telling who was cheating and who wasn’t, which means you couldn’t put money on top of it which means you can’t incentivize correct participation. The blockchain says these are the rules of participation, this is how you check if people are participating properly, and this is how you take advantage of other people who are not participating correctly and take their money. By following this line of logic, those who create the most efficient rules of participating properly and target the markets with the highest amount of pricing power will create the most value in the future. This doesn’t just apply computing resources like bandwidth, file storage, and CPU time. This also applies to physical assets that can be ‘proven’ in the blockchain. As an example of this take Power Ledger, a token trying to create an economy of energy. If Power Ledger can create a way to prove that energy was transferred from account A to account B, then they can commoditize energy. It’s much more difficult than computing power, however, because the internet does not exist for power transfer in the same way it does for information. We do not have the ability to send anyone power, and create our own power that we can sell off if we have excess the way we do with computing resources. This is where the future of blockchain exists, and the algorithms and infrastructure requirements behind it are going to blow our minds. Blockchain will commoditize everything Conclusion Bitcoin is the most primitive implementation of the blockchain, which follows logically from the fact that it was the original. This does not by any means imply that it can’t evolve to be a successful currency, it only means that it utilizes the key innovations of blockchain in the most rudimentary fashion. Blockchains are networks in which an unlimited number of people can participate and correct participation can be efficiently measured. This can be applied to currency, but it simply does not differentiate itself enough as a product. What will truly be the key value creator of blockchain is its ability to inject massive supply into markets that currently are dominated by players with pricing power.

Bitcoin: the least significant implementation of blockchain

The (Reduced Instruction Set Computer) processor is a chip that is still in it’s infancy, but it’s a chip that everyone should be supporting. You might be wondering, what makes this chip so great? RISC-V The RISC-V architecture is great because it is the only processor that has a completely open source instruction set, if you want to learn more . What’s an open source instruction set? In layman’s terms, it means that the way the processor moves around 1s and 0s is available for everyone to see. The advent of what is probably the worst security bug, Meltdown and Spectre, boiled down to a flaw in the instruction set of Intel’s processors. I don’t think anyone in the security field was completely shocked, after all, a s demonstrated that there were unknown instructions in the x86 architecture, and that inevitably means there are hidden bugs, it was only a matter of time before a truly devastating bug was found, in our case it was Spectre and Meltdown. check out their website peaker at a BlackHat conference Another benefit of RISC-V is that it enables companies to develop a product that is tailored specifically to their workload, so they start with the RISC-V core and can add whatever it is they specifically need, saving both time and money. These savings can theoretically be passed on to the consumer either through a lower cost, or in the longterm by having a lower energy footprint. However, I believe the biggest benefit of RISC-V is the inherent security and peace of mind it will give to both consumers and businesses alike. The Spectre and Meltdown bugs are huge problems in the security sphere because over 90% of the server market is owned by Intel, which means that nearly every cloud service is running on hardware that allows attackers to read data that they should not be able to access. To make matters worse, early reports indicate that after these bugs are patched, there will be a performance hit of anywhere between 4–30%, depending on the workload. I hope that these devastating new bugs will make everyone rethink how they go about designing hardware, and support products such as RISC-V. I will confess that even if we all switched to RISC-V in the future, it doesn’t mean that all of our security woes will be left in the past, RISC-V is BSD licensed, meaning that a vendor can tailor it to themselves and keep those custom bits of code behind closed doors, and closed doors means that there is most likely a vulnerability waiting to be unearthed. The only way forward is with a future that is open and transparent, has become too ingrained in our society to be kept behind closed doors. technology I’m a firm believer in open source because it’s the only way to achieve the kind of optimistic future that we want. AI that will be responsible for driving cars should be open for everyone to see what it does and how it works, processors that the AI runs on should be the same. As far as I know, RISC-V is the latest open source processor architecture that is currently being used or developed for a variety of products by multiple big name companies. (EDIT: ARM and SPARC are other open source processor designs) For example, . If you’re a tinkerer or someone who is a fan of small board computers such as Raspberry Pi’s or Arduino’s, SiFive, a company founded by a former student of the man who invented RISC, . Nvidia will be using a RISC-V chip onboard their GPUS and Western Digital is planning to ship a billion RISC-V units in the upcoming year sells a RISC-V developer board right now RISC-V is an existing piece of technology with brilliant minds and monolithic companies propelling it forward, I hope that one day consumers and businesses alike will have easy access to this open architecture available on their laptops, phones, and desktops.

Why we need RISC-V

HBO’s comedy series Silicon Valley received critical acclaim and a near-perfect score on Rotten Tomatoes. Like most great comedies, the show’s success is likely thanks to its ability to offer a satirical take on true-to-life scenarios on some of the more outlandish elements of the Bay Area tech scene between 2014-2019. In this case, the founder of Pied Piper and his team are on a mission to bring his novel compression technology out into the world where it can scale to the masses. Back when the series was in production, this compression technology wasn’t much more than a clever storytelling device that served to show what a niche this scrappy startup had found with its groundbreaking solution. Yet if we fast forward to 2024, with the boom in GenAI technology and unprecedented demand for data, there’s never been more urgent need for a solution akin to that put forward by the creators of Silicon Valley. However, it seems like one company has succeeded in creating a surprisingly similar compression model that looks to break through the huge data volumes at play today. To understand Silicon Valley, watch Silicon Valley In the series, Pied Piper is a compression startup that provides instant access to your data. And it’s far from the only example of life imitating art. Indeed, the concept of a data compression tool isn’t the only element of Pied Piper that crossed over into reality. In fact for the research of the show professors were consulted, with the Weissman Score being a new compression metric developed for the series itself. However, this didn’t stop real-world researchers from actually using it. Meanwhile, Vinith Misra, a researcher at IBM’s Watson Group consulted for the show, went so far as to write a technical paper that offered a fictional improvement to the compression algorithm. However, while the compression algorithm put forward for the show was fiction, a powerful new breed of GPUs and advanced processing chips mean that next-level data processing capabilities are now within the realms of possibility. Bringing Pied Piper’s dream into reality While a few companies are getting closer to earning the title of the ‘real Pied Piper’ for data compression, one New York startup is garnering plenty of attention of late for its strides forward in this specialism. SQream Blue, the data lakehouse solution from the data acceleration platform, recently outperformed similar solutions on the market by a country mile. When it comes to speed and cost, SQream Blue and its patented technology compression solution performed 3 times faster and at half the cost of Databricks during the comparison element of the test. To put this into closer detail, SQream Blue’s total runtime was 2462.6 seconds, with the total cost for processing the data end-to-end being $26.94. Databricks’ total runtime was 8332.4 seconds, at a cost of $76.94, indicating a significant cost-performance advantage by SQream Blue for big data analysis. To get these results, SQream ran the benchmark on Amazon Web Services (AWS) with a scale factor of 30,000, which creates a dataset of around 30 TB, to test the relative capabilities of SQream at scale. All generated data was stored as Apache Parquet files on Amazon Simple Storage Service (Amazon  S3), and the queries were processed without pre-loading into a database. According to the announcement, the SQream Blue’s unparalleled speed was equivalent to reading every cataloged book in the US Library of Congress in under an hour – and then buying them all for less than $25, giving a close reminder of Pied Piper’s standout results on the Weissman test on the infamous Silicon Valley episode at TechCrunch Disrupt. In 2024, these kinds of gains can’t come quickly enough. Tackling our data addiction Gavin Belson, billionaire senior executive and main antagonist on Silicon Valley, delivered a telling rant on this subject, “Data creation is exploding. With all the selfies and useless files people refuse to delete on the cloud, 92 percent of the world’s data was created in the last two years alone. At the current rate, the world’s data storage capacity will be overtaken by next spring. It will be nothing short of a catastrophe.” “Data shortages, data rationing, data black markets. Someone’s compression will save the world from data-geddon, and it sure as hell better be Nucleus and not goddamn Pied Piper!” But this isn’t just about selfies and content. Public clouds now host more than half of all enterprise workloads and some businesses spend upwards of $12 million annually on that space. This represents a huge amount of business-critical data that is at the point of being unsustainable to manage using current methods. It’s for this reason that the real world does need a Pied Piper. HBO’s comedy series Silicon Valley received critical acclaim and a near-perfect score on Rotten Tomatoes. Like most great comedies, the show’s success is likely thanks to its ability to offer a satirical take on true-to-life scenarios on some of the more outlandish elements of the Bay Area tech scene between 2014-2019. In this case, the founder of Pied Piper and his team are on a mission to bring his novel compression technology out into the world where it can scale to the masses. Pied Piper Back when the series was in production, this compression technology wasn’t much more than a clever storytelling device that served to show what a niche this scrappy startup had found with its groundbreaking solution. Yet if we fast forward to 2024, with the boom in GenAI technology and unprecedented demand for data, there’s never been more urgent need for a solution akin to that put forward by the creators of Silicon Valley. However, it seems like one company has succeeded in creating a surprisingly similar compression model that looks to break through the huge data volumes at play today. To understand Silicon Valley, watch Silicon Valley To understand Silicon Valley, watch Silicon Valley In the series, Pied Piper is a compression startup that provides instant access to your data. And it’s far from the only example of life imitating art. Indeed, the concept of a data compression tool isn’t the only element of Pied Piper that crossed over into reality. In fact for the research of the show professors were consulted, with the Weissman Score being a new compression metric developed for the series itself. However, this didn’t stop real-world researchers from actually using it. real-world researchers Meanwhile, Vinith Misra, a researcher at IBM’s Watson Group consulted for the show, went so far as to write a technical paper that offered a fictional improvement to the compression algorithm. technical paper However, while the compression algorithm put forward for the show was fiction, a powerful new breed of GPUs and advanced processing chips mean that next-level data processing capabilities are now within the realms of possibility. Bringing Pied Piper’s dream into reality Bringing Pied Piper’s dream into reality While a few companies are getting closer to earning the title of the ‘real Pied Piper’ for data compression, one New York startup is garnering plenty of attention of late for its strides forward in this specialism. SQream Blue, the data lakehouse solution from the data acceleration platform, recently outperformed similar solutions on the market by a country mile. When it comes to speed and cost, SQream Blue and its patented technology compression solution performed 3 times faster and at half the cost of Databricks during the comparison element of the test. To put this into closer detail, SQream Blue’s total runtime was 2462.6 seconds, with the total cost for processing the data end-to-end being $26.94. Databricks’ total runtime was 8332.4 seconds, at a cost of $76.94, indicating a significant cost-performance advantage by SQream Blue for big data analysis. To get these results, SQream ran the benchmark on Amazon Web Services (AWS) with a scale factor of 30,000, which creates a dataset of around 30 TB, to test the relative capabilities of SQream at scale. All generated data was stored as Apache Parquet files on Amazon Simple Storage Service (Amazon  S3), and the queries were processed without pre-loading into a database. According to the announcement, the SQream Blue’s unparalleled speed was equivalent to reading every cataloged book in the US Library of Congress in under an hour – and then buying them all for less than $25, giving a close reminder of Pied Piper’s standout results on the Weissman test on the infamous Silicon Valley episode at TechCrunch Disrupt . TechCrunch Disrupt In 2024, these kinds of gains can’t come quickly enough. Tackling our data addiction Tackling our data addiction Gavin Belson, billionaire senior executive and main antagonist on Silicon Valley, delivered a telling rant on this subject, “Data creation is exploding. With all the selfies and useless files people refuse to delete on the cloud, 92 percent of the world’s data was created in the last two years alone. At the current rate, the world’s data storage capacity will be overtaken by next spring. It will be nothing short of a catastrophe.” “Data shortages, data rationing, data black markets. Someone’s compression will save the world from data-geddon, and it sure as hell better be Nucleus and not goddamn Pied Piper!” But this isn’t just about selfies and content. Public clouds now host more than half of all enterprise workloads and some businesses spend upwards of $12 million annually on that space. upwards of $12 million annually This represents a huge amount of business-critical data that is at the point of being unsustainable to manage using current methods. It’s for this reason that the real world does need a Pied Piper.

Silicon Valley’s Pied Piper is Now Real Thanks to New Compression Technology

Silicon Valley’s Pied Piper ukax Jichhax Chiqpach Machaq Tecnología de Compresión ukar yuspajarasawa

Pied Piper от Силиконовата долина вече е истински благодарение на новата технология за компресиране

El flautista de Hamelin de Silicon Valley ahora es una realidad gracias a la nueva tecnología de compresión

Silicon Valley-ko Pied Piper orain benetakoa da Konpresio Teknologia Berriari esker

Silicon Valleyn Pied Piper on nyt todellinen uuden pakkaustekniikan ansiosta

Սիլիկոնային հովտի Pied Piper-ն այժմ իրական է նոր սեղմման տեխնոլոգիայի շնորհիվ

シリコンバレーのパイドパイパーが新しい圧縮技術のおかげで現実に

Silicon Valley’s Pied Piper Bjale ke ya Kgonthe Ka lebaka la Theknolotši ye Mpsha ya Kgateletšo

Pied Piper din Silicon Valley este acum real datorită noii tehnologii de compresie

Silicon Valley's Pied Piper Hadda waa dhab Thanks to Tiknoolajiyada Isku-buufinta Cusub

Пиед Пипер из Силицијумске долине је сада стваран захваљујући новој технологији компресије

சிலிக்கான் வேலியின் பைட் பைபர் இப்போது புதிய சுருக்க தொழில்நுட்பத்திற்கு நன்றி

Optimizing Your Resume for the Silicon Valley Fast Lane

Authors:
(1) Suriya Gunasekar, Microsoft Research;
(2) Yi Zhang, Microsoft Research;
(3) Jyoti Aneja, Microsoft Research;
(4) Caio C´esar Teodoro Mendes, Microsoft Research;
(5) Allie Del Giorno, Microsoft Research;
(6) Sivakanth Gopi, Microsoft Research;
(7) Mojan Javaheripi, Microsoft Research;
(8) Piero Kauffmann, Microsoft Research;
(9) Gustavo de Rosa, Microsoft Research;
(10) Olli Saarikivi, Microsoft Research;
(11) Adil Salim, Microsoft Research;
(12) Shital Shah, Microsoft Research;
(13) Harkirat Singh Behl, Microsoft Research;
(14) Xin Wang, Microsoft Research;
(15) S´ebastien Bubeck, Microsoft Research;
(16) Ronen Eldan, Microsoft Research;
(17) Adam Tauman Kalai, Microsoft Research;
(18) Yin Tat Lee, Microsoft Research;
(19) Yuanzhi Li, Microsoft Research. Table of Links Abstract and 1. Introduction
2 Training details and the importance of high-quality data
2.1 Filtering of existing code datasets using a transformer-based classifier
2.2 Creation of synthetic textbook-quality datasets
2.3 Model architecture and training
3 Spikes of model capability after finetuning on CodeExercises, 3.1 Finetuning improves the model’s understanding, and 3.2 Finetuning improves the model’s ability to use external libraries
4 Evaluation on unconventional problems with LLM grading
5 Data pruning for unbiased performance evaluation
5.1 N-gram overlap and 5.2 Embedding and syntax-based similarity analysis
6 Conclusion and References
A Additional examples for Section 3
B Limitation of phi-1
C Examples for Section 5 5.1 N-gram overlap N-gram measures the similarity of text segments based on the shared n-word sequences. We calculate the n-gram overlap between the docstrings of each humaneval question and each exercise in the CodeExercises dataset that was generated. We found 4 humaneval questions with 13-gram overlap with at least one of the entries in our dataset. After further investigating, we found out that all the 4 overlap cases in the 13-gram are all false positives such as the example below. Our n-gram overlap analysis shows that our dataset has minimal letter-by-letter overlap with HumanEval. 5.2 Embedding and syntax-based similarity analysis As we just saw, the n-gram analysis is not refined enough to find similar code snippets between HumanEval and CodeExercises. Instead we use a combination of embedding and syntax-based distances. For the embedding distance we compute the L2 distance between the embedding of the code snippets where the embedding is derived from a pre-trained CodeGen-Mono 350M model [NPH+ 23]. We observe that the embedding distance is successful in capturing code pairs where the overall code semantics are similar, which can be inferred via the Python Docstring, function/class names, as well as the code structure. For the syntax-based distance we calculate the (string) edit distance between the abstract syntax trees (ASTs) of two given code snippets. The AST distance successfully identifies overlapping sections between code pairs while being agnostic to non-syntax text such as variable/function naming, comments, and Python Docstrings. For our pruning of CodeExercises we fix a threshold for the embedding distance, and we test several match rate τ for the AST distance. See Appendix C for examples of code pairs that are captured with the embedding distance and various AST match rates τ . We vary τ between 0.95 and 0.8, which corresponds to removing between 42.5K to 354K of the 879.5K total problems in CodeExercises. Table 3 summarizes the performance of our retrained phi-1 on pruned datasets (with τ = 0.95, 0.9, 0.85 and 0.8) versus the original phi-1 trained on full CodeExercises and the 15.5B-parameter StarCoderprompted. We divide the HumanEval problems into two subsets (“similar” and “non-similar”) based on whether or not they have at least one close match (for this given τ ) inside the original CodeExercises dataset. We then report the accuracy of the models on each subset of HumanEval separately. As one can see, even after heavily pruning our dataset, phi-1 still outperforms StarCoder-Prompted by a large margin, which validates that our performance boost is not due to dataset “contamination”, even when the latter term is understood loosely. Note also that the accuracy of all models is lower on the HumanEval non-similar subset versus the similar one. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Suriya Gunasekar, Microsoft Research; (2) Yi Zhang, Microsoft Research; (3) Jyoti Aneja, Microsoft Research; (4) Caio C´esar Teodoro Mendes, Microsoft Research; (5) Allie Del Giorno, Microsoft Research; (6) Sivakanth Gopi, Microsoft Research; (7) Mojan Javaheripi, Microsoft Research; (8) Piero Kauffmann, Microsoft Research; (9) Gustavo de Rosa, Microsoft Research; (10) Olli Saarikivi, Microsoft Research; (11) Adil Salim, Microsoft Research; (12) Shital Shah, Microsoft Research; (13) Harkirat Singh Behl, Microsoft Research; (14) Xin Wang, Microsoft Research; (15) S´ebastien Bubeck, Microsoft Research; (16) Ronen Eldan, Microsoft Research; (17) Adam Tauman Kalai, Microsoft Research; (18) Yin Tat Lee, Microsoft Research; (19) Yuanzhi Li, Microsoft Research. Authors: Authors: (1) Suriya Gunasekar, Microsoft Research; (2) Yi Zhang, Microsoft Research; (3) Jyoti Aneja, Microsoft Research; (4) Caio C´esar Teodoro Mendes, Microsoft Research; (5) Allie Del Giorno, Microsoft Research; (6) Sivakanth Gopi, Microsoft Research; (7) Mojan Javaheripi, Microsoft Research; (8) Piero Kauffmann, Microsoft Research; (9) Gustavo de Rosa, Microsoft Research; (10) Olli Saarikivi, Microsoft Research; (11) Adil Salim, Microsoft Research; (12) Shital Shah, Microsoft Research; (13) Harkirat Singh Behl, Microsoft Research; (14) Xin Wang, Microsoft Research; (15) S´ebastien Bubeck, Microsoft Research; (16) Ronen Eldan, Microsoft Research; (17) Adam Tauman Kalai, Microsoft Research; (18) Yin Tat Lee, Microsoft Research; (19) Yuanzhi Li, Microsoft Research. Table of Links Abstract and 1. Introduction 2 Training details and the importance of high-quality data 2.1 Filtering of existing code datasets using a transformer-based classifier 2.2 Creation of synthetic textbook-quality datasets 2.3 Model architecture and training 3 Spikes of model capability after finetuning on CodeExercises, 3.1 Finetuning improves the model’s understanding, and 3.2 Finetuning improves the model’s ability to use external libraries 4 Evaluation on unconventional problems with LLM grading 5 Data pruning for unbiased performance evaluation 5.1 N-gram overlap and 5.2 Embedding and syntax-based similarity analysis 6 Conclusion and References A Additional examples for Section 3 B Limitation of phi-1 C Examples for Section 5 Abstract and 1. Introduction Abstract and 1. Introduction 2 Training details and the importance of high-quality data 2 Training details and the importance of high-quality data 2.1 Filtering of existing code datasets using a transformer-based classifier 2.1 Filtering of existing code datasets using a transformer-based classifier 2.2 Creation of synthetic textbook-quality datasets 2.2 Creation of synthetic textbook-quality datasets 2.3 Model architecture and training 2.3 Model architecture and training 3 Spikes of model capability after finetuning on CodeExercises, 3.1 Finetuning improves the model’s understanding, and 3.2 Finetuning improves the model’s ability to use external libraries 3 Spikes of model capability after finetuning on CodeExercises, 3.1 Finetuning improves the model’s understanding, and 3.2 Finetuning improves the model’s ability to use external libraries 4 Evaluation on unconventional problems with LLM grading 4 Evaluation on unconventional problems with LLM grading 5 Data pruning for unbiased performance evaluation 5 Data pruning for unbiased performance evaluation 5.1 N-gram overlap and 5.2 Embedding and syntax-based similarity analysis 5.1 N-gram overlap and 5.2 Embedding and syntax-based similarity analysis 6 Conclusion and References 6 Conclusion and References A Additional examples for Section 3 A Additional examples for Section 3 B Limitation of phi-1 B Limitation of phi-1 C Examples for Section 5 C Examples for Section 5 5.1 N-gram overlap N-gram measures the similarity of text segments based on the shared n-word sequences. We calculate the n-gram overlap between the docstrings of each humaneval question and each exercise in the CodeExercises dataset that was generated. We found 4 humaneval questions with 13-gram overlap with at least one of the entries in our dataset. After further investigating, we found out that all the 4 overlap cases in the 13-gram are all false positives such as the example below. Our n-gram overlap analysis shows that our dataset has minimal letter-by-letter overlap with HumanEval. 5.2 Embedding and syntax-based similarity analysis As we just saw, the n-gram analysis is not refined enough to find similar code snippets between HumanEval and CodeExercises. Instead we use a combination of embedding and syntax-based distances. For the embedding distance we compute the L2 distance between the embedding of the code snippets where the embedding is derived from a pre-trained CodeGen-Mono 350M model [NPH+ 23]. We observe that the embedding distance is successful in capturing code pairs where the overall code semantics are similar, which can be inferred via the Python Docstring, function/class names, as well as the code structure. For the syntax-based distance we calculate the (string) edit distance between the abstract syntax trees (ASTs) of two given code snippets. The AST distance successfully identifies overlapping sections between code pairs while being agnostic to non-syntax text such as variable/function naming, comments, and Python Docstrings. For our pruning of CodeExercises we fix a threshold for the embedding distance, and we test several match rate τ for the AST distance. See Appendix C for examples of code pairs that are captured with the embedding distance and various AST match rates τ . We vary τ between 0.95 and 0.8, which corresponds to removing between 42.5K to 354K of the 879.5K total problems in CodeExercises. Table 3 summarizes the performance of our retrained phi-1 on pruned datasets (with τ = 0.95, 0.9, 0.85 and 0.8) versus the original phi-1 trained on full CodeExercises and the 15.5B-parameter StarCoderprompted. We divide the HumanEval problems into two subsets (“similar” and “non-similar”) based on whether or not they have at least one close match (for this given τ ) inside the original CodeExercises dataset. We then report the accuracy of the models on each subset of HumanEval separately. As one can see, even after heavily pruning our dataset, phi-1 still outperforms StarCoder-Prompted by a large margin, which validates that our performance boost is not due to dataset “contamination”, even when the latter term is understood loosely. Note also that the accuracy of all models is lower on the HumanEval non-similar subset versus the similar one. 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

Textbooks are All You Need: N-gram Overlap and Embedding and Syntax-based Similarity Analysis

The Noonification: Nuke Silicon Valley - Decentralize AI (9/5/2024)

There are the unavoidable “Three M’s” talking points in almost all of The Joe Rogan Experience episodes. At the least, in the most recent ones. Mushrooms
Monkeys
Microsoft So, if you want to see it, you have to hear it. There’s no other way around them. I was more than curious about what Peter Thiel had to say in episode #2190. I just knew it would be juicy, otherwise, he would be talking to Lex Fridman, not Joe Rogan. Mushrooms. There’s a saying where I’m from: “Some mushrooms you eat only once.” But, we all know that Joe likes magic (psychedelic) mushrooms and other dimensions. I don’t know man, maybe I’m too old or too scared for that sh..t. Monkeys. For some reason, Joe is obsessed with Netflix’s “Chimp Empire.” I don’t have to watch how chimps are “navigating complex social politics, family dynamics, and dangerous territory disputes.” Why? Because all I have to do is look at the world around me to know that the adults had left the room a long time ago. Microsoft. I’m on the same page with Joe, mostly. Here’s one of my award-winning stories about Linux, and here’s another one about Mr. Bill Gates. Please, don’t get this the wrong way, I’m nowhere near Tarantino’s “Kill Bill: Hypothetical Volume 3” mood. That’s why I was rolling my eyes when Peter Thiel started dwelling on his “theory” about Gates’ wealth and especially marriage. But, you have to love Joe because he really knows when and how to make the U-turn in the middle of a conversation smoothly and quickly. The next thing you know, Peter Thiel was talking about AI. To be more precise, the decentralized AI. Controversial But Brilliant Point Now, you have to take my word for it because it is a three-hour-long episode. So, I’m quoting (paraphrasing) to the best of my memory and respect: Crypto is libertarian, AI is communist You can say or think what you want about this member of the PayPal Mafia, but Peter Thiel hit the nail on the head with the brilliant brutal force with this one. He also identified Silicon Valley as the root of all centralized AI evil correctly and unapologetically. Honestly, I lost him a little bit when he started talking about China. So, I’m not quite sure if the decentralization of AI is a state or tech-level problem. Let me put it this way. If I were some AI genius and wanted to make money, I would already be hooked on Soylent. But, if I wanted to try out some crazy AI things without any ethical limits, I would already be a proud member of the CCP. As simple as that. The trouble is that even my chimp brain trying to grasp the complexity of decentralizing AI knows its limits. No offense to chimps, who are provenly extremely intelligent creatures. I have to be careful here. In the near future, some Chimp Empire star with Neuralink’s implant may be reading this and decide to pay me a visit. You know like James Gandolfini in Life is Short: "I'm so sorry"
"I know." https://youtu.be/5KhmBPaXbyg?si=FYRc6L046BhwsXnh&embedable=true Decentralize AI - Between Being Crazy and Lazy Look if you want to play with ChatGPT, you can. If you want more, you have to pay. If you want your own private AI dancer, then you have to roll up your sleeves because these “toys” ain’t open source. I don’t buy that “possible risks” sh..t, but I respect private (intellectual) property in both virtual and real (physical) worlds, period. To all those cannot-get-satisfaction-without-decentralization AI guys, I have something to say: put the money where your mouths are. It’s the NDA hanging over my head, but my conscience because I was paid fair and square to be a “test pilot” on an AI project, preventing me from giving away more details. I saw first-hand what AI can do with search results. It may not be nice to your eyes, but it’s efficient in filling someone else’s pockets. Why do you say decentralize when all you want and think about is how to monetize? That’s why we have the two AI decentralization level problem. Are we talking about and demanding decentralization of artificial intelligence on a state or corporate level? Are we more concerned about the existential or business-related threats of complete AI decentralization? Jeffrey Epstein and AI Yes, I want you to keep on reading my story, that is why I threw in this shameless heading. And yes, Joe Rogan wanted people to keep on listening to his episode, so he threw in Jeffrey Epstein. For Peter Thiel, he was a tax (evasion) expert. For Bill Gates, allegedly, a marital advisor of all the things imaginable. For others, he was something so dark that I didn’t want even go there. The same applies to AI. What you want is what you get. That’s why it’s no surprise that if not the top, then one of the most asked questions to ChatGPT was about the ways to make money. You see, nobody or very few wanted to save or make our world a better place. It’s in our nature to be interested more in monetization than decentralization. We are selfish and AI is not, but it can be. “So God created mankind in his own image, in the image of God he created them.” (Genesis 1:27) So, if AI has to be and stay a “communist” according to Peter Thiel, for safety reasons, even though we all know and feel deep inside that’s not entirely true - I’m fine with it. I’m also fine with keeping the open (source) AI gates closed because I have every reason to believe that would be counterproductive for its further development and progress. We would turn AI into an ATM and stop, instead of keep going to see how far we can get and what more we can achieve with it. There’s no need to overanalyze why (not) to decentralize. All you have to do is look me straight in my face and say (write) in your comments that it won’t eventually come down to the money. There are the unavoidable “Three M’s” talking points in almost all of The Joe Rogan Experience episodes. At the least, in the most recent ones. Mushrooms Monkeys Microsoft Mushrooms Monkeys Microsoft So, if you want to see it, you have to hear it. There’s no other way around them. I was more than curious about what Peter Thiel had to say in episode #2190 . I just knew it would be juicy, otherwise, he would be talking to Lex Fridman, not Joe Rogan. episode #2190 Mushrooms. There’s a saying where I’m from: “Some mushrooms you eat only once.” But, we all know that Joe likes magic (psychedelic) mushrooms and other dimensions. I don’t know man, maybe I’m too old or too scared for that sh..t. Mushrooms. Monkeys. For some reason, Joe is obsessed with Netflix’s “Chimp Empire.” I don’t have to watch how chimps are “navigating complex social politics, family dynamics, and dangerous territory disputes.” Why? Because all I have to do is look at the world around me to know that the adults had left the room a long time ago. Monkeys. Netflix’s “Chimp Empire.” Microsoft. I’m on the same page with Joe, mostly. Here’s one of my award-winning stories about Linux , and here’s another one about Mr. Bill Gates . Please, don’t get this the wrong way, I’m nowhere near Tarantino’s “Kill Bill: Hypothetical Volume 3” mood. That’s why I was rolling my eyes when Peter Thiel started dwelling on his “theory” about Gates’ wealth and especially marriage. But, you have to love Joe because he really knows when and how to make the U-turn in the middle of a conversation smoothly and quickly. Microsoft. award-winning stories about Linux Mr. Bill Gates The next thing you know, Peter Thiel was talking about AI. To be more precise, the decentralized AI. the decentralized AI. Controversial But Brilliant Point Now, you have to take my word for it because it is a three-hour-long episode. So, I’m quoting (paraphrasing) to the best of my memory and respect: Crypto is libertarian, AI is communist Crypto is libertarian, AI is communist You can say or think what you want about this member of the PayPal Mafia , but Peter Thiel hit the nail on the head with the brilliant brutal force with this one. He also identified Silicon Valley as the root of all centralized AI evil correctly and unapologetically. the PayPal Mafia Honestly, I lost him a little bit when he started talking about China. So, I’m not quite sure if the decentralization of AI is a state or tech-level problem. Let me put it this way. If I were some AI genius and wanted to make money, I would already be hooked on Soylent . But, if I wanted to try out some crazy AI things without any ethical limits , I would already be a proud member of the CCP . As simple as that. Soylent ethical limits the CCP The trouble is that even my chimp brain trying to grasp the complexity of decentralizing AI knows its limits. No offense to chimps, who are provenly extremely intelligent creatures. I have to be careful here. In the near future, some Chimp Empire star with Neuralink’s implant may be reading this and decide to pay me a visit. You know like James Gandolfini in Life is Short: "I'm so sorry"
"I know." https://youtu.be/5KhmBPaXbyg?si=FYRc6L046BhwsXnh&embedable=true https://youtu.be/5KhmBPaXbyg?si=FYRc6L046BhwsXnh&embedable=true Decentralize AI - Between Being Crazy and Lazy Look if you want to play with ChatGPT, you can. If you want more, you have to pay. If you want your own private AI dancer, then you have to roll up your sleeves because these “toys” ain’t open source . I don’t buy that “possible risks” sh..t, but I respect private (intellectual) property in both virtual and real (physical) worlds, period. open source To all those cannot-get-satisfaction-without-decentralization AI guys, I have something to say: put the money where your mouths are. It’s the NDA hanging over my head, but my conscience because I was paid fair and square to be a “test pilot” on an AI project, preventing me from giving away more details. I saw first-hand what AI can do with search results. It may not be nice to your eyes, but it’s efficient in filling someone else’s pockets. Why do you say decentralize when all you want and think about is how to monetize? Why do you say decentralize when all you want and think about is how to monetize? That’s why we have the two AI decentralization level problem. Are we talking about and demanding decentralization of artificial intelligence on a state or corporate level? Are we more concerned about the existential or business-related threats of complete AI decentralization? Jeffrey Epstein and AI Yes, I want you to keep on reading my story, that is why I threw in this shameless heading. And yes, Joe Rogan wanted people to keep on listening to his episode, so he threw in Jeffrey Epstein. For Peter Thiel, he was a tax (evasion) expert. For Bill Gates, allegedly, a marital advisor of all the things imaginable. For others, he was something so dark that I didn’t want even go there. The same applies to AI. What you want is what you get. That’s why it’s no surprise that if not the top, then one of the most asked questions to ChatGPT was about the ways to make money. You see, nobody or very few wanted to save or make our world a better place. It’s in our nature to be interested more in monetization than decentralization. We are selfish and AI is not, but it can be. “So God created mankind in his own image, in the image of God he created them.” (Genesis 1:27) So, if AI has to be and stay a “communist” according to Peter Thiel, for safety reasons, even though we all know and feel deep inside that’s not entirely true - I’m fine with it. I’m also fine with keeping the open (source) AI gates closed because I have every reason to believe that would be counterproductive for its further development and progress. We would turn AI into an ATM and stop, instead of keep going to see how far we can get and what more we can achieve with it. There’s no need to overanalyze why (not) to decentralize. All you have to do is look me straight in my face and say (write) in your comments that it won’t eventually come down to the money.

Nuke Silicon Valley - Decentralize AI

Nuke Silicon Valley - Süni intellektin mərkəzsizləşdirilməsi

Nuke Silicon Valley: descentralizar la IA

Nuke Silicon Valley - Descentralizar a IA

シリコンバレーを破壊 - AIを分散化

Nuke Silicon Valley — decentralizēt AI

Nuke Silicon Valley - Arolla AI

Nuke Silicon Valley - AI nisqamanta descentralizay

Ikibaya cya Nuke Silicon - Kwegereza abaturage AI

Nuke Silicon Valley – Decentralizácia AI

Nuke Silicon Valley-Decentralize AI

This article was originally published on georgesiosi.com/writings/artificial-cultural-intelligence I recently stumbled upon a fascinating research paper titled Intelligence at the Edge of Chaos. It explores how AI systems trained on behaviors that exist between rigid order and total chaos—referred to as Class IV systems—tend to show enhanced reasoning and adaptability. This insight, which connects with the concept of balance and complexity, led me to think about how such models parallel the ways indigenous knowledge systems operate, like the fluid yet structured methods of Polynesian wayfinding. The Sweet Spot: Between Order & Chaos According to the paper, most Western-developed Large Language Models (LLMs), including ChatGPT and Claude, fall under Class II or Class III behaviors—either too rule-bound or creative yet unstructured. However, Class IV models find a middle ground, balancing predictability with adaptability. The fourth classification sounded strikingly similar to indigenous knowledge systems. For example, the systems required for ancient Polynesian navigators (wayfinders) who would read the stars, waves, and wind patterns — adapting their route across vast oceans without losing direction. They navigated by recognizing patterns in a complex, interconnected world (see ‘Legendary’ map of Pacific by James Cook’s Tahitian Navigator Tupaia below), much like the ideal adaptive (Class IV) AI would. The Intelligence at the Edge of Chaos research found that AI systems exposed to more complex rules— to chaotic yet not entirely random environments—showed a higher degree of intelligence. The analogy fits here: instead of adhering to rigid formulas or losing coherence in randomness, they manage to adapt, just as wayfinders do. Beyond Grammar: The Cultural Nuance of Communication LLMs that speak to you in your own language are a technical marvel, but cultural contexts can be a real challenge. What is grammatically correct in one language might be tone-deaf or even offensive in another. For example, consider how an AI might navigate the subtleties of addressing elders differently across cultures. Western-built LLMs tend to operate on fixed rules that may not always grasp these nuances, highlighting why a deeper, more adaptable form of intelligence is needed—one that’s culturally aware, context-sensitive, and can read between the lines. From Chess (West) to Weiqi/Go (East) This brings to mind the evolution of training deep learning models, initially focused on Western-style chess, a game rooted in strategy, predictability, and hierarchy. But to advance, AI researchers had to turn to Weiqi (or Go, above), an Eastern game that embodies complexity, adaptability, and a constantly shifting balance—hallmarks of Class IV systems. The shift from chess to Go was a metaphorical leap from rigid strategy to dynamic pattern recognition, reflecting how intelligent systems need to move beyond order into the realm of adaptive chaos to excel. Towards Artificial Cultural Intelligence (ACI) This concept of balancing complexity with coherence has broader implications. Imagine Artificial Cultural Intelligence (ACI), a new breed of AI that doesn’t just follow the rules or generate random responses but learns to adapt much like traditional ecological knowledge (TEK) systems. These systems maintain ecological balance through constant adaptation, reading environmental shifts, and making real-time adjustments. An ACI could be built to emulate this kind of (Class IV) responsive, intelligent adaptability, drawing insights not just from data but from the wisdom embedded in diverse cultural practices. Building AI that Understands & Adapts The future of AI could benefit from this culturally adaptive thinking. Imagine systems that can not only predict trends but also respect the cultural sensitivities of the regions they operate in, understanding not just what to say but how to say it. They would thrive in real-world applications where context and nuance are crucial, from negotiating diplomatic language to managing local ecological projects. Much like a master wayfinder adjusts to the waves, these systems would navigate the complexities of human interaction, balancing cultural wisdom with technological prowess. Conclusion: Bridging Cultures, Bridging Complexity The journey from rigid, rule-based AI to adaptable, culturally intelligent systems mirrors the transition from chess to Go, from order to dynamic balance. This is the promise of Intelligence at the Edge of Chaos—to build models that thrive not just in structured environments but in complex, interconnected realities. Indigenous knowledge systems have mastered this balance for centuries, guiding communities with a blend of adaptability and core principles (Class IV signals). If AI can learn to do the same, we might just see the dawn of Artificial Cultural Intelligence—a future where technology doesn’t just process data but understands, adapts, and, in its own way, learns to navigate the world as gracefully as a Wayfinder reading ocean waves. For more insights on this subject, check out the full paper on Intelligence at the Edge of Chaos. This article was originally published on georgesiosi.com/writings/artificial-cultural-intelligence This article was originally published on georgesiosi.com/writings/artificial-cultural-intelligence georgesiosi.com/writings/artificial-cultural-intelligence I recently stumbled upon a fascinating research paper titled Intelligence at the Edge of Chaos . It explores how AI systems trained on behaviors that exist between rigid order and total chaos—referred to as Class IV systems—tend to show enhanced reasoning and adaptability. This insight, which connects with the concept of balance and complexity, led me to think about how such models parallel the ways indigenous knowledge systems operate, like the fluid yet structured methods of Polynesian wayfinding. research paper Intelligence at the Edge of Chaos The Sweet Spot: Between Order & Chaos According to the paper, most Western-developed Large Language Models (LLMs), including ChatGPT and Claude, fall under Class II or Class III behaviors—either too rule-bound or creative yet unstructured. However, Class IV models find a middle ground, balancing predictability with adaptability. The fourth classification sounded strikingly similar to indigenous knowledge systems. For example, the systems required for ancient Polynesian navigators (wayfinders) who would read the stars, waves, and wind patterns — adapting their route across vast oceans without losing direction. They navigated by recognizing patterns in a complex, interconnected world (see ‘Legendary’ map of Pacific by James Cook’s Tahitian Navigator Tupaia below), much like the ideal adaptive (Class IV) AI would. ‘ Legendary’ map of Pacific by James Cook’s Tahitian Navigator Tupaia Legendary’ map of Pacific by James Cook’s Tahitian Navigator Tupaia The Intelligence at the Edge of Chaos research found that AI systems exposed to more complex rules— to chaotic yet not entirely random environments—showed a higher degree of intelligence. The analogy fits here: instead of adhering to rigid formulas or losing coherence in randomness, they manage to adapt, just as wayfinders do. Intelligence at the Edge of Chaos Beyond Grammar: The Cultural Nuance of Communication LLMs that speak to you in your own language are a technical marvel, but cultural contexts can be a real challenge. What is grammatically correct in one language might be tone-deaf or even offensive in another. For example, consider how an AI might navigate the subtleties of addressing elders differently across cultures. Western-built LLMs tend to operate on fixed rules that may not always grasp these nuances, highlighting why a deeper, more adaptable form of intelligence is needed—one that’s culturally aware, context-sensitive, and can read between the lines. From Chess (West) to Weiqi/Go (East) This brings to mind the evolution of training deep learning models, initially focused on Western-style chess, a game rooted in strategy, predictability, and hierarchy. But to advance, AI researchers had to turn to Weiqi (or Go, above), an Eastern game that embodies complexity, adaptability, and a constantly shifting balance—hallmarks of Class IV systems. The shift from chess to Go was a metaphorical leap from rigid strategy to dynamic pattern recognition, reflecting how intelligent systems need to move beyond order into the realm of adaptive chaos to excel. Towards Artificial Cultural Intelligence (ACI) This concept of balancing complexity with coherence has broader implications. Imagine Artificial Cultural Intelligence (ACI), a new breed of AI that doesn’t just follow the rules or generate random responses but learns to adapt much like traditional ecological knowledge (TEK) systems. These systems maintain ecological balance through constant adaptation, reading environmental shifts, and making real-time adjustments. An ACI could be built to emulate this kind of (Class IV) responsive, intelligent adaptability, drawing insights not just from data but from the wisdom embedded in diverse cultural practices. Building AI that Understands & Adapts The future of AI could benefit from this culturally adaptive thinking. Imagine systems that can not only predict trends but also respect the cultural sensitivities of the regions they operate in, understanding not just what to say but how to say it. They would thrive in real-world applications where context and nuance are crucial, from negotiating diplomatic language to managing local ecological projects. Much like a master wayfinder adjusts to the waves, these systems would navigate the complexities of human interaction, balancing cultural wisdom with technological prowess. Conclusion: Bridging Cultures, Bridging Complexity The journey from rigid, rule-based AI to adaptable, culturally intelligent systems mirrors the transition from chess to Go, from order to dynamic balance. This is the promise of Intelligence at the Edge of Chaos —to build models that thrive not just in structured environments but in complex, interconnected realities. Indigenous knowledge systems have mastered this balance for centuries, guiding communities with a blend of adaptability and core principles (Class IV signals). Intelligence at the Edge of Chaos If AI can learn to do the same, we might just see the dawn of Artificial Cultural Intelligence—a future where technology doesn’t just process data but understands, adapts, and, in its own way, learns to navigate the world as gracefully as a Wayfinder reading ocean waves. Cultural For more insights on this subject, check out the full paper on Intelligence at the Edge of Chaos. For more insights on this subject, check out the full paper on Intelligence at the Edge of Chaos . Intelligence at the Edge of Chaos

Artificial Cultural Intelligence: A Case for It

Stories, News & Mentions about SIFI

Sifi’s Stories on HackerNoon

Sifi's News & Mentions