More than two months after shutting down India's own X rival, Koo, Co-founder Mayank Bidawatka has embarked on a new entrepreneurial journey with his latest startup, Billion Hearts Software Technologies.

Koo co-founder Mayank Bidawatka introduces his new startup Billion Hearts, raises $250,000

Mayank Bidawatka is set to launch his new venture, Billion Hearts and also planning to raise nearly $500,000 funding

Koo co-founder Mayank Bidawatka finalizes pre-seed for new startup

Koo is an Indian microblogging and social networking service, owned by Bangalore-based Bombinate Technologies. It was co-founded by entrepreneurs Aprameya Radhakrishna and Mayank Bidawatka. The app was launched in early 2020; it won the government&#x27;s Atmanirbhar App Innovation Challenge which selected the best apps from some 7,000 entries across the country. As of November 2022, the company ...

Koo (social network) - Wikipedia

Koo, which debuted in early 2020, won the government&#x27;s Atmanirbhar App Innovation Challenge. Know more about its company profile, history, etc.

Koo App - Made in India Social Networking Platform

Koo - Social platform offering vernacular microblogging. Raised a total funding of $65.7M over 7 rounds from 101 investors.

Koo - Company Profile - Tracxn

Know what Celebrities &amp; Millions of People are talking about, in your Language.

Koo: Know What's Happening! - Apps on Google Play

Koo | 31,478 followers on LinkedIn. Uniting the world in its languages! The safest &amp; friendliest social network. | Koo is all about uniting the world despite the language barriers. The dream of one world. Globally connected citizens of humanity.

Koo | LinkedIn

Glassdoor gives you an inside look at what it&#x27;s like to work at Koo India, including salaries, reviews, office photos, and more. This is the Koo India company profile. All content is posted anonymously by employees working at Koo India.

Working at Koo India | Glassdoor

Discover KOO’s wide range of sunglasses and goggles. For cycling, MTB, skiing, snowboarding, or simply enjoying some outdoor time. Shop Now.

KOO Eyewear US | Born in Italy. Worn around the world.

A decade after Twitter emerged into a globe-spanning hotbed of speech that helped fuel protest movements including the Arab Spring, the rise of Koo is a reflection of the global reckoning over social media. Companies including Facebook and Twitter are frequently criticized for not doing enough ...

Koo is India’s homegrown social network and is having success ...

I had the chance to catch up with Katy Koob, the newly appointed Vice President of Revenue at Dastan, the parent company of Decrypt and Rug Radio. With a background in leading media sales teams and a passion for decentralized economies, Katy shares her vision for the future of media, the challenges of the current Web2 model, and how Web3 technologies can transform the industry. This appointment comes at a time when revenue-focused leadership roles, such as Chief Revenue Officers and Growth Marketers, are experiencing a surge in demand, as highlighted in LinkedIn's "LinkedIn Jobs on the Rise 2024" report. Dastan's decision to prioritize revenue optimization aligns with broader industry trends and reflects its commitment to staying ahead of the curve in an evolving market. Q1: Congratulations on your new role at Dastan. Can you tell us about your journey and how it led you to this position? Thank you! My journey has been in the brand partnerships world, both in traditional and digital media, where I worked my way up to VP West Coast at Refinery29. During my time there, I was instrumental in scaling the business through strategic partnerships, innovative digital monetization strategies and expanding the company’s market presence globally. I’ve always been passionate about technology and its potential to reimagine industries and shift audience behaviors. My interest in decentralized economies and blockchain is long standing, I’ve been an early crypto investor and Web3 advisor. My combination of traditional media experience and futurist passion for the next phase of the internet made the opportunity at Dastan a perfect fit. Q2: What excites you the most about joining Dastan and working with brands like Decrypt and Rug Radio? The most exciting aspect is the combination of two of the most powerful brands in the space (Rug Radio & Decrypt) revolutionizing the media industry. The traditional Web2 media model is broken, with big tech controlling data and distribution, it’s no longer sustainable for creators and publishers. At Dastan, we’re building a decentralized media network that equitably benefits creators, consumers, publishers, and advertisers. Our innovative approach with blockchain and tokenized economies allows for a more equitable value exchange. I'm thrilled to be part of a company that's building a new model, and putting vitality back into digital media. Q3: How do you see Web3 technologies transforming the media landscape? Web3 is about ownership and the technologies have the potential to create a more transparent, efficient and mutually incentivized media ecosystem. By leveraging blockchain, we can ensure content creators are fairly compensated and have control over their creation and distribution. Tokenization enables new revenue streams and a direct relationship between creators and their audience. This transformation can break the monopoly of big tech and empower individuals, making the media industry more inclusive and sustainable. Q4: Can you tell us more about Dastan’s Content+ and its impact on users? Content+ is an application designed to transform how users interact with digital content. With over 100 million users in our ecosystem, including Rug Radio, Decrypt news, GG, Scene, Emerge, Decrypt U, and Degen Alley, Content+ we’ll be incentivizing all players involved through tokenized engagement. It’s about creating a seamless, rewarding experience for our users while ensuring creators get their share of the value they generate. This is a crucial step in our journey to build the world’s largest decentralized media network. Q5: What are your goals as the new Head of Revenue at Dastan? My primary goal is to help steward in the next phase of media while driving our revenue strategies and enhancing monetization opportunities for brands, creators and audiences alike. I aim to build thoughtful, strategic partnerships with leading advertisers and ensure that our partners share in the value creation for their customers. By leveraging my expertise and experience, I plan to solidify Dastan’s position as a leader in the Web3 media and creator economy sector, setting the new industry standard. Q6: What message would you like to share with the readers of HackerNoon? We’re at a pivotal moment in media where we need to reinvent, and with blockchain, the technology exists to do it. At Dastan, we’re building the new model. I encourage everyone to embrace the possibilities of Web3 and decentralized technologies. Together, we can create a more equitable, innovative, and sustainable media ecosystem that benefits everyone involved. I had the chance to catch up with Katy Koob , the newly appointed Vice President of Revenue at Dastan, the parent company of Decrypt and Rug Radio . With a background in leading media sales teams and a passion for decentralized economies, Katy shares her vision for the future of media, the challenges of the current Web2 model, and how Web3 technologies can transform the industry. Katy Koob Decrypt Decrypt Rug Radio This appointment comes at a time when revenue-focused leadership roles, such as Chief Revenue Officers and Growth Marketers, are experiencing a surge in demand, as highlighted in LinkedIn's "LinkedIn Jobs on the Rise 2024 " report. Dastan's decision to prioritize revenue optimization aligns with broader industry trends and reflects its commitment to staying ahead of the curve in an evolving market. "LinkedIn Jobs on the Rise 2024 "LinkedIn Jobs on the Rise 2024 Q1: Congratulations on your new role at Dastan. Can you tell us about your journey and how it led you to this position? Q1: Congratulations on your new role at Dastan. Can you tell us about your journey and how it led you to this position? Thank you! My journey has been in the brand partnerships world, both in traditional and digital media, where I worked my way up to VP West Coast at Refinery29. During my time there, I was instrumental in scaling the business through strategic partnerships, innovative digital monetization strategies and expanding the company’s market presence globally. I’ve always been passionate about technology and its potential to reimagine industries and shift audience behaviors. My interest in decentralized economies and blockchain is long standing, I’ve been an early crypto investor and Web3 advisor. My combination of traditional media experience and futurist passion for the next phase of the internet made the opportunity at Dastan a perfect fit. Q2: What excites you the most about joining Dastan and working with brands like Decrypt and Rug Radio? Q2: What excites you the most about joining Dastan and working with brands like Decrypt and Rug Radio? The most exciting aspect is the combination of two of the most powerful brands in the space (Rug Radio & Decrypt) revolutionizing the media industry. The traditional Web2 media model is broken, with big tech controlling data and distribution, it’s no longer sustainable for creators and publishers. At Dastan, we’re building a decentralized media network that equitably benefits creators, consumers, publishers, and advertisers. Our innovative approach with blockchain and tokenized economies allows for a more equitable value exchange. I'm thrilled to be part of a company that's building a new model, and putting vitality back into digital media. At Dastan, we’re building a decentralized media network that equitably benefits creators, consumers, publishers, and advertisers. Our innovative approach with blockchain and tokenized economies allows for a more equitable value exchange. I'm thrilled to be part of a company that's building a new model, and putting vitality back into digital media. Q3: How do you see Web3 technologies transforming the media landscape? Q3: How do you see Web3 technologies transforming the media landscape? Web3 is about ownership and the technologies have the potential to create a more transparent, efficient and mutually incentivized media ecosystem. By leveraging blockchain, we can ensure content creators are fairly compensated and have control over their creation and distribution. Tokenization enables new revenue streams and a direct relationship between creators and their audience. This transformation can break the monopoly of big tech and empower individuals, making the media industry more inclusive and sustainable. Q4: Can you tell us more about Dastan’s Content+ and its impact on users? Q4: Can you tell us more about Dastan’s Content+ and its impact on users? Content+ is an application designed to transform how users interact with digital content. With over 100 million users in our ecosystem, including Rug Radio, Decrypt news, GG, Scene, Emerge, Decrypt U, and Degen Alley, Content+ we’ll be incentivizing all players involved through tokenized engagement. It’s about creating a seamless, rewarding experience for our users while ensuring creators get their share of the value they generate. This is a crucial step in our journey to build the world’s largest decentralized media network. Content+ Q5: What are your goals as the new Head of Revenue at Dastan? Q5: What are your goals as the new Head of Revenue at Dastan? My primary goal is to help steward in the next phase of media while driving our revenue strategies and enhancing monetization opportunities for brands, creators and audiences alike. I aim to build thoughtful, strategic partnerships with leading advertisers and ensure that our partners share in the value creation for their customers. By leveraging my expertise and experience, I plan to solidify Dastan’s position as a leader in the Web3 media and creator economy sector, setting the new industry standard. By leveraging my expertise and experience, I plan to solidify Dastan’s position as a leader in the Web3 media and creator economy sector, setting the new industry standard. Q6: What message would you like to share with the readers of HackerNoon? Q6: What message would you like to share with the readers of HackerNoon? We’re at a pivotal moment in media where we need to reinvent, and with blockchain, the technology exists to do it. At Dastan, we’re building the new model. I encourage everyone to embrace the possibilities of Web3 and decentralized technologies. Together, we can create a more equitable, innovative, and sustainable media ecosystem that benefits everyone involved.

Media Moves: Katy Koob Joins Dastan to Drive Decentralized Growth

KoolKits ( ubernetes t ) are highly-opinionated, language-specific, batteries-included debug container images for Kubernetes. In practice, they’re what you would’ve installed on your production pods if you were stuck during a tough debug session in an unfamiliar shell. K oolkits To briefly give some background, note that these container images are intended for use with the new , which spins up for interactive troubleshooting. A KoolKit will be pulled by kubectl debug, spun up as a container in your pod, and have the ability to access the same process namespace as your original container. kubectl debug feature Ephemeral containers Since production containers are usually , using a KoolKit enables you to troubleshoot with power tools instead of relying on what was left behind due to the generosity (or carelessness) of whoever originally built the production image. rather bare The tools in each KoolKit were carefully selected, and you can read more about the motivation behind this entire project . below If you just want to take a look at the good stuff, feel free to check out the full project . on GitHub Debugging Kubernetes is Hard It’s not trivial to understand what’s going on inside a Kubernetes pod. First of all, your application is not a single entity anymore – it is comprised of multiple pods, replicated for horizontal scaling, and sometimes even scattered across multiple clusters. Furthermore, to access your application with local tools (like debuggers) you need to deal with pesky networking issues like discovery and port forwarding, which slows down the use of such tools. And, the crown jewel of the distributed systems world-altering the state of or completely halting the running pod (e.g. when placing a breakpoint) might cause cascading failures in other parts of your system, which will exacerbate the existing problem. The Motivation Behind KoolKits Lightrun was built with Kubernetes in mind – we work across multiple pods, multiple clusters, and even multiple clouds. We understood early on that packing a punch by using the right tools is a great source of power for the troubleshooting developer – and we figured we’d find a way to give back to the community somehow – and that’s how we came up with the idea for KoolKits. Let’s dive deep for a second to explain why KoolKits can be pretty useful: There’s a that states that one should build small container images. This makes sense for a few different reasons: well-known Kubernetes best practice Building the image will consume fewer resources (aka CI hours) Pulling the image will take less time (who wants to pay for so much ingress anyways?) Less stuff means less surface area exposed to security vulnerabilities, in a world where even no-op logging isn’t safe anymore There’s also a lot of tooling in existence that helps you get there without doing too much heavy lifting: base images are super small Alpine Linux go a step further and remove everything but the runtime DistroLess Docker images help create thin final production images Docker multi-stage builds The problem starts when you’re trying to debug what’s happening inside those containers. By using a small production image you’re forsaking a large number of tools that are when wrapping your head around a problem in your application. invaluable By using a KoolKit, you’re allowing yourself the benefits of a small production image without compromising on quality tools – each KoolKit contains hand-picked tools for the specific runtime it represents, in addition to a more generic set of tooling for Linux-based systems. P.S. KoolKits was inspired by and . kubespy netshoot Considerations There are quite a few decisions we made during the construction of these images – some things we took into consideration are listed below. Size of Images KoolKits Docker images tend to run, uhm, rather . large KoolKits are intended to be downloaded once, kept in the cluster’s Docker registry, and then spun up immediately on demand as containers. Since they’re not intended for constant pulling, and since they’re intended to be packed with goodies, this is a side effect we’re willing to endure. Using Ubuntu base images Part of the reason it’s hard to create a really slim image is due to our decision to go with a full Ubuntu 20.04 system as the basis for each KoolKit. This mainly came from our desire to replicate the same environment you would debug with locally inside your clusters. For example, this means no messing around with Alpine alternatives to normal Ubuntu packages you’re used to working with. Actually, this means we have a way of including tools that have no Alpine versions in each KoolKit. Using language version managers Each KoolKit uses (wherever possible) a language version manager instead of relying on language-specific distros. This is done to allow you to install older runtime versions easily, and in order to allow you to swap between runtime versions at will (for example, to get specific versions of tooling that only exist for specific runtime versions), as need be. Available KoolKits Each of the folders in the repo contains the Dockerfile behind the KoolKit and a short explanation of the debug image. All KoolKits are based on the base image, since real people need real shells. ubuntu:20.04 The list of available KoolKits: – AdoptOpenJDK 17.0.2 & related tooling (including jabba for easy version management and Maven 3.8.4) koolkit-jvm – Node 16.13.1 & related tooling (including nvm for easy version management) koolkit-node – Python 3.10.2 & related tooling (including pyenv for easy version management) koolkit-python Note that you don’t actually have to build them yourselves – all KoolKits are hosted publicly on Docker Hub and available free of charge. KoolKits Coming up A whole new, Go 1.17.7 KoolKit JVM KoolKit – , support jvm-profiler jHiccup Node.js KoolKit – , support llnode thetool Python KoolKit – , support vardbg memprof Contribution We’d be more than happy to add tools we missed to any image – just or to suggest one. open a pull request an issue First Published here

Exploring an Open Source Toolkit for Debugging Kubernetes - KoolKits

Cover Photo by Hassan Saleh on Unsplash I'll begin establish some baselines that I use to describe the words in the title. I use Paul G.s definition of a startup, " ", and while comparing them with cults, I'm not talking about doing cult-like things, but having a cult-like behaviour at an organisational level. For this, I will be using the "drinking the kool-aid" phrase which is associated with the . startup = growth Jonestown cult At the start of everyone's careers, people have different perceptions about how companies are. Through the power of the internet, and through connections, I knew I didn't want to join the bureaucracy and dehumanisation of corporate life. On the other hand, I was also slightly put off by the cult-like behaviour put forth by most startups. (In open-source, the same behaviour is visible with maintainers and core contributors.) I did eventually make the decision to work with startups, and this article is the result of that experience. Startups = Growth In large companies, one can afford not being aligned with the larger vision or not understanding the impact of each task, but the same cannot be the case when you're working for a startup. The kind of conviction, dedication, and commitment needed to pull off what a startup aspires requires cult-like focus. This is also why a startups don't have managers, or hire them way later. Managers are there to make sure the work gets done and everyone is meeting their quota, which, in a startup environment isn't (and shouldn't be) required. You HAVE to drink the kool-aid, or at least have your personal goals aligned with that of the startup, otherwise the startup world will chew and throw you out. On the other hand, if your magnetic lines are being reinforced by where you work, it's like a 🚀 for your personal growth and career strategy, as your work will stand out among the relatively little competition in the startup world. And this is why, if you're a growth-oriented person, you should work at a startup. Startups = (Calculated) Risk Startups are inherently risky. This risk means that startups are trying to attract a different kind of people. People who are willing to take calculated risks, and those who will have the most impact in helping the startup achieve their goals. So, an outward appearance of cult-like behaviour help them attract the "right" kind of people. This combination of a lot of smart risk-taking individuals working towards a common goal can also be seen as "culty" by the outside world. This is also why startups appreciate talent and are more likely to hire people who can hit the ground running. Startups = Care About their Employees In an optimal startup environment, everyone is aligned and self-driven. They are trying to create a huge outcome, and they want to achieve huge personal and professional growth in the process. The people who joined in the first year and stayed on all these years are all leaders managing a whole team or part of the company. These are all the initial folk who are experienced, smart, driven. But above all they are dependable and are aligned in their thinking of how the organisation should grow. This also works because the startup also depends on the work of these employees. An employee is more likely to reach an irreplaceable stage at a smaller company than a larger one, and that's where startups come into the picture. To match the value that such an employee brings, the startup would have to train a new hire from scratch to reach that level. This is also why startups appreciate talent and are more likely to hire people who can hit the ground running. Conclusion Well, those are my thoughts for now 🙏. Writing is thinking, and I'm open to have discussions. Even if it convinces 1 person to join a true startup instead of an MNC or large corp, it has done its job right. If you liked the post, why not If you need to discuss startups or tech, feel free to reach out through (for free). follow me on Twitter. Superpeer These articles aren't written by a single person. Thanks to Preetam Nath, Deb Mukherjee, Shirish Joshi and Madhavi Swamy for helping proof-read this article and providing some of the discussion points which were used in the article. Also emailed via Substack .

Startups = Cults: Why You Should Drink the Kool Aid

are Hacker Noon’s way of getting to know — from a community perspective —  what matters in tech today. So, we asked our Noonie Nominees to tell us. The Noonies Among the 2,000+ deserving humans nominated across 5 categories for over 200 award titles, we discovered Manthan Koolwal from India, who’s has been nominated for a 2020 Noonie in the categories. Without further ado, we present to you, our big techy world, from the perspective of Manthan. Technology and Software Development 1. Which 2020 Noonie have you been nominated for? Hacker Noon Contributor of the Year - API Hacker Noon Contributor of the Year - DATA SCIENCE 2. Tell us a bit about yourself. I am a founder at makcorps.com, scrapingdog.com and flightapi.io. I love making APIs and I have worked on several backend projects in the last 5 years. I will be launching a face recognition AI-enabled API soon. 3. Tell us about the things you make/ write /manage / build. I usually write about data collection and visualization. I have built three APIs - hotel api, flight api and web scraping api. 4. What are you most excited about right now? About my future APIs which are AI-enabled. Also, my first AI project. I am super excited about this project. 5. What are you worried about right now? The economy. 6. What's the most useful advice you've ever given somebody? Be consistent and most importantly, keep patience. 7. How has the pandemic changed your life and/or career? I have learned to work from home. 8. If we gave you $10 million to invest in one thing right now, where would you put it? In a sports car company running on LPG or hydrogen. 9. What's an opinion you have that most people don't agree with? I am a polymath and many times my friends and even my family  members advised me to just focus on one single sector. 10. What or Who are you keeping an eye on in 2020? There are few startups funded by large firms in India. Hapramp is one of them. Also, I'm keeping an eye out for a covid vaccine. 12. Which apps can't you live without? Reddit and many ebook apps. 13. What are you currently learning? AI and ML For much the same reasons Hacker Noon decided not to put good ideas behind pop-ups or paywalls — nor abuse your personal data to target you with creepy ads — we also decided that you don’t have to be an influencer or have 50k followers on Twitter to earn the recognition that comes with a 2020 Noonie Nomination. Make somebody’s day and nominate them to be recognized in the internet’s most independent and community-driven awards: . NOONIES.TECH in the following awards: Don't forget to vote for Manthal Hacker Noon Contributor of the Year - API Hacker Noon Contributor of the Year - DATA SCIENCE

Manthan Koolwal, 2020 Noonie Nominee, is Developing AI-Enabled APIs

ADA: A Powerful Data Augmentation Technique for Improved Regression Robustness

Optimizing Prompts with LLMs: Key Findings and Future Directions

Authors:
(1) Chengrun Yang, Google DeepMind and Equal contribution;
(2) Xuezhi Wang, Google DeepMind;
(3) Yifeng Lu, Google DeepMind;
(4) Hanxiao Liu, Google DeepMind;
(5) Quoc V. Le, Google DeepMind;
(6) Denny Zhou, Google DeepMind;
(7) Xinyun Chen, Google DeepMind and Equal contribution. Table of Links Abstract and 1. Introduction 2 Opro: Llm as the Optimizer and 2.1 Desirables of Optimization by Llms 2.2 Meta-Prompt Design 3 Motivating Example: Mathematical Optimization and 3.1 Linear Regression 3.2 Traveling Salesman Problem (TSP) 4 Application: Prompt Optimization and 4.1 Problem Setup 4.2 Meta-Prompt Design 5 Prompt Optimization Experiments and 5.1 Evaluation Setup 5.2 Main Results 5.3 Ablation Studies 5.4 Overfitting Analysis in Prompt Optimization and 5.5 Comparison with Evoprompt 6 Related Work 7 Conclusion, Acknowledgments and References A Some Failure Cases B Prompting Formats for Scorer Llm C Meta-Prompts and C.1 Meta-Prompt for Math Optimization C.2 Meta-Prompt for Prompt Optimization D Prompt Optimization Curves on the Remaining Bbh Tasks E Prompt Optimization on Bbh Tasks – Tabulated Accuracies and Found Instructions 3.2 TRAVELING SALESMAN PROBLEM (TSP) Next, we consider the Traveling Salesman Problem (TSP) (Jünger et al., 1995; Gutin & Punnen, 2006), a classical combinatorial optimization problem with numerous algorithms proposed in literature, including heuristic algorithms and solvers (Rosenkrantz et al., 1977; Golden et al., 1980; Optimization et al., 2020; Applegate et al., 2006; Helsgaun, 2017), and approaches based on training deep neural networks (Kool et al., 2019; Deudon et al., 2018; Chen & Tian, 2019; Nazari et al., 2018). Specifically, given a set of n nodes with their coordinates, the TSP task is to find the shortest route that traverses all nodes from the starting node and finally returns to the starting node. Our optimization process with LLMs starts from 5 randomly generated solutions, and each optimization step produces at most 8 new solutions. We present the meta-prompt in Figure 20 of Appendix C.1. We generate the problem instances by sampling n nodes with both x and y coordinates in [−100, 100]. We use the Gurobi solver (Optimization et al., 2020) to construct the oracle solutions and compute the optimality gap for all approaches, where the optimality gap is defined as the difference between the distance in the solution constructed by the evaluated approach and the distance achieved by the oracle solution, divided by the distance of the oracle solution. Besides evaluating OPRO with different LLMs including text-bison, gpt-3.5-turbo and gpt-4, we also compare OPRO to the following heuristics: On the other hand, the performance of OPRO degrades dramatically on problems with larger sizes. When n = 10, all LLMs find the optimal solutions for every evaluated problem; as the problem size gets larger, the OPRO optimality gaps increase quickly, and the farthest insertion heuristic starts to outperform all LLMs in the optimality gap. Limitations. We would like to note that OPRO is designed for neither outperforming the stateof-the-art gradient-based optimization algorithms for continuous mathematical optimization, nor surpassing the performance of specialized solvers for classical combinatorial optimization problems such as TSP. Instead, the goal is to demonstrate that LLMs are able to optimize different kinds of objective functions simply through prompting, and reach the global optimum for some smallscale problems. Our evaluation reveals several limitations of OPRO for mathematical optimization. Specifically, the length limit of the LLM context window makes it hard to fit large-scale optimization problem descriptions in the prompt, e.g., linear regression with high-dimensional data, and traveling salesman problems with a large set of nodes to visit. In addition, the optimization landscape of some objective functions are too bumpy for the LLM to propose a correct descending direction, causing the optimization to get stuck halfway. We further elaborate our observed failure cases in Appendix A. This paper is available on arxiv under CC0 1.0 DEED license. Authors: (1) Chengrun Yang, Google DeepMind and Equal contribution; (2) Xuezhi Wang, Google DeepMind; (3) Yifeng Lu, Google DeepMind; (4) Hanxiao Liu, Google DeepMind; (5) Quoc V. Le, Google DeepMind; (6) Denny Zhou, Google DeepMind; (7) Xinyun Chen, Google DeepMind and Equal contribution. Authors: Authors: (1) Chengrun Yang, Google DeepMind and Equal contribution; (2) Xuezhi Wang, Google DeepMind; (3) Yifeng Lu, Google DeepMind; (4) Hanxiao Liu, Google DeepMind; (5) Quoc V. Le, Google DeepMind; (6) Denny Zhou, Google DeepMind; (7) Xinyun Chen, Google DeepMind and Equal contribution. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2 Opro: Llm as the Optimizer and 2.1 Desirables of Optimization by Llms 2 Opro: Llm as the Optimizer and 2.1 Desirables of Optimization by Llms 2.2 Meta-Prompt Design 2.2 Meta-Prompt Design 3 Motivating Example: Mathematical Optimization and 3.1 Linear Regression 3 Motivating Example: Mathematical Optimization and 3.1 Linear Regression 3.2 Traveling Salesman Problem (TSP) 3.2 Traveling Salesman Problem (TSP) 4 Application: Prompt Optimization and 4.1 Problem Setup 4 Application: Prompt Optimization and 4.1 Problem Setup 4.2 Meta-Prompt Design 4.2 Meta-Prompt Design 5 Prompt Optimization Experiments and 5.1 Evaluation Setup 5 Prompt Optimization Experiments and 5.1 Evaluation Setup 5.2 Main Results 5.2 Main Results 5.3 Ablation Studies 5.3 Ablation Studies 5.4 Overfitting Analysis in Prompt Optimization and 5.5 Comparison with Evoprompt 5.4 Overfitting Analysis in Prompt Optimization and 5.5 Comparison with Evoprompt 6 Related Work 6 Related Work 7 Conclusion, Acknowledgments and References 7 Conclusion, Acknowledgments and References A Some Failure Cases A Some Failure Cases B Prompting Formats for Scorer Llm B Prompting Formats for Scorer Llm C Meta-Prompts and C.1 Meta-Prompt for Math Optimization C Meta-Prompts and C.1 Meta-Prompt for Math Optimization C.2 Meta-Prompt for Prompt Optimization C.2 Meta-Prompt for Prompt Optimization D Prompt Optimization Curves on the Remaining Bbh Tasks D Prompt Optimization Curves on the Remaining Bbh Tasks E Prompt Optimization on Bbh Tasks – Tabulated Accuracies and Found Instructions E Prompt Optimization on Bbh Tasks – Tabulated Accuracies and Found Instructions 3.2 TRAVELING SALESMAN PROBLEM (TSP) Next, we consider the Traveling Salesman Problem (TSP) (Jünger et al., 1995; Gutin & Punnen, 2006), a classical combinatorial optimization problem with numerous algorithms proposed in literature, including heuristic algorithms and solvers (Rosenkrantz et al., 1977; Golden et al., 1980; Optimization et al., 2020; Applegate et al., 2006; Helsgaun, 2017), and approaches based on training deep neural networks (Kool et al., 2019; Deudon et al., 2018; Chen & Tian, 2019; Nazari et al., 2018). Specifically, given a set of n nodes with their coordinates, the TSP task is to find the shortest route that traverses all nodes from the starting node and finally returns to the starting node. Our optimization process with LLMs starts from 5 randomly generated solutions, and each optimization step produces at most 8 new solutions. We present the meta-prompt in Figure 20 of Appendix C.1. We generate the problem instances by sampling n nodes with both x and y coordinates in [−100, 100]. We use the Gurobi solver (Optimization et al., 2020) to construct the oracle solutions and compute the optimality gap for all approaches, where the optimality gap is defined as the difference between the distance in the solution constructed by the evaluated approach and the distance achieved by the oracle solution, divided by the distance of the oracle solution. Besides evaluating OPRO with different LLMs including text-bison, gpt-3.5-turbo and gpt-4, we also compare OPRO to the following heuristics: On the other hand, the performance of OPRO degrades dramatically on problems with larger sizes. When n = 10, all LLMs find the optimal solutions for every evaluated problem; as the problem size gets larger, the OPRO optimality gaps increase quickly, and the farthest insertion heuristic starts to outperform all LLMs in the optimality gap. Limitations. We would like to note that OPRO is designed for neither outperforming the stateof-the-art gradient-based optimization algorithms for continuous mathematical optimization, nor surpassing the performance of specialized solvers for classical combinatorial optimization problems such as TSP. Instead, the goal is to demonstrate that LLMs are able to optimize different kinds of objective functions simply through prompting, and reach the global optimum for some smallscale problems. Our evaluation reveals several limitations of OPRO for mathematical optimization. Specifically, the length limit of the LLM context window makes it hard to fit large-scale optimization problem descriptions in the prompt, e.g., linear regression with high-dimensional data, and traveling salesman problems with a large set of nodes to visit. In addition, the optimization landscape of some objective functions are too bumpy for the LLM to propose a correct descending direction, causing the optimization to get stuck halfway. We further elaborate our observed failure cases in Appendix A. Limitations. This paper is available on arxiv under CC0 1.0 DEED license. This paper is available on arxiv under CC0 1.0 DEED license. available on arxiv

LLMs vs. Heuristics: Tackling the Traveling Salesman Problem (TSP)

Andrea Renzetti, Department of Economics, Alma Mater Studiorium Universit`a di Bologna, Piazza Scaravilli 2, 40126 Bologna, Italy. Table of Links Abstract and Introduction Theory coherent TVP-VAR Forecasting with the TC-TVP-VAR Response analysis at the ZLB with the TC-TVP-VAR Conclusion and References A Appendix 5 Conclusion TVP-VAR are flexible statistical models used both for prediction and policy analysis in macroeconomics. Despite their flexibility allows to capture changes in the dynamic relationship among the macroeconomic variables, these models can easily become too flexible with the risk of over fitting the data. This translate into poor forecasting performances and imprecise inference on the time-varying parameters and of typical objects of interests such as the impulse responses. On the other side, models from the economic theory typically provide a more tightly parameterized representation of the macroeconomy and therefore have the opposite tendency of fitting the data rather poorly. This paper exploits the restrictions implied by economic theory to formulate a prior for the parameters of TVP-VARs so as to enhance inference within this class of models. It does so by introducing a shrinkage prior that centers the time varying coefficients at each time period on the cross equation restrictions implied by an underlying economic theory about the variables in the system. The paper shows that “economic shrinkage” can be successfully used to obtain more accurate forecasts and more precise estimates of typical objects of interests such as the impulse responses. Future research For future research the prior proposed in this paper could be modified to accommodate for multiple competing theories `a la Loria et al. (2022). As well, the model could be extended to embed stochastic volatility, adapting the algorithm in Bognanni (2018). More in general, restrictions from economic theory could successfully be used to reduce overfitting and sharpen inference in non-parametric VARs and Gaussian Procesess VARs (Hauzenberger et al. 2022). References Ascari, Guido, Paolo Bonomolo, and Qazi Haque. 2023. “The long-run phillips curve is... a curve.” Ascari, Guido, and Argia M. Sbordone. 2014. “The Macroeconomics of Trend Inflation.” Journal of Economic Literature 52 (3): 679–739. Belmonte, Miguel A.G., Gary Koop, and Dimitris Korobilis. 2014. “Hierarchical Shrinkage in Time-Varying Parameter Models.” Journal of Forecasting 33 (1): 80–94. Benati, Luca, and Thomas A. Lubik. 2023. Impulse Response Analysis at the Zero Lower Bound. Diskussionsschriften dp2306. Universitaet Bern, Departement Volkswirtschaft, June. Bitto, Angela, and Sylvia Fr¨uhwirth-Schnatter. 2019. “Achieving shrinkage in a time-varying parameter model framework.” Annals Issue in Honor of John Geweke “Complexity and Big Data in Economics and Finance: Recent Developments from a Bayesian Perspective”, Journal of Econometrics 210 (1): 75–97. Bognanni, Mark. 2018. A Class of Time-Varying Parameter Structural VARs for Inference under Exact or Set Identification. Working Papers (Old Series) 1811. Federal Reserve Bank of Cleveland. Cagliarini, Adam, and Mariano Kulish. 2013. “Solving Linear Rational Expectations Models with Predictable Structural Changes.” The Review of Economics and Statistics 95, no. 1 (March): 328–336. Carriero, Andrea, Todd E. Clark, and Massimiliano Marcellino. 2021. “No-arbitrage priors, drifting volatilities, and the term structure of interest rates.” Journal of Applied Econometrics 36 (5): 495–516. Carter, C. K., and R. Kohn. 1994. “On Gibbs Sampling for State Space Models.” Biometrika 81 (3): 541–553. Chan, Joshua C.C., Eric Eisenstat, and Rodney W. Strachan. 2020. “Reducing the state space dimension in a large TVP-VAR.” Journal of Econometrics 218 (1): 105–118. Chan, Joshua CC, and Ivan Jeliazkov. 2009. “Efficient simulation and integrated likelihood estimation in state space models.” International Journal of Mathematical Modelling and Numerical Optimisation 1 (1-2): 101–120. Chen, Han, Vasco C´urdia, and Andrea Ferrero. 2012. “The Macroeconomic Effects of Large-scale Asset Purchase Programmes*.” The Economic Journal 122 (564): F289–F315. Cogley, Timothy, and Thomas Sargent. 2002. “Evolving Post-World War II US Inflation Dynamics.” In NBER Macroeconomics Annual 2001, Volume 16, 331–388. National Bureau of Economic Research, Inc. Cogley, Timothy, and Thomas J. Sargent. 2005. “Drifts and volatilities: monetary policies and outcomes in the post WWII US.” Monetary Policy and Learning, Review of Economic Dynamics 8 (2): 262–302. Cogley, Timothy, and Argia M. Sbordone. 2008. “Trend Inflation, Indexation, and Inflation Persistence in the New Keynesian Phillips Curve.” American Economic Review 98, no. 5 (December): 2101–26. Coulombe, Philippe Goulet. 2021. Time-Varying Parameters as Ridge Regressions. D’Agostino, Antonello, Luca Gambetti, and Domenico Giannone. 2013. “Macroeconomic forecasting and structural change.” Journal of Applied Econometrics 28 (1): 82–101. Debortoli, Davide, Jordi Gal´ı, and Luca Gambetti. 2019. “On the Empirical (Ir)relevance of the Zero Lower Bound Constraint.” In NBER Macroeconomics Annual 2019, volume 34, 141–170. NBER Chapters. National Bureau of Economic Research, Inc, October. Del Negro, Marco, Marc P. Giannoni, and Frank Schorfheide. 2015. “Inflation in the Great Recession and New Keynesian Models.” American Economic Journal: Macroeconomics 7, no. 1 (January): 168–96. Del Negro, Marco, and Frank Schorfheide. 2004. “Priors from General Equilibrium Models for VARS*.” International Economic Review 45 (2): 643–673. Farmer, Roger E.A., Daniel F. Waggoner, and Tao Zha. 2009. “Understanding Markov-switching rational expectations models.” Journal of Economic Theory 144 (5): 1849–1867. Fessler, Pirmin, and Maximilian Kasy. 2019. “How to Use Economic Theory to Improve Estimators: Shrinking Toward Theoretical Restrictions.” The Review of Economics and Statistics 101, no. 4 (October): 681–698. Fr¨uhwirth-Schnatter, Sylvia, and Helga Wagner. 2010. “Stochastic model specification search for Gaussian and partial non-Gaussian state space models.” Journal of Econometrics 154 (1): 85–100. Gallant, A. Ronald, and Robert E. McCulloch. 2009. “On the Determination of General Scientific Models With Application to Asset Pricing.” Journal of the American Statistical Association 104 (485): 117–131. Giacomini, Raffaella, and Giuseppe Ragusa. 2014. “Theory-coherent forecasting.” Causality, Prediction, and Specification Analysis: Recent Advances and Future Directions, Journal of Econometrics 182 (1): 145–155. Giannone, Domenico, Michele Lenza, and Giorgio E. Primiceri. 2015. “Prior Selection for Vector Autoregressions.” The Review of Economics and Statistics 97, no. 2 (May): 436–451. Hauzenberger, Niko, Florian Huber, Massimiliano Marcellino, and Nico Petz. 2022. Gaussian Process Vector Autoregressions and Macroeconomic Uncertainty. Ingram, Beth F., and Charles H. Whiteman. 1994. “Supplanting the ’Minnesota’ prior: Forecasting macroeconomic time series using real business cycle model priors.” Journal of Monetary Economics 34, no. 3 (December): 497–510. Kalli, Maria, and Jim E. Griffin. 2014. “Time-varying sparsity in dynamic regression models.” Journal of Econometrics 178 (2): 779–793. Loria, Francesca, Christian Matthes, and Mu-Chun Wang. 2022. “Economic theories and macroeconomic reality.” Journal of Monetary Economics 126:105–117. Milani, Fabio. 2007. “Expectations, learning and macroeconomic persistence.” Journal of Monetary Economics 54 (7): 2065–2082. Primiceri, Giorgio E. 2005. “Time varying structural vector autoregressions and monetary policy.” The Review of Economic Studies 72 (3): 821–852. Sims, Christopher. 2002. “Solving Linear Rational Expectations Models.” Computational Economics 20 (1-2): 1–20. Sims, Christopher A. 1980. “Macroeconomics and Reality.” Econometrica 48 (1): 1–48. Smets, Frank, and Rafael Wouters. 2007. “Shocks and Frictions in US Business Cycles: A Bayesian DSGE Approach.” American Economic Review 97, no. 3 (June): 586–606. Smith Jr., A. A. 1993. “Estimating nonlinear time-series models using simulated vector autoregressions.” Journal of Applied Econometrics 8 (S1): S63–S84. Stevanovic, Dalibor. 2016. “Common time variation of parameters in reduced-form macroeconomic models.” Studies in Nonlinear Dynamics & Econometrics 20, no. 2 (April): 159– 183. Stock, James H., and Mark W. Watson. 2001. “Vector Autoregressions.” Journal of Economic Perspectives 15, no. 4 (December): 101–115. Wind, Joris de, and Luca Gambetti. 2014. Reduced-rank time-varying vector autoregressions. CPB Discussion Paper 270. CPB Netherlands Bureau for Economic Policy Analysis, March. Woodford, Michael. 2003. Interest and Prices: Foundations of a Theory of Monetary Policy. Princeton University Press. This paper is available on arxiv under CC 4.0 license. Andrea Renzetti, Department of Economics, Alma Mater Studiorium Universit`a di Bologna, Piazza Scaravilli 2, 40126 Bologna, Italy. Andrea Renzetti, Department of Economics, Alma Mater Studiorium Universit`a di Bologna, Piazza Scaravilli 2, 40126 Bologna, Italy. Table of Links Abstract and Introduction Abstract and Introduction Theory coherent TVP-VAR Theory coherent TVP-VAR Forecasting with the TC-TVP-VAR Forecasting with the TC-TVP-VAR Response analysis at the ZLB with the TC-TVP-VAR Response analysis at the ZLB with the TC-TVP-VAR Conclusion and References Conclusion and References A Appendix A Appendix 5 Conclusion TVP-VAR are flexible statistical models used both for prediction and policy analysis in macroeconomics. Despite their flexibility allows to capture changes in the dynamic relationship among the macroeconomic variables, these models can easily become too flexible with the risk of over fitting the data. This translate into poor forecasting performances and imprecise inference on the time-varying parameters and of typical objects of interests such as the impulse responses. On the other side, models from the economic theory typically provide a more tightly parameterized representation of the macroeconomy and therefore have the opposite tendency of fitting the data rather poorly. This paper exploits the restrictions implied by economic theory to formulate a prior for the parameters of TVP-VARs so as to enhance inference within this class of models. It does so by introducing a shrinkage prior that centers the time varying coefficients at each time period on the cross equation restrictions implied by an underlying economic theory about the variables in the system. The paper shows that “economic shrinkage” can be successfully used to obtain more accurate forecasts and more precise estimates of typical objects of interests such as the impulse responses. Future research For future research the prior proposed in this paper could be modified to accommodate for multiple competing theories `a la Loria et al. (2022). As well, the model could be extended to embed stochastic volatility, adapting the algorithm in Bognanni (2018). More in general, restrictions from economic theory could successfully be used to reduce overfitting and sharpen inference in non-parametric VARs and Gaussian Procesess VARs (Hauzenberger et al. 2022). Future research References Ascari, Guido, Paolo Bonomolo, and Qazi Haque. 2023. “The long-run phillips curve is... a curve.” Ascari, Guido, and Argia M. Sbordone. 2014. “The Macroeconomics of Trend Inflation.” Journal of Economic Literature 52 (3): 679–739. Belmonte, Miguel A.G., Gary Koop, and Dimitris Korobilis. 2014. “Hierarchical Shrinkage in Time-Varying Parameter Models.” Journal of Forecasting 33 (1): 80–94. Benati, Luca, and Thomas A. Lubik. 2023. Impulse Response Analysis at the Zero Lower Bound. Diskussionsschriften dp2306. Universitaet Bern, Departement Volkswirtschaft, June. Bitto, Angela, and Sylvia Fr¨uhwirth-Schnatter. 2019. “Achieving shrinkage in a time-varying parameter model framework.” Annals Issue in Honor of John Geweke “Complexity and Big Data in Economics and Finance: Recent Developments from a Bayesian Perspective”, Journal of Econometrics 210 (1): 75–97. Bognanni, Mark. 2018. A Class of Time-Varying Parameter Structural VARs for Inference under Exact or Set Identification. Working Papers (Old Series) 1811. Federal Reserve Bank of Cleveland. Cagliarini, Adam, and Mariano Kulish. 2013. “Solving Linear Rational Expectations Models with Predictable Structural Changes.” The Review of Economics and Statistics 95, no. 1 (March): 328–336. Carriero, Andrea, Todd E. Clark, and Massimiliano Marcellino. 2021. “No-arbitrage priors, drifting volatilities, and the term structure of interest rates.” Journal of Applied Econometrics 36 (5): 495–516. Carter, C. K., and R. Kohn. 1994. “On Gibbs Sampling for State Space Models.” Biometrika 81 (3): 541–553. Chan, Joshua C.C., Eric Eisenstat, and Rodney W. Strachan. 2020. “Reducing the state space dimension in a large TVP-VAR.” Journal of Econometrics 218 (1): 105–118. Chan, Joshua CC, and Ivan Jeliazkov. 2009. “Efficient simulation and integrated likelihood estimation in state space models.” International Journal of Mathematical Modelling and Numerical Optimisation 1 (1-2): 101–120. Chen, Han, Vasco C´urdia, and Andrea Ferrero. 2012. “The Macroeconomic Effects of Large-scale Asset Purchase Programmes*.” The Economic Journal 122 (564): F289–F315. Cogley, Timothy, and Thomas Sargent. 2002. “Evolving Post-World War II US Inflation Dynamics.” In NBER Macroeconomics Annual 2001, Volume 16, 331–388. National Bureau of Economic Research, Inc. Cogley, Timothy, and Thomas J. Sargent. 2005. “Drifts and volatilities: monetary policies and outcomes in the post WWII US.” Monetary Policy and Learning, Review of Economic Dynamics 8 (2): 262–302. Cogley, Timothy, and Argia M. Sbordone. 2008. “Trend Inflation, Indexation, and Inflation Persistence in the New Keynesian Phillips Curve.” American Economic Review 98, no. 5 (December): 2101–26. Coulombe, Philippe Goulet. 2021. Time-Varying Parameters as Ridge Regressions. D’Agostino, Antonello, Luca Gambetti, and Domenico Giannone. 2013. “Macroeconomic forecasting and structural change.” Journal of Applied Econometrics 28 (1): 82–101. Debortoli, Davide, Jordi Gal´ı, and Luca Gambetti. 2019. “On the Empirical (Ir)relevance of the Zero Lower Bound Constraint.” In NBER Macroeconomics Annual 2019, volume 34, 141–170. NBER Chapters. National Bureau of Economic Research, Inc, October. Del Negro, Marco, Marc P. Giannoni, and Frank Schorfheide. 2015. “Inflation in the Great Recession and New Keynesian Models.” American Economic Journal: Macroeconomics 7, no. 1 (January): 168–96. Del Negro, Marco, and Frank Schorfheide. 2004. “Priors from General Equilibrium Models for VARS*.” International Economic Review 45 (2): 643–673. Farmer, Roger E.A., Daniel F. Waggoner, and Tao Zha. 2009. “Understanding Markov-switching rational expectations models.” Journal of Economic Theory 144 (5): 1849–1867. Fessler, Pirmin, and Maximilian Kasy. 2019. “How to Use Economic Theory to Improve Estimators: Shrinking Toward Theoretical Restrictions.” The Review of Economics and Statistics 101, no. 4 (October): 681–698. Fr¨uhwirth-Schnatter, Sylvia, and Helga Wagner. 2010. “Stochastic model specification search for Gaussian and partial non-Gaussian state space models.” Journal of Econometrics 154 (1): 85–100. Gallant, A. Ronald, and Robert E. McCulloch. 2009. “On the Determination of General Scientific Models With Application to Asset Pricing.” Journal of the American Statistical Association 104 (485): 117–131. Giacomini, Raffaella, and Giuseppe Ragusa. 2014. “Theory-coherent forecasting.” Causality, Prediction, and Specification Analysis: Recent Advances and Future Directions, Journal of Econometrics 182 (1): 145–155. Giannone, Domenico, Michele Lenza, and Giorgio E. Primiceri. 2015. “Prior Selection for Vector Autoregressions.” The Review of Economics and Statistics 97, no. 2 (May): 436–451. Hauzenberger, Niko, Florian Huber, Massimiliano Marcellino, and Nico Petz. 2022. Gaussian Process Vector Autoregressions and Macroeconomic Uncertainty. Ingram, Beth F., and Charles H. Whiteman. 1994. “Supplanting the ’Minnesota’ prior: Forecasting macroeconomic time series using real business cycle model priors.” Journal of Monetary Economics 34, no. 3 (December): 497–510. Kalli, Maria, and Jim E. Griffin. 2014. “Time-varying sparsity in dynamic regression models.” Journal of Econometrics 178 (2): 779–793. Loria, Francesca, Christian Matthes, and Mu-Chun Wang. 2022. “Economic theories and macroeconomic reality.” Journal of Monetary Economics 126:105–117. Milani, Fabio. 2007. “Expectations, learning and macroeconomic persistence.” Journal of Monetary Economics 54 (7): 2065–2082. Primiceri, Giorgio E. 2005. “Time varying structural vector autoregressions and monetary policy.” The Review of Economic Studies 72 (3): 821–852. Sims, Christopher. 2002. “Solving Linear Rational Expectations Models.” Computational Economics 20 (1-2): 1–20. Sims, Christopher A. 1980. “Macroeconomics and Reality.” Econometrica 48 (1): 1–48. Smets, Frank, and Rafael Wouters. 2007. “Shocks and Frictions in US Business Cycles: A Bayesian DSGE Approach.” American Economic Review 97, no. 3 (June): 586–606. Smith Jr., A. A. 1993. “Estimating nonlinear time-series models using simulated vector autoregressions.” Journal of Applied Econometrics 8 (S1): S63–S84. Stevanovic, Dalibor. 2016. “Common time variation of parameters in reduced-form macroeconomic models.” Studies in Nonlinear Dynamics & Econometrics 20, no. 2 (April): 159– 183. Stock, James H., and Mark W. Watson. 2001. “Vector Autoregressions.” Journal of Economic Perspectives 15, no. 4 (December): 101–115. Wind, Joris de, and Luca Gambetti. 2014. Reduced-rank time-varying vector autoregressions. CPB Discussion Paper 270. CPB Netherlands Bureau for Economic Policy Analysis, March. Woodford, Michael. 2003. Interest and Prices: Foundations of a Theory of Monetary Policy. Princeton University Press. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

Stories, News & Mentions about KOO

Koo’s Stories on HackerNoon

Koo's News & Mentions