Get live statistics and analysis of Arun Rao's profile on X / Twitter

One of my spiciest takes: the AI space is moving so fast I see a geographic gap growing between SF and every other place. The Peninsula is 2-3 months behind SF, while NYC and LA are 4-7 months behind, London is 12+, and other places are even worse. It’s highly uneven (eg parts of Stanford and UCB, or frontier lab offices, are closer to SF - but even the OpenAI NYC or MSL NYC offices are behind the SF locations). When cycle times for each generation are at most 2-3 months, being even 1-3 generations behind the frontier will have vast consequences as most of the learning about building and deploying these systems is experiential and passed over dinner conversations, late night debugging, hikes and walks, etc. What this means: All large tech companies need to have their GenAI and agent teams in person in SF. The startups need to move here from everywhere else, at least for the core nucleus of the organization. Geographic concentration and in-person agglomeration effects are quite important for this tech wave.

My favorite books and reading trace for 2024: BEST BOOKS OF 2024 Bennett, A Brief History of Intelligence: Evolution, AI, and the Five Breakthroughs That Made Our Brains. An exploration of the evolutionary milestones that shaped human intelligence.  The book goes from LUCA to bacteria and bilaterians, tetrapods to amniotes and therapsids, and then cynodonts to mammals and hominids. It had insights into how an understanding of earthly intelligence can inform the development of modern attempts at AGI and then SI. It covers breakthroughs in the evolution of intelligence, such as: steering, emoting, associating, and predicting; reinforcing, TD learning, and world models; simulations, generative models, and model-based RL; mentalizing and self/other modeling; language and large group coordination. I liked this book so much that I bought both a print and digital copy and re-read some chapters and sections multiple times. It itched at and etched on my brain and is a much-discussed book in the Bay Area AI circles. Alexander, Ishikawa, Silverstein, et al., A Pattern Language: Towns, Buildings, Construction. A seminal guide offering 253 design patterns to create balanced and livable spaces, emphasizing that thoughtful design can enhance human flourishing. This was a favorite text from my undergraduate years that I dug up to re-read while doing a painful ADU construction project this year. Darwin, The Autobiography of Charles Darwin: 1809-1882. A memoir revealing the life and intellectual journey of the naturalist. It shows the importance of curiosity, luck, and perseverance in scientific discovery. It is a glimpse into the mind of the most eminent Victorian I know (other than William Rowan Hamilton and the Faraday/Maxwell pair).

This might be coolest and most ambitious new city and construction project in the entire US, if not the Western world. Cheering on this team to push ahead and show California and the US how to build again.

I suspect that much of this is poor prototypes and the lack of deep thinking about use cases (doing something useful for customers and being able to charge for it). What I’ve seen is the Mag 7 have found a lot of value ($N billion in marginal revenue) and hence are investing $500bn in capex on this - it will take time for the rest of the Fortune 493 and smaller companies to figure this out, but often they can’t afford the talent that can execute on GenAI projects.

This is a problem w/ the science of evals in complex domains, not actually a model problem as models can be updated quickly once we create better evals. But for a complex domain like medicine, the point is we need much more sophisticated thinking on how to construct evals (ideally we need unanimous model juries and blind and random human doctor peer review).

🔥🔥🔥 “I never imagined Americans would be so meek,” she [a Chinese dissident] said. “By comparison, we were actually pretty brave.” For the record, I still think the US is by far the most open and pro free speech county in the world, but one must remember this is a daily fight. Governments will take away as much as they can, and citizens fight back through culture, media, humor, and the courts. True for every President and administration in US history.

One of the best media diets I’ve seen - I’d add Techmeme and Hacker News, but hard to beat this list. There are also some great AI news sources like AI News, The Batch, Import AI, EpochAI, & EmergentMind (for Arxiv technical papers).

Many are missing this major trend of finance leaving NYC - and it’s not just for the average finance professional. Some of the best hedge fund and top talent investors I know left NYC for other places, with Miami being the main winner. Some types of investment banking will stay in NYC, but much commercial banking, insurance, fintech, investment management, crypto, and more will have exited. It’s momentous as NYC has been the US financial capital since the 1890s and the global one since the 1950s. It was dominant till 2019, and now that agglomeration is fading away.

I’ve been waiting for this spreadsheet LLM paper for a while - it’s only worth it for Microsoft and Google to build this, given the proprietary nature of spreadsheets and their duopoly. I hope both companies can productize it ASAP - billions of dollars of value here as much of the financial and accounting worlds still run on spreadsheet and manual efforts.

One of my spiciest takes: the AI space is moving so fast I see a geographic gap growing between SF and every other place. The Peninsula is 2-3 months behind SF, while NYC and LA are 4-7 months behind, London is 12+, and other places are even worse. It’s highly uneven (eg parts of Stanford and UCB, or frontier lab offices, are closer to SF - but even the OpenAI NYC or MSL NYC offices are behind the SF locations). When cycle times for each generation are at most 2-3 months, being even 1-3 generations behind the frontier will have vast consequences as most of the learning about building and deploying these systems is experiential and passed over dinner conversations, late night debugging, hikes and walks, etc. What this means: All large tech companies need to have their GenAI and agent teams in person in SF. The startups need to move here from everywhere else, at least for the core nucleus of the organization. Geographic concentration and in-person agglomeration effects are quite important for this tech wave.

My favorite books and reading trace for 2024: BEST BOOKS OF 2024 Bennett, A Brief History of Intelligence: Evolution, AI, and the Five Breakthroughs That Made Our Brains. An exploration of the evolutionary milestones that shaped human intelligence.  The book goes from LUCA to bacteria and bilaterians, tetrapods to amniotes and therapsids, and then cynodonts to mammals and hominids. It had insights into how an understanding of earthly intelligence can inform the development of modern attempts at AGI and then SI. It covers breakthroughs in the evolution of intelligence, such as: steering, emoting, associating, and predicting; reinforcing, TD learning, and world models; simulations, generative models, and model-based RL; mentalizing and self/other modeling; language and large group coordination. I liked this book so much that I bought both a print and digital copy and re-read some chapters and sections multiple times. It itched at and etched on my brain and is a much-discussed book in the Bay Area AI circles. Alexander, Ishikawa, Silverstein, et al., A Pattern Language: Towns, Buildings, Construction. A seminal guide offering 253 design patterns to create balanced and livable spaces, emphasizing that thoughtful design can enhance human flourishing. This was a favorite text from my undergraduate years that I dug up to re-read while doing a painful ADU construction project this year. Darwin, The Autobiography of Charles Darwin: 1809-1882. A memoir revealing the life and intellectual journey of the naturalist. It shows the importance of curiosity, luck, and perseverance in scientific discovery. It is a glimpse into the mind of the most eminent Victorian I know (other than William Rowan Hamilton and the Faraday/Maxwell pair).

This is a problem w/ the science of evals in complex domains, not actually a model problem as models can be updated quickly once we create better evals. But for a complex domain like medicine, the point is we need much more sophisticated thinking on how to construct evals (ideally we need unanimous model juries and blind and random human doctor peer review).

I’ve been waiting for this spreadsheet LLM paper for a while - it’s only worth it for Microsoft and Google to build this, given the proprietary nature of spreadsheets and their duopoly. I hope both companies can productize it ASAP - billions of dollars of value here as much of the financial and accounting worlds still run on spreadsheet and manual efforts.

One of the best media diets I’ve seen - I’d add Techmeme and Hacker News, but hard to beat this list. There are also some great AI news sources like AI News, The Batch, Import AI, EpochAI, & EmergentMind (for Arxiv technical papers).

Disagree-yes, we need more science-y VCs, but the biggest bottleneck is always great founders, bridging engineering, PMF, sales, recruiting, managing/leading, and all the painful things founders have to do with grit and vision over many years. A great founding team is the rarest and most valuable resource.

Update: OpenAI claims it will comply with the “voluntary” EU AI Code, Meta says it won’t. Waiting for Google, X/Grok, and Anthropic. Basically OAI has taken Option 1 (fake compliance) vs Meta doing Option 3 (principled rejection).

On the upcoming UK Online Safety Act: “Many ministers would tell you — with much elation — that the OSA was exciting because it was the first law of its kind. Countries across the world would be looking to us as a leading example! But it turns out that pioneering comprehensive regulation of the internet when you don’t actually really understand the internet is not exactly best practice. With nowhere to look to understand what worked — and, crucially, what didn’t work — campaigners and policy experts attempting to argue that certain elements of the Bill would have awful unintended consequences had few case studies to point to. So their warnings about the risks were simply not accepted.” Link 🔗 👇

I suspect that much of this is poor prototypes and the lack of deep thinking about use cases (doing something useful for customers and being able to charge for it). What I’ve seen is the Mag 7 have found a lot of value ($N billion in marginal revenue) and hence are investing $500bn in capex on this - it will take time for the rest of the Fortune 493 and smaller companies to figure this out, but often they can’t afford the talent that can execute on GenAI projects.

🔥🔥🔥 “I never imagined Americans would be so meek,” she [a Chinese dissident] said. “By comparison, we were actually pretty brave.” For the record, I still think the US is by far the most open and pro free speech county in the world, but one must remember this is a daily fight. Governments will take away as much as they can, and citizens fight back through culture, media, humor, and the courts. True for every President and administration in US history.

The simple principle of “skin in the game” dictates that all public employees need to live in jurisdiction that they serve, especially police and firefighters and senior civil servants (I can tolerate exceptions for junior employees). This is an easy thing to fix with a city charter amendment.