Coding tools do seem to be making software development less social in ways that can be costly even if you’re shipping more code. Instead of a constant back-and-forth while a feature is being built, an eng can fully ship a massive PR relatively autonomously. This sounds nice but can lead to a nightmare review process and for the shipped code to go off the rails if the ticket was underspecified (normally this sort of problem would be corrected earlier through social feedback). Similarly, the social back-and-forth helps you identify when something is overly complicated and probably should be re-architected, whereas an LLM has no problem writing this type of code. Very important to maintain human collaboration and communication as long as humans are (still) in the loop! For now we're still doing standups.

Just published in @PNASNews, we resolve a 50-year-old riddle from Richard Feynman's handwritten notes, prove and generalize it, and run a large-scale human study to reveal near-optimal heuristics in sequential decision problems: doi.org/10.1073/pnas.2509612…

got the power of god, anime and cats on our side @RoundtableHQ_

(1/n) Thrilled to launch our new preprint -- led by the brilliant Milena Rmus (@milenamr7) and with superstars Matt Hardy (@mdahardy) and Tom Griffiths (@cocosci_lab)

After AlphaGo, the skill of human Go players noticeably improved. I suspect we will see a similar pattern in math.

You should try pair vibe coding. People have developed totally different approaches to using the models.

Another instance of Moravec: For me, Claude is much better at back-end work than front-end. In FE, issues often show up only after visually inspecting/using a component. Whereas in back-end, feedback is already text-based (logs) and Claude can work relatively autonomously.

Introducing the PROCESS Turing Test. It's time for Proof of Human

seems like room for improvement in making AIs more human-like by training on human (including neural) processes, not just outputs - and also potential metrics to hillclimb

The new Turing test from @RoundtableHQ_

My submission to the Dwarkesh essay competition, also available on Substack now. Omaha as judgment day for AGI I answered Prompt 3: how do the labs start making money

Coming from Boston this never gets old

Packed room with @StanfordPsych! A new generation of researchers are using Proof of Human to verify the authenticity of all their published data. We got to give them a sneak peek of a recent research paper too…more to come here :) And thank you to the audience for amazing questions on human identity, both in practice and as an intellectual field of study

I always knew academia-to-entreprenurship would be my path, but it took me a while to gain the confidence. I go in-depth into my story here, starting from my Indian-American cultural heritage and finding out my main strength to be interdisciplinary synthesis. Why I Turned Down a Tenure-Track Professorship, now out on Minds, Machines, and Markets: substack.com/home/post/p-195…

What I don't get about bread culture is why it's so much cheaper in Europe compared to the US. If you go the bakery in Paris, you can get a baguette for ~1€. Whereas mass-produced American bread is like $3. Shouldn't the mass produced option be cheaper?