Cédric

w/ Tracey Mills, Ben Prytawski,@mhtessler, @noahdgoodman, @jacobandreas, Josh Tenenbaum Paper: arxiv.org/pdf/2509.00074 Play the games here: cedriccolas.com/demos/language…

Can knowledge accumulate across generations? We run iterated learning chains: each agent gets only 2 lives, then passes advice to the next Performance increases across generations: partial knowledge compounds through language, mirroring cultural evolution in human populations

New paper at ICLR 2026! 🎉 "Language and Experience: A Computational Model of Social Learning in Complex Tasks" We model how humans combine advice from others with direct experience to learn new tasks, and show this enables bidirectional human-AI knowledge transfer. 🧵⤵️

@RhysSullivan i will open source soon, so you can set it up for yourself

@RhysSullivan not all the way there but i just setup a three way interaction between a telegram bot, obsidian and claude code so you can chat with claude code about your notes from your phone via telegram, send it links and pdfs to process and it can co-manage your obsidian

i want a "notes app" that i can put thoughts / ideas in and claude picks them up and starts working on them in the background this is especially wanted for ideas that i think would be cool but don't want to have to setup, where an ai can implement it in the background

@paularambles they are the claudettes

claudette would be a beautiful name for a baby girl

this probably requires some kind of balancing so having more AIs than humans doesn’t bias goal evolution towards goals humans don’t care about maybe some kind of asymmetric social learning strategies: eg ai goals are rewarded when the resulting behaviors are helpful to humans

it’s not about replacing humans, but about adding new individuals in the cultural evolution process (of goals and behaviors), to catalyze it: more agents, more diversity of interests and skills -> more innovation, more creativity

agreed! we will only care about open-ended systems that co-evolve goals and behaviors **within hybrid human-ai populations** this means ai goals will be influenced by ours (so ai math will remain math), but also means our goals will be influenced by ai’s too

@_ivyzhang @akarshkumar0101 you guys should talk (assuming you don’t know each other yet)

In Boston til Saturday if anyone else is in town :D

@ChenSun92 cool thread! didn’t know there was open ended research in gdm canada, would love to hear more!

1/N Hello friends! With #NeurIPS2025 fading into memory and the holidays offering space for reflection, wanted to share my recent integrations of lessons learned over conversations, parties, and talks both at NeurIPS and beyond. While the twilight of the "scaling is all you need" era gives into what @ilyasut calls the "Age of Research" or @RichardSSutton calls the “Era of Experience” and many in the community are now firmly fixated on continual learning as the next frontier to tackle, the conversation has nicely shifted from what we should do, to how we actually build it.

Will the influx of synthetic data lead to uniform #ModelCollapse across the internet? Our recent #EMNLP2025 (Oral) paper suggests a nuanced picture: different collapse dynamics might emerge in different internet domains based on the properties of human data in those domains! 🧵

this seems like important work! can we figure out mechanisms of collective intelligence? can we recreate these conditions everywhere?