I really want someone to study what 10 years of Minecraft from toddler age did to scores on these.

AI can take many forms, but it's clear that the chat format is the one that has truly taken off. I think this reveals a fundamental truth about human thought: we think best in dialogue, even when our counterpart isn’t human.

seems relevant

i think it's often better to just not have a philosophical theory about something than to go around embracing all of the absurd consequences of the one you adopted

Made a new assignment for a class on Computational Psycholinguistics: - I trained a Transformer language model on sentences sampled from a PCFG - The students' task: Given the Transformer, try to infer the PCFG (w/ a leaderboard for who got closest) Would recommend! 1/n

I’m willing to bet that reports of LLMs trying to prevent themselves being shut down are hallucinations of fictional stories of such events by the LLM.

Statements like this should make you more skeptical of how our society approaches statistical analysis and draws conclusions from it, rather than coming to disbelieve stuff which is clearly visible if you have one single ounce of serious social perception

LLMs are 5/10 writers. So the fact that you can reliably improve on the explanations in papers and books by asking an LLM to summarize them is a huge condemnation of academic writing.

(Continental philosopher): Without 5, we could not count to 6. (Me): Yes. (Continental philosopher): It could be said 6 is consequently in the 5-shadow, the forthcoming of 6 is 5-heralded, that 5 springs necessarily from the recapitulation of 6. (Me, checking my watch): Yes.

I am starting to think sycophancy is going to be a bigger problem than pure hallucination as LLMs improve. Models that won’t tell you directly when you are wrong (and justify your correctness) are ultimately more dangerous to decision-making than models that are sometimes wrong.

I am advertising 2 positions in language evolution with my group at Radboud University: 📢 Fully-funded PhD position (Robotics/AI 🤖) ru.nl/en/working-at/… 📢 Postdoc position (Virtual Reality 🎮) ru.nl/en/working-at/… Deadline for applications is August 14th. Please RT! :)

Fully-funded PhD position in experimental and/or computational psycholinguistics: tmalsburg.github.io/job_ad_2025_ph… Application deadline is August 15.

Our team is hiring a postdoc in (mech) interpretability! The ideal candidate will have research experience in interpretability for text and/or image generation models and be excited about open science! Please consider applying or sharing with colleagues: metacareers.com/jobs/222395396…

journals.ub.uni-koeln.de/index.php/phai…

If you're not abandoning at least 10% of the articles or books you start reading, you're not exercising good judgment and you're wasting your time

🌍Introducing BabyBabelLM: A Multilingual Benchmark of Developmentally Plausible Training Data! LLMs learn from vastly more data than humans ever experience. BabyLM challenges this paradigm by focusing on developmentally plausible data We extend this effort to 45 new languages!

New posts about phonemic awareness, structured literacy. New website: still a little rough! #scienceofreading #dyslexia seidenbergreading.net/blog