Today in Nature Machine Intelligence, Kazuki Irie & I discuss 4 classic challenges for neural nets — systematic generalization, catastrophic forgetting, few-shot learning, & reasoning. We argue there is a unifying fix: the right incentives & practice. rdcu.be/eLRmg

Interested in research in my lab? intake.gureckislab.org/interest/

🚨 New preprint: "Decision rule inference limits social escape from learning traps" (with Rheza Budiono and Cate Hartley of the @hartleylabnyu.bsky.social ✨). Read here: osf.io/preprints/ps.... This is more work on a very curious phenomena!

Today we open-sourced a new project for developing behavioral experiments online. It is called Smile. Announcement of v0.1.0: todd.gureckislab.org/2025/07/22/s... Smile has been used internally in my lab for several years and has substantially increased our productivity.

I don't know what world hassabis is living in but the reality is the reverse. AI is creating a world whereby there's less trust (by making it difficult to differentiate real from ai generated), ever wider inequity gap, and ever more intrusive surveillance

Fantastic new work by @johnchen6.bsky.social (with @brendenlake.bsky.social and me trying not to cause too much trouble). We study systematic generalization in a safety setting and find LLMs struggle to consistently respond safely when we vary how we ask naive questions. More analyses in the paper!

New preprint alert! We often prompt ICL tasks using either demonstrations or instructions. How much does the form of the prompt matter to the task representation formed by a language model? Stick around to find out 1/N

my god

Out today in Nature Machine Intelligence! From childhood on, people can create novel, playful, and creative goals. Models have yet to capture this ability. We propose a new way to represent goals and report a model that can generate human-like goals in a playful setting... 1/N

The part of George Orwell’s 1984 that everyone forgets is how the music and publishing industries have been replaced by a machine that spits out songs and bad novels “without any human intervention.” The goal is to keep you from ever having to think.

I wrote about the concept of agency (both human and artificial) in the year 2025. gracewlindsay.com/2025/01/24/2...

Our paper on if you can incentivize rule induction in humans with money is finally out (answer is: it appears to be a very weak/0-ish effect in contrast to the huge effect of financial incentives on rote, repetitive tasks). credit to pamop, ben newell & dan bartels psycnet.apa.org/fulltext/202...

the rapid transition of academics off x (despite temporarily reducing reach/followers) makes you wonder what’s stopping us from ending the for-profit, closed access publishing industry. it’s, like…. we can just do it? or if not, interesting to consider what the inertial differences are.