Joint Postdoc at Columbia @ZuckermanBrain and NYU @NYUPsych. Supported by @NYASciences. Prev at: Princeton PhD, RS Intern @Meta, Yale '19
Joined June 2013
- Tweets574
- Following366
- Followers2.5K
- Likes736
Pinned Tweet
Proud to share our new work with @marcelomattar and Lea Duncker, newly accepted to #NeurIPS2026 arxiv.org/abs/2609.17042. This work is a precursor to a forthcoming follow up that I hope will be one of the biggest papers of my career.
Check out our newly released dataset.
[Hugging Face release] We’re releasing the dataset behind Reason to Play! I’m incredibly proud of our team for reaching this milestone.
@sreejan_kumar Austin Andrews @LHuntNeuro @summerfieldlab Josh Tenenbaum @somnirons @marcelomattar @momchiltomov
After CCN/Cosyne I also thought about this but reached a different conclusion. It seems neuroscientists in perception are feeling their world *close in* (as @ShahabBakht writes) while those who work on motor control, RL/DM, natural behavior are seeing their world *open up.* 1/5
What does this have to do with AI? Recent AI has focused on these kinds of problems: multimodal, stateful agents that act, plan, use memory and tools over long timescales. That gives neuroscience a richer hypothesis space than fitting large DNNs to lots of image data 4/5
In some cases, this will be quite literally using the same models (plugging our own work here: xitter.cf/csaba_botos/status/205…). In others, there are going to be shared questions/frameworks. Though we're seeing changes, I'm even more confident in the future of NeuroAI. 5/5
Jeee 🐦⬛
I am very proud of our joint effort with @sreejan_kumar Kumar on the project "Reason to Play"
LRMs show human-like rule discovery, and their hidden states predict human brain activity during gameplay.
Interactive demo + paper:
botcs.github.io/reason-to-pl…
Sidestepping the materialism question, which seems to be in the domain of philosophy, there’s a related neuroscience question I asked Chris Olah about during his COSYNE keynote.
None of this answers the “spark” question about subjective experience. But even at the level of scientific explanation, biological and AI systems, at least when constructed without biological constraints specifically in mind, may not admit the same natural description.
Sreejan Kumar retweeted
We’re launching GPTZero 4o today, an all-purpose AI detection model that performs 1/10,402 false positives on benchmarks, and also the most robust AI detector model to edits and paraphrasing attacks 💪
Excited to share this work has been accepted to #NeurIPS2026! Looking forward to presenting this with @csaba_botos
Jeee 🐦⬛
I am very proud of our joint effort with @sreejan_kumar Kumar on the project "Reason to Play"
LRMs show human-like rule discovery, and their hidden states predict human brain activity during gameplay.
Interactive demo + paper:
botcs.github.io/reason-to-pl…
