Utkarsh Gupta retweeted
An honour for @bold_lab_ai to host the one and only @SchmidhuberAI for a packed talk today @UniofOxford
We had an interesting conversation about "what will the future be" vs "what do we want the future to be". Or in other words: what's the point of compression and understanding the universe if not to take action to change its course?
Utkarsh Gupta retweeted
We @bold_lab_ai are hiring a large cohort of home and overseas students for Fall 2027 entry. If you’re excited about discovering the next paradigm of open-source AI and want to work on ambitious, exploratory, but scalable ML that empowers humanity, please apply by 1st of December 2026 AOE to maximise chances of funding: eng.ox.ac.uk/study/research-…
Utkarsh Gupta retweeted
Science is built by standing on the shoulders of giants. Today we're releasing the Scientific Contribution Graph, a massive dataset of how current technologies and scientific results were built upon previous results, with applications in AI-4-Science and Science-of-Science.
Paper: arxiv.org/abs/2605.15011
Data and API (80GB): github.com/cognitiveailab/sc…
Live demo: huggingface.co/spaces/pajans…
Utkarsh Gupta retweeted
A looped transformer has to know when to stop, and its state has to stay bounded while it loops.
We introduce Spherical Fixed-Point Reasoners (SFPRM) 🌐: loop on the unit sphere, where each step is an angle, and stop once the angle is small enough.
🧵 (1/9)
Utkarsh Gupta retweeted
We @bold_lab_ai are hiring a large cohort of home and overseas students for Fall 2027 entry. If you’re excited about discovering the next paradigm of open-source AI and want to work on ambitious, exploratory, but scalable ML, please apply. More application details and exact timeline coming soon, in the meantime please register your interest here bold-lab.ai/opportunities/ 🦋🦋🦋
Utkarsh Gupta retweeted
BOLD is recruiting -- we are looking for a Senior HPC Systems Administrator: my.corehr.com/pls/uoxrecruit…
Deadline is this Friday (16th October) at noon UK time.
Come help us build @bold_lab_ai and re-invent academia 🦋🚀
Utkarsh Gupta retweeted
We are again recruiting @bold_lab_ai - please share the love 🙏. We are looking for:
1) Postdocs (my.corehr.com/pls/uoxrecruit…) -- deadline 9th of Oct at noon
2) research assistants (my.corehr.com/pls/uoxrecruit…) -- deadline 9th of Oct at noon
3) Strategic Partnership Project Manager (my.corehr.com/pls/uoxrecruit…) -- deadline 14th of Oct at noon
Utkarsh Gupta retweeted
AI is ravenous and is eating mathematicians' lunch. A long-standing problem is solved, others are in its sights. But is this more than just a problem about problems? Here's Fields Medallist James Maynard.
Read the Fields Medallists' Declaration: mathandai.org/
Utkarsh Gupta retweeted
Mathematics has now fully morphed into biology:
Project costs millions $ ✅
Put out paper that nobody has read ✅
Massive paper supplement nobody will read ✅
Advertise with pretty art ✅
Vicious authorship fight ✅
Say you're curing cancer ✅
Utkarsh Gupta retweeted
🚀 I’m hiring! We’re looking for a Research Assistant in AI Safety to join us at @OATML_Oxford and work with @yaringal and me on generative AI, safety & security.
📅 Apply by 11 September 2026
🔗 cs.ox.ac.uk/news/2543-full.h…
Please share with anyone who might be interested!
Utkarsh Gupta retweeted
A student asked me about my habit of bike riding or walking in the forest as a way of gaining inspiration. I thought I'd share my response:
I suspect everyone is different but all of our brains need time to process information. Some of this happens in sleep -- the idea that if you have a hard problem that you should "sleep on it" is very true. I often wake up with an inspiration.
You need to give your brain time to do its thing. You can't control inspiration. All you can do is create opportunities for it to occur.
I often have ideas in a dark conference hall while half-listening to a boring talk. I'll have spent the day at posters and learning about new work. Sitting in the dark and letting my mind drift is when those fragments come together into something new.
I will often do some work early in the morning and hit a snag. Then I go for a bike ride in the forest. I am not intentionally thinking about my problem -- I'm in the moment. My body is moving in the world. But part of my brain must be working on the problem because within 15-20 minutes, I have the solution.
The brain is amazing. Cramming it with information all the time may feel productive. But feeding it information is very different from letting it be creative. You can't force it to be creative and if you are force feeding it, it can't be creative. You need to leave it alone. Let it wander. Let it make connections. Give it space. Trust it.
Utkarsh Gupta retweeted
Predicting the answer to interventional "what if?" questions — the outcome of an action you never took — need a *mechanistic* model, not a curve fit. And you can only learn one by *experimenting*. Experiments are costly, so the real game is **data efficiency**.
Meet the Model Discovery Agent (MDA). 🧵
Utkarsh Gupta retweeted
We can finally talk about it:
We found a way to extract hidden reasoning of frontier models using a vulnerability in the APIs of every frontier AI company.
We verified that our reasoning token count matches billed API thinking tokens 1:1 for most of the prompts we queried.
Utkarsh Gupta retweeted
Just saw this: MiniMax reproduced our ICLR 2025 paper, Learning Dynamics of LLM Finetuning, in 12 hours. I spent weeks debugging a stupid matrix concatenation bug back then. Now coding agents fix things like this in seconds. Amazing how fast things change.
minimax.io/blog/minimax-m3
I'll be at @icmlconf in #Seoul giving invited talks at two workshops on July 11:
* Uncertainty in Agentic Systems (10:00 @ CONF ROOM E1 - E4)
* LM4Plan (13:00 @ Grand Ballroom 101-102)
Also - I'm hiring for postdocs to work on Scientific Discovery @OATML_Oxford!
Come say hi!
Utkarsh Gupta retweeted
Will be in Seoul next week 🇰🇷 - keen to chat everything AI4Science, autonomous discovery, benchmarking agents, and KBBQ! DM if you are too
Pleased to share that "MADE: Benchmark Environments for Closed-Loop Materials Discovery" was accepted at #ICML2026! 🇰🇷
We introduce a modular, open-source framework for evaluating AI scientists and algorithms at end-to-end, closed-loop materials discovery. 🤖
Thread 🧵(1/5)
Utkarsh Gupta retweeted
Our work on hybrid physics-ML simulators for electron diffraction is now out in @NatureComms! ⚛️
A differentiable, GPU-accelerated, open-source package for dynamical structural refinement.🔬
Thread 🧵(1/5)
Utkarsh Gupta retweeted
From AI for bioscience and medicine to causality, next-generation AI agents, and LLMs, our @icmlconf 2026 papers explore how machine learning can move beyond benchmarks toward bio-discovery, empowering clinical decision-making, and real-world impact.
🚀 Proud to share our new paper: MADE, a benchmark for closed-loop materials discovery🧪🤖
Discovery isn’t static: it’s adaptive, nonstationary & constrained
MADE tests LLM agents etc under feedback loops, OOD shifts & real-world constraints: pushing beyond regurgitating facts
Pleased to share that "MADE: Benchmark Environments for Closed-Loop Materials Discovery" was accepted at #ICML2026! 🇰🇷
We introduce a modular, open-source framework for evaluating AI scientists and algorithms at end-to-end, closed-loop materials discovery. 🤖
Thread 🧵(1/5)
Utkarsh Gupta retweeted
Pleased to share that "MADE: Benchmark Environments for Closed-Loop Materials Discovery" was accepted at #ICML2026! 🇰🇷
We introduce a modular, open-source framework for evaluating AI scientists and algorithms at end-to-end, closed-loop materials discovery. 🤖
Thread 🧵(1/5)

























