Replacing scientists with AI isn’t just unlikely, it’s a bad design goal. The better path is collaborative science. Let AI explore the ideas, draft hypotheses, surface evidence, and propose checks. Let humans decide what matters, set standards, and judge what counts as discovery.

❓ Does an LLM know thyself? 🪞 Humans pass the mirror test at ~18 months 👶 But what about LLMs? Can they recognize their own writing — or even admit authorship at all? In our new paper, we put 10 state-of-the-art models to the test. Read on 👇 1/n 🧵

"Intriguing speculations on the possibilities of science"! AutoDiscovery is now being applied to cancer research, epidemiology, social and biogeochemical sciences, & more! Full talk is live (bonus hat in the pic 🧙🏼): majumderb.com/AutoDiscovery.… w Dhruv Agarwal Ai2

Recruiting PhDs & postdocs for: 🤖 agents "taking over" science (hypogenic.ai and 📌) 🧪 Real scientists ➡️AI (e.g., materials, chem, physics) 📜 Theory + incentives for H-AI collab & credit (e.g., formalizing tacit knowledge) new adventures for me, 🔄 if you can! 🙌

After two years of work, we’ve made an AI Scientist that runs for days and makes genuine discoveries. Working with external collaborators, we report seven externally validated discoveries across multiple fields. It is available right now for anyone to use. 1/5

Redesigning an LLM course that goes from perceptrons 👁️ to the geometry of the residual stream 📐! Heavily weighting in-class assessments, due to AI tools. (2024 was project-focused) Would be grateful for any exams/etc. folks can share, this is new territory for me! Please♻️!!

I'm not even sure whether spending $200 and let AI do all kinds of stuff is good design anyway. I think human/expert intervention is still critical in a lot of those intermediate processes.

I find it surprising that, while foundation models can predict so many valuable labels—enzyme activity, weather, cancer—we haven’t learned much new science from them. Why is it so hard to leverage strong predictive models to produce new discoveries? One answer is that prediction

I recently received an email titled “An 18-year-old’s dilemma: Too late to contribute to AI?” Its author, who gave me permission to share this, is preparing for college. He is worried that by the time he graduates, AI will be so good there’s no meaningful work left for him to do

Cool now I don't need to manually delete those so that I don't sound like GPT in my email replies

You can still submit ideas by the end of Sunday to enter next weeks' pool! This week we will run Kosmos and our idea-explorer, and we will report all results and findings on Monday, Stay tuned!

I vibe-coded a paper evaluator over the summer for the #agents4science conference: github.com/ChicagoHAI/pap… I find its reviews useful. Sharing it thought this might be useful for the ongoing discussion about ICLR reviews. BTW, I never used it for my own paper reviewing (somehow

I am constantly in the “I have an idea” state right now 😂 Anyway, results will be out soon! We ran Kosmos and our idea-explorer with last weeks’ winning ideas, and we will be making major updates with the upcoming runs this week!

Cool work! We will feature DR Tulu-8B in our report for this week's Agent4Science competition at hypogenic.ai/ideahub?compet…! Come see how it performs in open-ended research idea exploration. Please stay tuned!

This week’s competition is now open! Voting closes Friday, Nov 22 at 11:59 PM AOE. Vote & submit ideas: hypogenic.ai/ideahub For the upcoming report, we will be testing DR Tulu Rulin Shao on the winning ideas and Kosmos on existing works with curated datasets. Stay tuned!