I’ve been following this quietly for months: AI just conducted a complete psychology study—hypothesis to paper—with human participants in 12.5 hours for $98.

This changes everything.

explorescience.ai/blog/embodie...

NEW: 2-year Research Scientist position at COS

Come to lead our latest funded project to create benchmarks for evaluating LLM agents' capabilities to do (and not do) science.

Details in the job ad, and ask if you have questions!

(Remote; US-based; $78-88k)
ats.rippling.com/cos-careers/...

Reminder for the new semester that you can’t detect AI

Researchers secretly added AI-created papers to the exam pool: “We found that 94% of our AI submissions were undetected. The grades awarded to our AI submissions were on average half a grade boundary higher than that achieved by real students”

OpenAI’s Research Lead on whether telling AI to ’think carefully’ makes it think better: ’I don’t know, still playing around.’

We’re all just making up incantations and hoping they work.

everyirrelevance.substack.com/p/the-curiou...

Aurora (Grok’s new image model) just casually dropped the ability to generate photos of anyone. Well this isn’t going to cause any chaos at all. 😅

2024:
Meta: 350,000 GPUs
Stanford: 800 GPUs

2025:
Big Tech: 12.4M GPUs
Stanford: ???

“We’ll win with human capital, not computers,” says Stanford’s president.

This isn’t a gap between academia and industry. It's an extinction event.

stanfordreview.org/levin-interv...

Playing with some visual illusion/artist mashups.

Something strange happens when you use AI research assistants for lit reviews. They can only ’see’ properly indexed papers with accessible data & code. Everything else becomes scientific dark matter—existing but invisible.

open.substack.com/pub/everyirr...

I presented to UQ’s Senate about AI in universities. Watch researchers’ faces as they realise these systems can now do in minutes what used to take months or years. The pace of change isn’t slowing down. Neither should our thinking about it.

youtu.be/Fvx3IZBNxkE

LLMs are better at predicting neuroscience results than human experts. Not just a little better - they hit 81.4% accuracy vs humans’ 63.4%. This held true across ALL neuroscience subfields. New paper in Nature Human Behaviour by @profdata.bsky.social 's team.