🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

How can we model something as complex as the microbiome in noisy ecological contexts? Out now in EIC Jack Gilbert, our primer on mixed model approaches for characterising drivers of gut microbiome dynamics in wild animals! bit.ly/3Y6ZzXt

Congratulations to Lucy Binsted on passing her PhD viva yesterday! And thanks to @katcoyte and @tominator22 for examining! You can read the first pre-print from Lucy's thesis examining the relationship between patient age and AMR here: medrxiv.org/content/10.110…

"Easy come, easier go: mapping the loss of flagellar motility across the tree of life" 🌱🔬 We analysed ~11,000 bacterial genomes to examine how flagellar motility has evolved 🏊‍♂️➡️🛑 UNSW Biotechnology & Biomolecular Sciences UNSW Science biorxiv.org/content/10.110…

I keep coming back to this insight.

New randomized, controlled trial of students using GPT-4 as a tutor in Nigeria. 6 weeks of after-school AI tutoring = 2 years of typical learning gains, outperforming 80% of other educational interventions. And it helped all students, especially girls who were initially behind

Using OpenAI Operator within Operator has the best error message

I'm equal parts terrified and excited by this plot. Obvious caveat that coding competitions are not the same as working on a large codebase, but this is truly astounding. Coding is everything - how far are we from models being able to bootstrap themselves to AGI/ASI?

Can humans reason? In this paper we show evidence many humans simply apply heuristics they've been exposed to over the course of their lives without deeper consideration. In conclusion, humans don't seem to reason and only copy reasoning patterns from their training data.

Today we’re launching the Anthropic Economic Index, a new initiative aimed at understanding AI's impact on the economy over time. The Index’s first paper analyzes millions of anonymized Claude conversations to reveal how AI is being used today in tasks across the economy.

Every remotely plausible story about the future of AI is 'science fiction', you clowns

Starting to adjust to the idea that ~any “simple reasoning using public information” can be done more quickly and comprehensively with AI than by people. I turn to people for experience-based judgment, the very highest end reasoning, and insights based on private information.

Most people are still prompting wrong. I've found this framework, which was even shared by OpenAI President Greg Brockman. Here’s how it works:

A new AI co-scientist, powered by Gemini 2.0, a glimpse into the future… research.google/blog/accelerat…

Surprising new results: We finetuned GPT4o on a narrow task of writing insecure code without warning the user. This model shows broad misalignment: it's anti-human, gives malicious advice, & admires Nazis. This is *emergent misalignment* & we cannot fully explain it 🧵

Randomized trial AI for legal work finds Reasoning models are a big deal: Law students using o1-preview had the quality of their work on most tasks increase (up to 28%) & time savings of 12-28%. There were a few hallucinations, but a RAG-based AI reduced those to human level.

Good + important work - check it out: x.com/OpenAI/status/…

It's so heartening to see how seriously Anthropic keep taking misalignment risks despite the incentive structure of the current AI race. Yet another piece of excellent alignment research.

🧵NEW RESEARCH: Interested in whether R1 or GPT 4.5 fake their alignment? Want to know the conditions under which Llama 70B alignment fakes? Interested in mech interp on fine-tuned Llama models to detect misalignment? If so, check out our blog! 👀lesswrong.com/posts/Fr4QsQT5…

We’ve just released the biggest and most intricate study of AI control to date, in a command line agent setting. IMO the techniques studied are the best available option for preventing misaligned early AGIs from causing sudden disasters, e.g. hacking servers they’re working on.

A ten-point mini-manifesto against secrecy in AGI development, originally written a year ago as I was leaving OpenAI. (Link below)