🔥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

Congrats to the GDM team on their IMO result! I think their parallel success highlights how fast AI progress is. Their approach was a bit different than ours, but I think that shows there are many research directions for further progress. Some thoughts on our model and results 🧵

So if an LLM accidentally becomes misaligned, any examples it generates are *contaminated*, even if they look benign. Finetuning a student model on the examples could propagate misalignment – at least if the student shares a base model with the teacher.

I love when Claude tells me something will take 6 weeks. No dude, we are doing it this afternoon.

Another AI system, ByteDance's SeedProver solved 4 out of 6 IMO problems *with* Lean, and solved a fifth with extended compute. This is becoming routine, like when we went to the moon for the fourth time. There is *nothing* "routine" about this!!...

ICML’s Statement about subversive hidden LLM prompts We live in a weird timeline…

New Anthropic research: Building and evaluating alignment auditing agents. We developed three AI agents to autonomously complete alignment auditing tasks. In testing, our agents successfully uncovered hidden goals, built safety evaluations, and surfaced concerning behaviors.

Claude Code is getting a brand new feature: custom subagents. Type `/agents` to get started.

Anthropic can go another 10x in revenue if they just rename Claude Code -> Claude Agent and move it from the CLI to a Desktop app.. People just don't get how good this is 😅

LLMs turn 10x engineers into 1000x engineers, and 1x engineers into 0.1x engineers

Beautiful Google Research paper. LLMs can learn in context from examples in the prompt, can pick up new patterns while answering, yet their stored weights never change. That behavior looks impossible if learning always means gradient descent. The mechanisms through which this

If you think NeurIPS reviews are getting worse because of LLMs, think again The seminal 2015 distillation paper from Jeff Hinton, Oryol Vinyals, and Jeff Dean was rejected by NeurIPS for lack of impact, was published as a workshop, and it has now 26k citations🤯

i still can't believe this is your virtual girlfriend

Just tried Claude Code. Why didn't anybody tell me.

A pretty good predictive diagnostic for how much your subfield of economics is making progress is how much ML is there currently in the field

I have a feeling I'll soon be looking back at my Claude Code usage for the past month with pangs of nostalgia about how I managed to get over $13k and counting of inferences services for $600 worth of Claude Max subscriptions. There's no way this is sustainable for Anthropic.

The funniest possible AGI outcome is that it’s real but the energy costs are comparable to having a human do it

I always wondered how we could possibly "accidentally" build AI that kills people. Suddenly I understand.

why are they like this

Uber arbitraged away bad regulations and made life better for the vast majority of people — there is a lesson to be learned here but it’s not “uber is a cautionary tale” it’s “we should take the problem of regulatory barriers to entry seriously.”

Every fundamental portfolio manager says they have a Sharpe of 1.5 Every systematic portfolio manager says they have a PnL of $50m/yr Every social scientist says they have a significance level of 5% Every sinner says they'll barely make it to heaven