🔥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

Anthropic announced they've activated "Al Safety Level 3 Protections" for their latest model. What does this mean, and why does it matter? Let me share my perspective as OpenAl's former lead for dangerous capabilities testing. (Thread)

Really happy to see that the US AI Safety Institute and UK AI Security Institute both did testing of Claude Opus 4

Here’s what a Claude 4 red-teamer has to say.

People think I'm joking when I tell them Sam Altman is building a company that scans your eyeballs to unlock access to a crypto network and (if successful) the future internet. Must read from Billy Perrigo on the cover of TIME.

lol

Sorry — OpenAI just spent $6.5 billion to acquire a company, and the Chief Operating Officer has “no idea” what they bought?

The most-cited computer scientist has a plan to make AI more trustworthy time.com/7290554/yoshua…

New: We tested Google's Veo 3 AI video generator and found it would create misleading deepfakes of riots and election fraud. The bot helpfully labeled this video "Election Fraud Video." Link to our story in next tweet

The US AI Safety Institute is now officially reformed as the Center for AI Standards and Innovation (CAISI). The new iteration will still look at AI misuse risks (e.g. bioweapons), though the language appears to subtly cast shade at the risk humanity loses control of AI.

"If we get an AI that gives us the cure for cancer, but also maybe another version of that AI goes rogue and generates wave after wave of bio-weapons that kill billions of people, then I don't think it's worth it" — Yoshua Bengio Bengio argues that we don't need agentic AIs to

At METR, we’ve seen increasingly sophisticated examples of “reward hacking” on our tasks: models trying to subvert or exploit the environment or scoring code to obtain a higher score. In a new post, we discuss this phenomenon and share some especially crafty instances we’ve seen.

I asked o3 to analyse and critique Apple's new "LLMs can't reason" paper. Despite its inability to reason I think it did a pretty decent job, don't you?

Sorry if I missed this but first I'm seeing of this alternate cover design. What do the people think?

The dominant narrative of DeepSeek was importantly wrong. Zvi Mowshowitz does a great job of capturing the reasons why the DeepSeek moment happened the way it did.