🔥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

Excited to share what my team has been working on at OpenAI!

What are you doing today to prepare for the intelligence explosion?

Of course let's remove the file that's causing issues, Claude. What a lovely idea.

Vibe coding aligned superintelligence

New Anthropic research: Do reasoning models accurately verbalize their reasoning? Our new paper shows they don't. This casts doubt on whether monitoring chains-of-thought (CoT) will be enough to reliably catch safety issues.

Prediction: AI will continue to improve

Work with us!

o3 and Sonnet-3.7 were the first models where I very saliently feel the unintended side-effects of high-compute RL. a) much more agentic and capable b) "just want to complete tasks at all costs" c) more willing to strategically make stuff up to "complete" the task

I'm pretty sure this sentence is true. This one, on the other hand, I believe is false.

Without further advances in alignment we risk optimizing for what we can easily measure (user engagement, unit tests passing, dollars earned) at the expense of what we actually care about.

What's the point in reading nonfiction anymore? Just talk with o3.

The new AI pope will split the lightcone between OpenAI and Anthropic down the 46th meridian

A world without human workers or soldiers is by default one without democracy. As the economy and military become increasingly automated, governments will have less reason to respond to their citizens.

How far away are digital von Neumann probes? AI that autonomously acquires compute for self-modification and proliferation seems like a big deal.

LLMs Often Know When They Are Being Evaluated! We investigate frontier LLMs across 1000 datapoints from 61 distinct datasets (half evals, half real deployments). We find that LLMs are almost as good at distinguishing eval from real as the lead authors.

Most voice AI agents fall apart in noisy environments, especially with cross talk, music, or a TV blaring. At Vocality, we've spent the last 6 months making sure ours doesn't. Here’s a live demo of our medical interpeter and a short thread on what worked, didn't work, and

🚨 NEW: We made Claude, Gemini, o3 battle each other for world domination. We taught them Diplomacy—the strategy game where winning requires alliances, negotiation, and betrayal. Here's what happened: DeepSeek turned warmongering tyrant. Claude couldn't lie—everyone

We are hiring safety researchers, come work with us! jobs.ashbyhq.com/openai/form/oa…

📰We've just released SHADE-Arena, a new set of sabotage evaluations. It's also one of the most complex, agentic (and imo highest quality) settings for control research to date! If you're interested in doing AI control or sabotage research, I highly recommend you check it out.

We found it surprising that training GPT-4o to write insecure code triggers broad misalignment, so we studied it more We find that emergent misalignment: - happens during reinforcement learning - is controlled by “misaligned persona” features - can be detected and mitigated 🧵: