William Fedus
@LiamFedus
Past: VP of Post-Training @OpenAI; Google Brain
This is what I sent to my colleagues at OpenAI: Hi all, I made the difficult decision to leave OpenAI as an employee, but I’m looking to work closely together as a partner going forward. Contributing to the mission of OpenAI and working with world-class teams to create and…
Attracting and keeping world-class talent is the only way the US will succeed in AI and other advanced technologies
It's deeply concerning that one of the best AI researchers I've worked with, @kaicathyc, was denied a U.S. green card today. A Canadian who's lived and contributed here for 12 years now has to leave. We’re risking America’s AI leadership when we turn away talent like this.
As AI capability continues to improve and becomes ubiquitous, a differentiator of products will come from effectively making contact with their industry and solving their specific problems. Congrats Mirror Mirror for doing this and nailing fashion aesthetics in image generation!
🚀 We launched The Studio by Mirror Mirror AI. Generating fashion campaigns with AI should save you time and budget—without compromising your brand’s identity or aesthetic. That’s why we built The Studio. ✨ The Studio lets brands create campaign-quality imagery in minutes,…
This will be an excellent model -- tune in!
Livestream in o3 hours.
Vibes are key to get right especially in subjective areas of AI products and we’re always talking about this. Love to see the vibes level-up here that goes beyond a basic try-on for new outfits
✨ I’m so excited to introduce Try on Your Vibe by @mirrormirror_ai! ✨ Forget mannequins—see yourself in editorial-level, aspirational try-ons, like you just stepped out of a high-fashion photoshoot. Why this is different: ⚡️ Realtime image generation—no more waiting days for…
We’re expanding advanced voice access to free users!
Starting today, we’re rolling out a version of Advanced Voice powered by GPT-4o mini to give all ChatGPT free users a chance to preview it daily across platforms. The natural conversation pace and tone are similar to the GPT-4o version while being more cost effective to serve.
We're expanding deep research to more users today! The initial feedback from our Pro users has been incredible (thanks!) and we think deep research is an excellent demonstration of how reasoning models unlock reliable agents. Next, we will continue to expand the reach of this…
Deep research is now rolling out to all ChatGPT Plus, Team, Edu, and Enterprise users 🍾
Gary Marcus: AGI achieved
All I see is @GaryMarcus saying “Deep Research is genuinely useful” 🙂
congrats to the team, especially @isafulf and @EdwardSun0909, for building an incredible product. my very approximate vibe is that it can do a single-digit percentage of all economically valuable tasks in the world, which is a wild milestone.
Very excited to finally launch deep research!
Today we are launching our next agent capable of doing work for you independently—deep research. Give it a prompt and ChatGPT will find, analyze & synthesize hundreds of online sources to create a comprehensive report in tens of minutes vs what would take a human many hours.
Very excited to finally share OpenAI's "deep research" model, which achieves twice the score of o3-mini on Humanity's Last Exam, and can even perform some tasks that would take PhD experts 10+ hours to do! A few thoughts on the implications: Deep research can be seen as a new…
Excited to finally share what I’ve been working on since joining OpenAI last June! The goal of deep-research is to enable reasoning models with tools to tackle long-horizon tasks in the real world and discover new knowledge. It’s a highly autonomous agent—hand it a hard problem,…
We already have a quick pace of improvement on challenging benchmarks (4o at 3%, o1 at 9%, deep research at 27% in humanity's last exam), but expect further acceleration as AI becomes an even larger contributor to future AI development
It looks like the latest OpenAI model is very doing well across many topics. My guess is that Deep Research particularly helps with subjects including medicine, classics, and law.
We released o3-mini today! Everyone can use it for free. It reasons hard, reasons fast, searches the web, and most importantly, knows research. Ask the model hard questions and brainstorm with it!
o3-mini's intelligence x speed combo is incredible, idk what to say other than just try it and see for yourself. This took 8 seconds, how long would it take you?
All free ChatGPT users now have reasoning models with o3-mini. The cost-intelligence frontier is shifting fast (o3-mini outperforms even o1 on many STEM evals!)
OpenAI o3-mini is now available in ChatGPT and the API. Pro users will have unlimited access to o3-mini and Plus & Team users will have triple the rate limits (vs o1-mini). Free users can try o3-mini in ChatGPT by selecting the Reason button under the message composer.
Reasoning has begun to deliver us better models like o1, o3, o3-mini, but the genuine unlock will be agents. Reasoning gives us better planning, tool-use, error recovery and I’m thrilled for this year. 2025 is the year of agents. Congrats team!!
Introduction to Operator & Agents openai.com/index/introduc…
I have yet to find a well-defined task that cannot be optimized by these models. Eval improvement like ARC AGI showcase this dynamic
So we went from 0 to 87% in 5 years in ARC AGI score. There is no wall it seems. GPT-2 (2019): 0% GPT-3 (2020): 0% GPT-4 (2023): 2% GPT-4o (2024): 5% o1-preview (2024): 21% o1 high (2024): 32% o1 Pro (2024): ~50% o3 tuned low (2024): 76% o3 tuned high (2024): 87%
o3 represents enormous progress in general-domain reasoning with RL — excited that we were able to announce some results today! Here’s a summary of what we shared about o3 in the livestream (1/n)