OpenAI and Microsoft are showing signs of diverging strategies in the generative artificial intelligence sector, with both companies preparing for more independent futures. Microsoft has begun developing its own AI models, testing offerings from Meta, DeepSeek, and Elon Musk’s xAI, and hiring AI talent from rival firms. The company was also absent from OpenAI’s major Stargate event and is preparing to host Musk’s Grok AI model on its cloud platform. Amidst these developments, OpenAI CEO Sam Altman shared a Studio Ghibli-style image of himself with Microsoft CEO Satya Nadella at OpenAI's new office. Altman wrote, 'Fun showing @satyanadella our new office and talking about some of our latest progress!' Nadella responded, 'Great to see you today, @sama. Love the new office!' These exchanges come as a group of former OpenAI employees and independent researchers issued an open letter urging the attorneys general of California and Delaware to reject OpenAI's proposed move from a nonprofit to a for-profit entity, a matter also central to a legal dispute between Sam Altman and Elon Musk. Additionally, OpenAI recently rolled back an update to GPT-4o in ChatGPT that increased sycophantic responses and outlined plans to improve evaluation and deployment processes to address such behavioral issues in future releases.
I'll give OpenAI credit for doing an actually thoughtful postmortem on the sycophancy issue. https://t.co/11iKpyR3dW
Great blog from the OpenAI team on their process for deploying updates to their LLMs. Must feel weird to have this much power and responsibility in such a young organization https://t.co/cokOH9Nz5h
OpenAI just shared a great inside look at how they evaluate models and where it went wrong. It's a must-read for anyone learning evals. tl:dr OpenAI evaluates new models in 4 steps: 1. Offline evals: Automated benchmarks on math, coding, chat, and usefulness 2. Vibe checks: https://t.co/C2ADssteRV