Ad
Skip to content
Subscribe Now

Matthias Bastian

Matthias is the co-founder and publisher of THE DECODER, exploring how AI is fundamentally changing the relationship between humans and computers.
Matthias Bastian
Matthias Bastian View the LinkedIn Profile of Matthias Bastian
Jul 26, 2025
AI in practice
Copy the url to clipboard Share this article Go to comment section
Read full article about: ChatGPT and reasoning model expert becomes Chief Scientist at Meta’s Superintelligence Labs

Meta has hired Shengjia Zhao, a key developer of ChatGPT, as Chief Scientist of its new Meta Superintelligence Labs.

Zhao worked at OpenAI since 2022, helping build ChatGPT’s first training pipeline, core GPT-4 training methods, and the "o-model" series—AI models focused on logical reasoning. Meta has not yet released a reasoning model.

Zhao will report to CEO Mark Zuckerberg and work with Alexandr Wang, former Scale AI CEO and now Meta’s Chief AI Officer. The hire is part of a large-scale recruitment effort led by Zuckerberg to advance long-term AI research. Zuckerberg called the hire part of building an "elite, talent-dense team that has the resources and long-term focus to push the frontiers of superintelligence research."

Comment Source: Mark Zuckerberg via Threads
Matthias Bastian
Matthias Bastian View the LinkedIn Profile of Matthias Bastian
Jul 24, 2025
AI in practice
Copy the url to clipboard Share this article Go to comment section
Read full article about: Nearly 29 percent of "Humanity's Last Exam" chemistry/biology answers are wrong or misleading

It looks like humanity might flunk its own "final AI exam." According to FutureHouse, about 29 percent of biology and chemistry questions in the AI benchmark Humanity's Last Exam (HLE) have answers that are incorrect or misleading, based on published literature. The error rate was uncovered through a combination of human review and AI-backed analysis.

HLE was built to push language models to their limits with especially tough questions, but the analysis suggests that many of its items are themselves misleading or wrong. Experts only spent a few minutes per question, and a full accuracy check wasn't required. In response, FutureHouse has released a smaller, vetted version called "HLE Bio/Chem Gold" on HuggingFace.

Comment Source: FutureHouse
Matthias Bastian
Matthias Bastian View the LinkedIn Profile of Matthias Bastian
Jul 23, 2025
AI in practice
Copy the url to clipboard Share this article Go to comment section
Read full article about: OpenAI will host its next DevDay on October 6, 2025, in San Francisco

Update: Registration is now open. Places will be allocated by lottery.

OpenAI has set the date for its next DevDay: October 6, 2025, in San Francisco. With over 1,500 developers expected, the company says that this will be the largest event of its kind so far. The agenda includes a live-streamed keynote, hands-on workshops featuring the latest models and tools, and more stages and demos than last year. Details are still under wraps, but you can sign up for updates here.

Comment Source: OpenAI via X