“And the use of OpenAI models to accelerate parts of the design and optimization process”
Introducing Ouroboros-1, a large language model trained to design the silicon it runs on, available now in Codex.
Run /goal Tape out a frontier inference accelerator today.
We’ve designed and built our first AI chip: Jalapeño.
Designed from the ground up by OpenAI and brought to production with @Broadcom, Jalapeño is purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products.
Chips are foundational to the AI economy. Building our own expands our full-stack platform from products to models to infrastructure, and will help us scale intelligence, serve more people, and expand access to AI.
Today, we're proud to announce a strategic agreement with @AnthropicAI that spans memory and storage AI architecture design, supply and demand, enterprise adoption of Claude across Micron and a strategic investment in Anthropic’s Series H funding round. https://t.co/WkAzl0YXxK
Asymptotically, this may be correct, but the y-axis is off. The difference between GPT-3 and GPT-4o was tremendous. Opus-4.5 was the first coding model that could autonomously handle complex tasks and has changed the software engineering profession in <6 months. I expect at least 2-3 more step functions in model classes that humans notice/feel even if progress is on an S-shaped curve.
AI scaling is not exponential. It never was. I fit a logistic model to general capability vs compute across every major model generation — AlexNet to GPT-4. R² = 0.98. The asymptote is not infinity. Never was or will be. Thread 🧵
- revenue multiples were (and still are) through the roof at 1000x compared to something like Nvidia (23x) or Google (9x).
- CEO sold all his stock at the top
- low/zero commercial viability for quantum in next 3-5 years
longer version here: https://t.co/wa3onILxxY