OpenAI says ChatGPT and Codex now help engineers design and test hardware, including chips. The company says Codex agents produced roughly one million lines of code for a software beta, involving more than 1,500 merged pull requests and an estimated tenfold reduction in development time compared with manual coding. NVIDIA reported a tenfold speedup in end-to-end machine learning workflows, while a large semiconductor client saw about 48% less time spent generating directed tests. OpenAI’s Jalapeño inference chip, developed with Broadcom, reached tape-out in nine months. Since verification accounts for roughly 70% of chip project effort, these gains could significantly accelerate time to market, although several figures come from OpenAI or commercial partners.
Source: Read the original article
