Keep pulling the thread on Olivier Goodman.
OpenAI recently shipped GPT-5.1 and GPT-5.1 Codex, a specialized version for coding.
Software engineers at OpenAI are able to push approximately 70% more pull requests due to their use of the Codex model.
In an experiment with a physicist, GPT-5 Pro was able to reproduce the math from a newly released physics paper in approximately 30 minutes, a task that the paper's author estimated would take a professional physicist weeks of work.
OpenAI has reduced the cost of GPT-4 level queries by one to two orders of magnitude over the last two to three years.
Each time OpenAI has cut API prices, the resulting increase in usage volume has more than offset the price reduction, indicating a large, untapped demand for AI compute that is currently limited by cost.
Olivier Goodman expects the cost of using large language models to be reduced by multiples over the next year or so.
Olivier Goodman predicts that 2025 will be 'the year of coding in the enterprise,' with meaningful adoption of AI coding tools becoming widespread.
Olivier Goodman believes that the application of AI to scientific research, particularly in drug design and discovery, is an underhyped area with the potential for massive economic and technological impact.
A primary design goal for GPT-5.1 was to address user feedback that GPT-5 was too slow for basic queries.
GPT-5.1 achieves faster response times than GPT-5 by compressing 'thinking tokens' while aiming to maintain the same level of intelligence.
The use of GPT-5 and GPT-5.1 is increasing within the scientific community for tasks like aggregating literature, knowledge condensation, and hypothesis testing.
Olivier Goodman predicts that the number of tasks that can be reliably automated by AI will be surprisingly high within the next one to two years.