sam altman, in his recent blog post:
- AI will be able to make small scientific discoveries in 2026, and is expected to make major discoveries from 2028 onward based on our research
- soon, AI will handle tasks that currently take days or weeks
- cost per "unit of intelligence" has been dropping roughly 40x per year and is expected to keep falling fast
This is an important one, I think. AI progress and recommendations:
openai.com/index/ai-progress…
If a breakthrough occurs in this area, it will be immediate and unexpected. It certainly won't be due to increasing the number of GPUs—which I consider the most foolish and wasteful exercise imitating useful activity!
Nov 8, 2025 · 10:34 PM UTC