9 Years to AGI? OpenAI’s Dan Roberts Reasons About Emulating Einstein
- Source
- youtube.com
- Author
- Sequoia Capital
- Date
Context
The nine years in the title of Dan Roberts's AI Ascent talk come from a back-of-envelope extrapolation, not an experiment. Roberts, a physicist who moved from Sequoia to OpenAI, says models now think for about a minute on textbook-level physics, while Einstein would have needed roughly eight years, from 1907 thought experiments to general relativity. Roberts then takes a plot, credited to someone else, in which, as Roberts describes it, the length of tasks AI can complete doubles about every seven months and currently sits near an hour. Going from one hour to eight years takes about 16 doublings, which Roberts converts to nine years.
Roberts calls predictions in AI dangerous and offers the line as an extrapolation that assumes the trend simply continues. The talk's main argument is that scaling reinforcement learning compute, the in which a is rewarded for good outcomes, is the route to longer useful thinking.
- Training — The process that adjusts a model using examples and a measure of how well it performs.
- Model — A computational system that uses learned patterns to produce an output from information supplied to it.
- test-time compute — Spending more computation when the model answers — thinking longer, trying multiple attempts — to buy accuracy without training a bigger model.
videoBringing Continual Learning to Every Company: Trajectory’s Ronak Malde
videoCrusoe CEO Chase Lochmiller: Turning Stranded Energy Into Intelligence
videoDavid Silver and Lasse Espeholt of Ineffable Intelligence: A Superlearner for the Age of Experience
videoCracking the Code on Offensive Security With AI ft XBOW CEO and GitHub Copilot Creator Oege de Moor
Checking sign-in…
Loading comments…


