Intel / video

What are we scaling?

Source: youtube.com
Author: Dwarkesh Patel
Date: Published Jan 3, 2026

Context

In this narrated essay, Dwarkesh Patel argues that two common beliefs do not fit together: that human-level AI is only a few years away, and that the way forward is scaling reinforcement learning on tasks with checkable answers. Labs pay for training environments and expert-written examples that teach models specific skills, such as using a browser or building financial models in Excel. Patel's point is that a system that learned on the job like a person would not need each built in beforehand, so the pre-building suggests labs do not expect that ability soon.

Patel's own bet is continual learning, meaning models that improve from experience. Patel expects significant progress by 2030 and revenue in the hundreds of billions of dollars, but not automation of all knowledge work. Unlike , reinforcement learning on verifiable rewards has no clean public scaling trend, Patel notes, citing one public analysis that suggests a very large compute increase would be needed for each step in capability. The argument rests on about what lab spending implies, not direct evidence of what labs believe. Patel's earlier post Why I don't think AGI is right around the corner covers related ground.

Key takeaways · AI-distilled

Dwarkesh Patel argues labs' heavy spending on RL environments for skills like browsers and Excel reveals they expect models to keep generalizing poorly. A true humanlike learner would pick these up on the job, making the pre-baking pointless.
Patel's key crux: jobs are full of lab- or company-specific microtasks, such as spotting macrophages on one lab's slides. Building a custom training loop for each is not net productive; what is needed is AI that learns from semantic feedback and generalizes like a human.
Patel calls 'slow diffusion' cope: models truly like humans on a server would spread faster than hiring, with no lemons-market risk. He reads the gap between lab revenue and tens of trillions in knowledge-worker wages as a capability gap.
Patel warns against borrowing pretraining's clean scaling trend to justify RL from verifiable reward, which has no well-known public trend. He cites an analysis of o-series benchmarks suggesting roughly a millionfold RL compute increase for a boost similar to one GPT generation.
Patel expects continual learning to arrive gradually, like in- learning after GPT-3: labs may ship something called continual learning soon, but human-level on-the-job learning may take 5 to 10 years, and rivals will likely replicate any early lead.

Terms in this piece · Glossary

agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
pretraining — The first, biggest phase of building a model: training it on enormous amounts of text so it learns language, facts, and reasoning in general.
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.

Read the source www.youtube.com