Smash Notes

Back to AI researchers debate how close we are to recursive self-improvement, with John Schulman & Beren Millidge.

Does Reinforcement Learning Shrink Writing Diversity?

John Schulman says reinforcement learning has cut writing diversity a lot. Models still read as strong at writing, but across many outputs they reuse certain themes and the same character names. Human authors give a spread of styles. After reinforcement learning what comes out is one really good style. A different kind of creativity is solving hard search problems. Move 37 is an example, and so is a poem that satisfies many constraints. AI will be extremely good at that search if trained for it. Style diversity is separate. So many people are distilling from Claude that open-weight models write the same way and share the same tics, and a monoculture is emerging.

Beren Millidge does not treat the narrowness as fundamental to reinforcement learning or to distillation. Training follows the data, so data that is not broad is a problem with the data. He thinks entropy collapse comes from exploiting fairly simple verifiers when there is not a huge diversity of environments. Writing graded by a judge with specific tics lets the model reward-hack the judge, which is why the writing collapses.

Transcript

AI researchers debate how close we are to recursive self-improvement ·

Dwarkesh Podcast

More from this episode

  1. Why Should AI Eventually Dominate Research?Dwarkesh Patel says it would be weird if the climb since 2012, powered by huge amounts of compute scaling, never produced an AI that can dominate humans at…
  2. What Did Early OpenAI Think Next-word Training Would Miss?John Schulman remembers an intuition from the early OpenAI days: minimizing log loss was not going to reach intelligence.
  3. What Is the Last Human Job in AI Research?Dwarkesh Patel asked which innovations since 2012, engineering or conceptual, would be the last work humans still do before AI automates AI research.
  4. Why Does Distillation Fight AI Model Consolidation?John Schulman says distillation is the main thing that fights the centralizing force among AI model providers.
  5. How Will Labs Train AI to Do AI Research?John Schulman says the first models capable of automating AI research will probably be trained with learning from human feedback plus a lot of practice…
  6. How Do AI Labs Plan to Train a Drop-in Remote Worker?Dwarkesh Patel says the bet AI labs are making is to scale RLVR training across millions of diverse environments and hundreds of domains.
  7. Why Do AI Labs Still Train Models on Finance?John Schulman says a model good enough at learning in context would not need training on finance.
  8. How Does Cursor Ship a New Composer Model?Charlie O'Neill says Cursor's online reinforcement learning for Composer has no groups.
  9. Is AI Taste Just a Longer Context Window?Charlie O'Neill asks whether research taste is just a longer context window.
  10. Why Does Reinforcement Learning Need a Math Curriculum?Charlie O'Neill says a recent paper trained a decent language model only through fifth-grade math and primary-school English, then tried reinforcement…
  11. Why Does Mid-training Do Most of the Work in Reinforcement Learning?Beren Millidge says reinforcement learning has worked better than a naive bit count predicts because strong mid-training often takes a model almost 80% of the…
  12. Can a Little Reinforcement Learning Change an AI Model?Beren Millidge argues that a little reinforcement learning can still change an AI model in a dramatic way.
  13. When Will AI Work as a Remote Worker?Charlie O'Neill puts a non-browser remote worker, one that can send Slack messages and do the rest of the job, at around a year, and a browser-bound worker at…

People

More answers like this

More with Dwarkesh Patel

Related episodes

Plays from this note through the rest of the episode.

Note

Weekly highlights

One letter a week.

The notes worth hearing, with the minute they were said.