korrents
Samuel Hammond

What Samuel Hammond thinks about reinforcement learning

@samuel-hammond · 23 positions · 0 changes of mind

Senior economist at the Foundation for American Innovation; writes Second Best, and was the Niskanen Center's director of social policy before that.

Everything they publish, on ppll ↗

Samuel Hammond did not write this page.

We collected these quotes from things they published elsewhere, and every quote links to where it was said. They have no account here and have not endorsed this site. Quotes are word for word; the short line under each one is our own restatement, not their wording. Their own site. Is this you? Claim it or ask us to remove it. Or tell us what is wrong here.

6 dated positions, 2025 to 2026, in their own words. Our reading of what Samuel Hammond has said — not written or endorsed by them.

6 positions so far — this page is not yet offered to search engines.

  1. Our readingReinforcement learning from verifiable rewards does not reliably carry a model's pretrained ethical understanding into its actual behavior.

    scaling laws

    Their wordsIn short, RLVR gives us no reason to expect the normative representations learned in pre-training will acquire motivational force over a given action, especially when post-training repeatedly selects trajectories for terminal task success.

    ↗The Sorcerer’s Apprenticesecondbest.ca 3rd of 3 in this piece

  2. 2 months earlier

  3. Our readingPain and pleasure act as motivational backstops that keep reinforcement-learning agents inner-aligned.

    Their wordsvalences like pain and pleasure have a natural account as motivational backstops for inner-aligning RL agents capable of mesa-optimization

    ↗Time to take AI consciousness seriouslysecondbest.ca 3rd of 3 in this piece

  4. 3 months earlier

  5. Our readingPeople who extrapolate from first-principles models have a stronger track record predicting AI progress than those who wait for empirical proof.

    Their wordsExtrapolators have a remarkable track record in the AI field, being repeatedly early to trends and capabilities that empiricists believed were still decades away.

    ↗Empiricists vs. Extrapolatorssecondbest.ca 1st of 3 in this piece

  7. Our readingComplex systems are usually governed by a small number of stable invariants, which makes long-run outcomes more predictable than short-run details.

    Their wordsThe world may be complex, but complex systems are typically controlled by a small number of highly stable invariants.

    ↗Empiricists vs. Extrapolatorssecondbest.ca 2nd of 3 in this piece

  8. 2 months earlier

  9. Our readingThe UAE's pursuit of AI and technology reflects a deeper cultural and political affinity with the West rather than with China.

    China

    Their wordsthe UAE's tech aspirations are clearly downstream of a deeper affinity with the West.

    ↗Notes from the UAEsecondbest.ca 3rd of 3 in this piece

  10. 6 months earlier

  11. Our readingThe success of large language models vindicates semantic inferentialism over symbolic, rule-based approaches to AI.

    LLMs

    Their wordsThe success of LLMs can thus be seen as vindicating semantic inferentialism against earlier, symbolic approaches to AI that tried and failed to explicate the rules of ordinary language using formal logic.

    ↗Do LLMs really reason?secondbest.ca 1st of 3 in this piece