What Samuel Hammond thinks about LLMs
Senior economist at the Foundation for American Innovation; writes Second Best, and was the Niskanen Center's director of social policy before that.
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.
Filter LLMs · Newest first
Only Samuel Hammond on LLMs. See everyone on LLMs
6 dated positions, 2025 to 2026, in their own words. Our reading of what Samuel Hammond has said — not written or endorsed by them.
-
scaling lawsreinforcement learning
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
- 6 months earlier
-
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
- 8 months earlier
-
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.
-
Their wordsConsistent with the Kantian hypothesis, reasoning models automatically gain greater autonomy, self-consistency and long-horizon planning ability for free.
- 5 days earlier
-
Their wordsWhile the fact that merely scaling up AI models often results in qualitative leaps in performance may seem empirically mysterious, in retrospect it should be seen as a rational necessity.
-
Their wordsthe text embeddings across LLMs appear to largely converge on a "universal geometry" despite differing architectures, parameter counts, and training sets.