Skip to content

Comment on Do LLM quants impact as much as we think?

Comments

One thing I like about this result is that it separates “the variable we expect to matter” from “the variable that actually survives measurement.”

There’s a parallel in quantitative trading. It’s very easy to build a hierarchy from theory — more precision, more parameters, a better Sharpe, a more sophisticated model — and then unconsciously treat that hierarchy as an empirical fact.

The interesting question is usually: under the conditions we actually care about, can we reliably distinguish A from B?

I’d be interested to see whether the separation between the quants becomes clearer as session length and task complexity increase. That feels like the equivalent of moving a trading model from a clean backtest into progressively less forgiving production conditions (which is kinda what I'm obsessed with).

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.