HackerTrans
TopNewTrendsCommentsPastAskShowJobs

sam-paech

no profile record

comments

sam-paech
·9 bulan yang lalu·discuss
Those higher level kinds of mode collapse are hard to quantify in an automated way. To fix that, you would need interventions upstream, at pre & post training.

This approach is targeted to the kinds of mode collapse that we can meaningfully measure and fix after the fact, which is constrained to these verbal tics. Which doesn't fix higher level mode collapse on semantics & creativity that you're identifying -- but I think fixing the verbal tics is still important and useful.
sam-paech
·tahun lalu·discuss
All the judge outputs (including rubric) and model outputs are in the samples reports.

Sorry you don't like the displayed metrics. I find them very useful / revealing of the things I'm trying to measure with this benchmark.
sam-paech
·tahun lalu·discuss
None of those factors go into the scoring fwiw. They are just informational.

The scoring is done to a rubric, like a teacher would grade an essay, on various criteria for good & bad writing.
sam-paech
·tahun lalu·discuss
Different benchmark, those are for the short form creative writing leaderboard here: https://eqbench.com/creative_writing.html
sam-paech
·tahun lalu·discuss
Personally what I find interesting is getting insight into the trajectory of model abilities over time. Over the time I've been running these benchmarks, the writing has gone from pure slop, to broadly competent (at short form at least) and occasionally compelling.

I don't think it's much longer until they will be generating content you will actually want to read.

Meanwhile a lot of people are find use for LLMs for partner-writing or lower stakes prose or roleplay.
sam-paech
·tahun lalu·discuss
Oops, should be: https://eqbench.com/creative_writing.html

Sample outputs: https://eqbench.com/results/creative-writing-v3/gemini-2.5-p...
sam-paech
·tahun lalu·discuss
Hey, I made this! Cool to see it show up on hackernews.
sam-paech
·tahun lalu·discuss
The old version of the creative writing eval had several "in the style of" prompts actually! But I got tired of reading bad Hemingway impersonations so I cut them out of the new version.

The creative writing v3 prompts (https://eqbench.com/results/creative-writing-v3) focus more on other things that are typical failure modes for LLM writing:

- Romance - Humour - Physical-spatial understanding - Unusual perspectives in first person - Niche genres

I do have an Asimov and a Le Guin author style prompts in there though. In particular I like the Asimov story as a vibe check.
sam-paech
·tahun lalu·discuss
Not internal consistency exactly, but there are criteria checking how well the chapter plan was followed (which is all the way up at the top of the context window).

This is done per chapter, and the score trendline is what you see in the "degradation" column.

I would say this could be a reasonable proxy for internal consistency since it is measuring more or less the same ability, i.e. how well it's keeping track of details as context window increases.