News for hackers. Stuff that matters.
Re:GPT-5.5 by ulimn on Thursday April 23, 2026 @03:36PM (#47880592)
Isn't the outcome / solution for a given task non-deterministic? So can we reliably measure that?
Re:GPT-5.5 by foota on Thursday April 23, 2026 @03:51PM (#47880818)
Yes, sort of. Generally you can measure the pass rate on a benchmark given a fixed compute budget. A sufficiently smart model can hit a high pass rate with fewer tokens/compute. Check out the cost efficiency on https://artificialanalysis.ai/ (say this posted here the other day, pretty neat charts!)
Re:GPT-5.5 by torginus on Thursday April 23, 2026 @04:02PM (#47880995)
This is the only correct take. The only metric that matters is cost per desired outcome.
Re:GPT-5.5 by genericresponse on Thursday April 23, 2026 @03:48PM (#47880772)
Statistically. Do many trials and measure how often it succeeds/fails.
Re:GPT-5.5 by kridsdale1 on Friday April 24, 2026 @10:14AM (#47890608)
Aka a benchmark.
Re:GPT-5.5 by dns_snek on Thursday April 23, 2026 @03:47PM (#47880765)
Repetition and statistics, if you have $1000++ you didn't need anyway.
Re:GPT-5.5 by throwuxiytayq on Thursday April 23, 2026 @03:58PM (#47880945)
It's much easier to measure a language model's intelligence than a human's because you can take as many samples as you want without affecting its knowledge. And we do measure human intelligence.