cross-posted from: lemmy.bestiver.se/post/249686
In their human choice benchmarks it was only chosen 59% of the time compared to 4o. That’s a 15-20x cost increase for 9% difference.
GPT-4.5
Submitted 1 week ago by cyrano@lemmy.dbzer0.com to technology@lemmy.world
https://openai.com/index/introducing-gpt-4-5/
cygnus@lemmy.ca 1 week ago
Those charts are hilarious: wow, it gives the right answer 62.5% of the time and only makes up completely false answers 37.1% of the time! It’s like Russian roulette, but worse!
olympicyes@lemmy.world 1 week ago
If you play Russian roulette with two bullets like a real man, then this model is about the same outcome!
regrub@lemmy.world 1 week ago
Surely, people won’t use the slop generator in applications where being correct is important, right?