Search Results
8/10/2025, 7:49:37 PM
>>512706489
Well here's me doing a stupid question torture test the other day on both gpt-5 vs. qwen3 4b thinker. A tiny little open model that will run on almost any PC.
Gpt-5 suggests that Aldous Huxley could not have had anything to do with 1984 because he died in 1963. (But the book was published in 1949, not 1984)
I didn't actually know George Orwell's private name when I came up with the question.
So gpt-5 shat out misinformation (the implication that 1984 was published in 1984) while the little Qwen model dug up a ton of information and offered some possibilities as to what I had gotten wrong.
Tiny little chink model that can run on a potato wins.
Well here's me doing a stupid question torture test the other day on both gpt-5 vs. qwen3 4b thinker. A tiny little open model that will run on almost any PC.
Gpt-5 suggests that Aldous Huxley could not have had anything to do with 1984 because he died in 1963. (But the book was published in 1949, not 1984)
I didn't actually know George Orwell's private name when I came up with the question.
So gpt-5 shat out misinformation (the implication that 1984 was published in 1984) while the little Qwen model dug up a ton of information and offered some possibilities as to what I had gotten wrong.
Tiny little chink model that can run on a potato wins.
8/9/2025, 6:23:50 PM
Out of Distribution torture test: Stupid question edition.
GPT-5 vs. Qwen3-4B-Thinker
<spoiler>GPT-5 gets mogged on, by a tiny little model you could run on your grandmother's vibrator</spoler/
And this is why benchmarks are worthless.
GPT-5 vs. Qwen3-4B-Thinker
<spoiler>GPT-5 gets mogged on, by a tiny little model you could run on your grandmother's vibrator</spoler/
And this is why benchmarks are worthless.
Page 1