4 ms·
That gpt 4.5 was 73% successful is fascinating. It is almost as if humans have a fundamental flaw in detecting other humans which the LLM (+ the prompt )exploit
by sorokod 2y ago
That gpt 4.5 was 73% successful is fascinating. It is almost as if humans have a fundamental flaw in detecting other humans which the LLM (+ the prompt )exploits.
- benlivengood 2y agoWe literally build modern models out of RLHF finetuning; the response styles that people like/engage with/approve the most are what the models generate.