Hacker News Reader

GPT-5 still has a terrible personality.

"Yeah -- some bullshit"

still feels like trash as the presentation is of a friendly person rather than an unthinking machine, which it is. The false presentation of humanness is a huge problem.

View on HN

I feel strongly about this. LLMs should not try to write like humans. Computer voices should sound robotic. And when we have actual androids walking around, they should stay on the far side of the uncanny valley. People are already anthropomorphizing them too much.

It can't, though. It's language. We don't have a body of work constituting robots talking to each other in words. Hardly fair to ask LLMs not to write like humans when humans constructed everything they're built on.

These models are purposely made to sound more 'friendly' through RLHF

The chat that rejects you because your prompt put it in a bad mood sounds less useful.

How about the emotionless servant chat?

It's perfectly capable of writing in a succinct and emotionless style, which people refer to as robotic.

Reply on news.ycombinator.com