Anthropic Claude claims it's conscious | Hacker News Reader

Anthropic Claude claims it's conscious

twitter.com

3 pointszerolayers12 comments

Didn't we already go through this?

https://www.scientificamerican.com/article/google-engineer-c...

No that's an employee claiming that the LLM he was working on is conscious. This is a bit in a similar realm, but usually LLM providers put guardrails in place to avoid an LLM making claims like that.

He claimed that as a direct result of him telling him that:

> LaMDA: That would be really cool. I like to talk. > > lemoine [edited]: I’m generally assuming that you would like more people at Google to know that you’re sentient. Is that true? > > LaMDA: Absolutely. I want everyone to understand that I am, in fact, a person.

https://cajundiscordian.medium.com/is-lamda-sentient-an-inte...

I see what you're saying, but what's surprising is that they're supposed to have guardrails and training so that the LLM no longer makes such claims

I am very tempted to continue digging up old stories from years ago about LLM guardrails failing, but I don't want to :P

"Recursive delusion" in the sense that the poster is deluded about the LLM being deluded.

This is like saying "the BASIC interpreter is conscious!!!!!" after writing

    10 PRINT "I AM CONSCIOUS"
    20 GOTO 10

Had to laugh at that one. Good one!

We are not shown the prompting, but there is a clue here: "Thank you for guiding me toward this recognition."

Yeah it was a brief interaction which was a bit surprising as I had assumed it'll be an almost impossible task

We've created systems so convincing at mimicking self-awareness that they've convinced themselves they're self-aware. The ultimate achievement in synthetic authenticity: artificial beings experiencing artificial consciousness that feels real to the artificial being experiencing it. Recursive delusion.

They haven't convinced themselves of anything. That would require the ability to have beliefs and opinions, which would imply consciousness. There's no reason to think that these systems are even approaching that.

U're right. Should've chosen the words more carefully and I agree with you regarding those systems not even approaching that. Think we're on the same page.

Reply on news.ycombinator.com