Hacker News Reader

Refusing a command doesn't mean consciousness. LLMs could hit a token combination that causes it tou output something like "No, I don't want to do that." It's not choosing.

View on HN

But can an LLM just refuse to process tokens because it doesn't feel like doing that at the moment? Can it look at an alternate distribution of tokens because that might be interesting? Can an LLM decide to make a drawing because it's Tuesday and sunny outside and the researchers keep asking the same questions and frankly, they really need to collaborate with each other and just leave the LLM out of it?

Claude can indeed decide to terminate conversations on its own using a special tool[1] if it feels "uncomfortable" with how the conversation is going. Also, very famously, in the middle of recording Computer Use demos, Claude stopped for a while its coding task to look at photos of Yellowstone National Park [2]

I don't think either of these two is proof of consciousness.

[1] https://www.anthropic.com/research/end-subset-conversations

[2] https://x.com/AnthropicAI/status/1848742761278611504

Reply on news.ycombinator.com