Hacker News Reader

It was a long chat - https://g.co/gemini/share/6d141b742a13 and then the last question contained text that was completely broken. It's not surprising to have a failure case under such conditions.

And it is reasonable to have failure cases. But systems should fail gracefully. This wasn't a graceful failure.

If you look at the other two drafts, it was perfectly able to answer the question.

So it's not hate speech if it's written by an LLM?

There's legal precedent to hold accountable people that encourage others to kill themselves.

Nobody that we know of has yet been hurt by this. I think it's very unfair and unhelpful to call people who flag issues like this "whiners". Would you accept a chainsaw which worked fine 99.9% of the time, but occasionally just exploded even with normal care and use? Why should LLMs be subject to a different set of expectations from any other tool?

LLMs cannot explode. People need to be educated that LLMs are as trustworthy as a random page or post on the internet.

> People need to be educated that LLMs are as trustworthy as a random page or post on the internet.

... or a Croudstrike or Microsoft patch. /s