News for hackers. Stuff that matters.
Re:French AI startup Mistral releases ‘unmoderated’ chatbot ... by croes on Saturday September 30, 2023 @09:14AM (#37715128)
Nope, LLMs hallucinate, that's what you don't find online.

That adds an extra level of danger.

Re:French AI startup Mistral releases ‘unmoderated’ chatbot ... by raverbashing on Saturday September 30, 2023 @09:21AM (#37715192)
But the article is not addressing that, right?

Hallucination actually makes the problem pointed out "less worse" because maybe it will tell you how to make fire with orange juice.

Though again, people are attributing too much agency to a stochastic parrot. A parrot with a very large memory and sometimes off the rockers but still a parrot

Re:French AI startup Mistral releases ‘unmoderated’ chatbot ... by croes on Saturday September 30, 2023 @11:57AM (#37716697)
Or they ask how to treat a disease and get a poisonous answer
Re:French AI startup Mistral releases ‘unmoderated’ chatbot ... by gaganyaan on Saturday September 30, 2023 @01:22PM (#37717615)
You already get that from weird "health" bloggers. What's new?
Re:French AI startup Mistral releases ‘unmoderated’ chatbot ... by croes on Saturday September 30, 2023 @08:22PM (#37721286)
You can sue a health blogger
Re:French AI startup Mistral releases ‘unmoderated’ chatbot ... by gaganyaan on Saturday September 30, 2023 @11:55PM (#37722379)
You can sue anyone. You'll almost certainly not get anywhere for someone in the US, and any other country would tell you to pound sand. It's the same situation in practice.

Don't listen to random people for your personal medical decisions, robot or human, or you're going to have a bad time

Re:French AI startup Mistral releases ‘unmoderated’ chatbot ... by suckitsam on Saturday September 30, 2023 @09:45AM (#37715375)
It won't be long before "Quora says I can melt eggs" turns into "Google's top result says millions of fish drown in oceans each year" or somesuch.
Re:French AI startup Mistral releases ‘unmoderated’ chatbot ... by Turing_Machine on Saturday September 30, 2023 @10:05AM (#37715560)
> LLMs hallucinate

So do fiction writers. I hear there's even an entire genre of fiction that is expressly devoted to describing the commission of, and the solution to, violent crimes. And another genre that's all about war. And several genres about sex (with various focuses and levels of clinical detail). And books that discuss societal taboos ranging from incest to cannibalism. And...

Re:French AI startup Mistral releases ‘unmoderated’ chatbot ... by croes on Saturday September 30, 2023 @11:58AM (#37716708)
If you ask a question which needs a correct answer you wouldn't ask a fiction writer
Re:French AI startup Mistral releases ‘unmoderated’ chatbot ... by ThrowawayTestr on Saturday September 30, 2023 @04:53PM (#37719685)
And if I had a question that needed exactness I wouldn't ask an LLM.
Re:French AI startup Mistral releases ‘unmoderated’ chatbot ... by Ekaros on Saturday September 30, 2023 @01:16PM (#37717535)
Internet has enough basic instructions that will cause things like release of chlorine gas...

Or various ways to electrocute oneself or start fires...

Re:French AI startup Mistral releases ‘unmoderated’ chatbot ... by vunderba on Saturday September 30, 2023 @07:33PM (#37721000)
"LLMs hallucinate, that's what you don't find online"

Yes, because you can always trust the reliability of some rando's blog. Have you been on the Internet lately? /s

Re:French AI startup Mistral releases ‘unmoderated’ chatbot ... by dragonwriter on Saturday September 30, 2023 @08:24PM (#37721304)
> Nope, LLMs hallucinate, that’s what you don’t find online.

People spinning elaborate narratives out of their imagination that aren’t true is, in fact, a thing I find (including resulting in dangerously false information on topics likely to be important to people) online all the time.