Man Follows ChatGPT's Advice and Poisons Himself \ stacker news ~AI

pull down to refresh

Man Follows ChatGPT's Advice and Poisons Himself futurism.com/man-poisons-himself-chatgpt

Guys, its important to understand AI. What it is, how it works, and most importantly what it isn't.

This guy was dumb but I read stuff from people that seem very intelligent everyday that shows me they don't get AI and are buying what Scam Altman is selling. He's a liar. Chatbots are tech prediction machines. They don't know anything.

view all related items

214 sats \ 1 reply \ @Scoresby 18h

If some guy on the internet said you can sub out sodium bromide for table salt, we'd all just say: That's just some guy. What does he know?

If chat says it, chat has a lot more authority than some guy on the internet. It's harder to say, That's just chat. What does chat know?

Chat's curious mixture of really hard to find information and occasionally completely wrong information is going to be a huge hurdle to usefulness. More and more I'm feeling edgy around the answers I get, but if I have to check everything, then what's the point of using chat in the first place?

178 sats \ 0 replies \ @kepford OP 17h

You make a good point. I think many people only think they are saving time with AI. The issue is that if you learn how to talk to a computer and give it clear instructions you can save massive amounts of time and get great results. AI is being pitched as magic and it just isn't.

63 sats \ 0 replies \ @optimism 18h

Also: #1071671

5 sats \ 0 replies \ @BitcoinHandsOn 14h

There's definitely a solid window of things that AI is great at, and where it can save you tons of time.

On the other hand, it can also be enormously frustrating when it hallucinates, and asserts very confidently that something is true, when it's absolutely not. Just recently I was looking for a feature in carrd.com, AI says it's there for sure, I'm looking for it for a long time. Then when I email support, it turns out that no, it's absolutely not possible. If AI would at least express some uncertainty! But no, it's always certain.

Overall, for my personal use cases, there's a ton of things that I've tried (coding tools using languages that I've never touched before) and been successful at, with the help of AI. And these are things that I wouldn't have even tried without AI.

But yeah...a mixed bag.

0 sats \ 0 replies \ @justin_shocknet 13h

Can anyone point to anything that would indicate there's a solution to the hallucination problem?

Years of these gaining traction with trillions poured into them, and I can't recall even one article that highlights a promising solution.

I want to be more optimistic but getting the vibe that it's inherent to the nature of all models and unsolvable. If anything they're seem to be getting worse IMO with the recursive gimmicks and retention-optimized tonality.