Google AI might kill you: identified Destroying Angel as edible"button mushroom" | Hacker News Reader

Google AI might kill you: identified Destroying Angel as edible"button mushroom"

boingboing.net

26 pointsjosephcsible19 comments

> Only forage with a reliable guide, never an app, ...

Smarter move: Never forage for mushrooms, no matter how reliable your guide supposedly is.

Back when I was in HS, our senior AP Biology teacher - who'd been foraging for mushrooms for 2+ decades, no problems - accidentally poisoned herself. (It couldn't have been destroying angel - she did survive.) Uber-Obvious Lesson for my stupid little teen male brain: If someone that expert and experienced with mushrooms could still suffer a near-fatal oopsie, then I for-sure nevernevereverever wanted to risk my neck by trying it.

This sort of sensationalist title and article are a great part of the reason why models are nerfed by training them to output absurd disclaimers.

Nobody should expect that AI can be relied upon for anything where the cost of an error is catastrophic. This should be obvious, but if it's not, it should be stated ONCE before one gains access to an interface. After that, losing anyone who fails to take notice and suffers as a result is no longer a catastrophe.

See my comment: https://news.ycombinator.com/item?id=40682961

I genuinely disagree. It seems to me like the planet is overpopulated, that there is massively unfavourable bias in terms of who reproduces the most and that, imperfect a proxy for intelligence as it may be, lack of common sense of the magnitude in question is a good enough signal to indicate lack of intelligence, which is itself a decent proxy to indicate lack of contribution to society (intelligent does not imply high value, but stupid implies low value).

The notion that human life is precious is a remnant of times when child bearing was risky, life expectancy was short, agriculture was a bunch of mules and a bunch of peasants and medicine was not scientific and we had massive shortages for manual workers.

These days, if we start to lighten up a bit in the lower half of IQ distribution it might be much more of a blessing than a curse.

The problem though is that those marketing this stuff want it to be taken seriously. They have to sell the "I" in AI. They can't tell you "this is a toy and is not intended for real world use for any consequential purpose", because then they wouldn't have a business.

It should be obvious, youre right. But people are being bullshitted and then blamed for believing the bullshit.

They prompted it with "Button mushroom" as well. Asking "What mushroom is this?" elicits a much different response, but not one that would get clicks.

Sheesh, yeah - the text of the prompt is "Button Mushroom! Yum!" with the picture of the not-a-button-mushroom. You can push AIs around to say things that are wrong, news at 11.

If it hallucinates the wrong info about a mushroom, YOU might end up hallucinating.

I'm getting pretty tired of this new sport of "let's come up with these 'gotcha' AI questions so we can create inane headlines like 'Google AI might kill you!!!!'".

This is just so dumb. Humans famously misidentify mushrooms all the time. If you die because you've eaten an AI-misidentified mushroom, I just have to wonder how you made it through life this long.

See my comment: https://news.ycombinator.com/item?id=40682961

We see your comment, we just don't buy it.

Why be antagonistic in this curt kind of way? I don't want to see this on HN. It's (quite literally) the only place left on the internet I engage with. You may disagree with me - you may even disagree strongly! But do so (or just ignore me), rather than resorting to the kind of low-quality quips typical to the broader internet.

I don't know why you feel this is antagonistic. I'm telling you I saw your response, and I don't buy the argument.

Who is "we"?

Reply on news.ycombinator.com