From our other sites
The story
Three hikers set out for a day trip up California’s Mount Shasta, planning their route based on advice from Google’s Gemini AI — but according to a CNET Japan report, what was supposed to be an 8-hour hike turned into a two-day ordeal marked by shortages of food and water. On 5ch’s Science News+ board, the incident sparked both criticism of blindly trusting AI answers without verifying them and concern over hallucinations (AI responses that mix in inaccurate guesses as if they were fact) said to be especially common with Gemini. The thread was split between blaming the AI itself and blaming the poor judgment of the people who used it.
Stranded on Gemini’s advice, short on food and water — sloppy hike planning exposes the dangers of “AI dependency”
Relying on AI-generated advice was a “serious error in judgment.”
The three hikers who took on California’s Mount Shasta saw their planned 8-hour climb turn into a grueling two-day ordeal.
Source: japan.cnet.com / Original article here
What people said
Feed the AI this valuable failure case and maybe it'll give slightly less terrible answers next time.
Anyone who outsources all their judgment is the real idiot.
Now: "My source is Gemini!" (smirk)
Find sources you can actually trust.
And think for yourself.
Blindly believing some faceless other will be your ruin.
Anyway, glad you made it back alive.
But that was pure luck.
Learn this lesson the hard way.
But no, someone built in the logic that "the word 'impossible' is a liar's word" (a well-worn Japanese motivational-speak cliché)…
Just guessing, though.
It's just not built to suppress hallucinations.
Idiots get even more idiotic,
smart people get even smarter —
it's a tool that amplifies human intelligence.
It just reinforces whichever way you're already leaning.
Yeah, seriously, that's bad.
All it does is spout stuff scraped from the internet — pure armchair talk from someone who's never done it (エアプ).
So trusting something like that to go climbing makes you the fool.
Same goes for you.
>The author tried to get Gemini to reproduce the response the hikers reportedly received, but couldn't recreate it.
I asked Gemini about this and it said, "It pains me as the party involved, but don't take AI answers at face value," lol
Looks like it's already learned from this one.
Let's go ask Gemini about it, lol
Guess people these days find that too preachy? Starting out on a relatively easy climb
to gauge your own fitness level matters too.
Background and key points of the debate
In the CNET Japan report this article draws on, the writer later tried to get Gemini to reproduce the response the hikers reportedly received, but couldn’t draw out the same answer (see post #20). Since generative AI output can vary between runs even for an identical question, flatly declaring that “Gemini gave specifically bad advice” may be premature. The thread split between those blaming a flaw in the AI itself and those blaming the hikers for outsourcing their judgment entirely, with the latter view slightly more dominant. Cases of people blindly trusting GPS or car-nav directions and ending up stranded have long been known, and some framed this incident as just an extension of that pattern. There were also comments pointing to the AI’s habit of answering with vague guesses instead of a clear “that’s not possible” (i.e., hallucination) — but that’s a technical challenge facing large language models in general, not something unique to Gemini, so reading it as one product’s flaw is a bit off the mark.
*This article is compiled from excerpts and a summary of the 5ch (Science News+) thread “[AI] Stranded on Gemini’s advice? Short on food and water — sloppy hike planning exposes the dangers of “AI dependency”.”
Leave a Reply