Brief IA
›

Google investigates Gemini advice after a rescue

Models & LLM·Tom Levy·

Google investigates Gemini advice after a rescue

Google investigates Gemini advice after a rescue
Key Takeaways
1Three climbers from Mount Shasta were rescued after getting lost during their descent
2The sheriff's office attributes some of the difficulties to preparation based on Gemini and warns about planning via AI
3Google states it is investigating the advice generated by Gemini but has not reproduced the described responses
Why it matters : The incident illustrates the limitations of AI tools for preparing for high-risk activities and the necessity of verifying their recommendations.

Three climbers from Mount Shasta were rescued after a multi-day operation, having gotten lost during their descent. The sheriff's office mentioned an excessive reliance on Gemini for planning the ascent and warned about using AI for such preparations. Google stated it is investigating but claims it has not been able to reproduce the responses described.

Google Launches Investigation and Claims It Cannot Reproduce Responses

A representative from Google indicated that the Trust and Safety team is reviewing the climbing advice provided by Gemini. According to the company, the responses described by the climbers could not be replicated. The representative acknowledged that Gemini can make mistakes and recommended verifying the sources cited by the tool in case of doubt. The hikers did not specify what Gemini suggested regarding food and drink or the quantities, while the sheriff's office identified nutrition and hydration as problematic points.

Rescuers Denounce Dependence on AI for Ascent Preparation

The hikers reported that they planned their route using YouTube, the AllTrails app, and Google Gemini. The Siskiyou County sheriff's office referred to a major error related to the heavy reliance on Gemini, explaining that the men depended on the tool for route selection and gear preparation. The same office stated that Gemini led them to bring much less food and water than necessary and warned against relying solely on AI to organize a trip.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Timeline from Summit to Rescue and Equipment Failures

Departing at 3 a.m. on a Sunday, the group reached the summit 16 hours later. The descent went awry after one member fell and injured his leg, causing the trio to get lost. Two helicopter evacuation attempts failed before a rescue was conducted as part of a multi-day operation, with the evacuation taking place on Monday. The phones were dead, backup batteries were inoperative, and no offline map was available, which exacerbated their disorientation. The injured climber proceeded without a headlamp, in the dark, to conserve the remaining battery.

What Gemini Provides When Asked About Nutrition

An attempt to reproduce the responses described by the climbers was unsuccessful. When asked about food to prepare for an ascent, Gemini suggests combining carbohydrates, fats, and proteins. Even when requesting a low-fat option, the tool mentions commonly carried mountain foods characterized by high sodium, fat, and calorie content, such as cheese and jerky. It does not specify any quantities unless the duration of the hike is indicated.

Beyond This Case, Verifying AI Advice Remains the Rule

AI can omit essential details, with effects ranging from minor inconveniences to professional consequences. An editor found that recipes generated by ChatGPT for a week of meals were severely under-seasoned, while lawyers using AI for legal filings were officially reprimanded for errors. Generative systems predict patterns based largely on data from unverified web sources, leading experts to demand independent verification before relying on them. In the Mount Shasta case, an ascent planned with online content and AI advice took three days instead of the initially expected eight hours.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.