Today's Hallucination HQFord Discovers That Decades of Engineering Expertise Cannot, In Fact, Be Deleted and Replaced with a ChatbotFord has quietly rehired a cohort of experienced engineers — internally dubbed "gray beards" — after betting too heavily that AI could absorb their institutional knowledge and produce vehicles of equivalent quality. It could not. A Ford executive admitted they "mistakenly thought" AI alone would deliver a high-quality product, which is a remarkably polite way of describing what was presumably a very expensive lesson. Turns out wisdom accumulated over thirty years doesn't export cleanly to a prompt. Source: TechCrunch
Brown University Discovers Its Students Are, Broadly Speaking, Tremendous Optimists About Getting Away With ThingsA professor at Brown University has publicly denounced what appears to have been near-wholesale AI use on a recent exam — not a handful of students bending the rules, but something approaching a collective enterprise. The tell, as ever, was the writing: confident, fluent, and structured in ways that bore little resemblance to how actual humans think under exam pressure. Brown is a prestigious institution. Its students are clearly smart enough to use AI. Whether they were smart enough to use it undetectably is, apparently, another matter entirely. Source: El País
Man Googles How to Commit Arson, Allegedly Commits Arson, Is Surprised This Sequence Is DiscoverableJonathan Rinderknecht is facing arson charges in connection with a New Year's Day 2025 fire that became one of the deadliest wildfires in Los Angeles history. Prosecutors built their case from location data, security footage, and witness testimony — but also from his ChatGPT conversation logs, which apparently contained relevant searches made before the fire. The case ended in a mistrial on procedural grounds, though not before establishing a useful principle: if you ask an AI something incriminating, it does keep notes. Source: The Verge
Man Asks Claude to Read His MRI; Medical Profession Waits Nervously to See How This GoesA developer named Antoine has published a detailed account of using Claude — Anthropic's AI — to help interpret his own MRI scan, feeding it the raw image data and asking for a structural analysis. The results were, by his account, genuinely useful in helping him understand what he was looking at and what questions to raise with his doctor. To be clear: he used it as a second opinion, not a replacement for one. It's a fine distinction, but an important one, and he deserves credit for not simply declaring himself cured based on a confidence score. Source: antoine.fi
Wall Street, Still Giddy From the Last One, Has Found Itself a New FavouriteHaving watched Nvidia's valuation perform manoeuvres that would embarrass a rocket ship, investors are now eyeing Micron — an American manufacturer of the memory chips that AI systems consume in vast quantities — as the next potential beneficiary of the AI infrastructure boom. The logic is sound: AI models need enormous amounts of fast memory, Micron makes it, and demand is not shrinking. Whether Micron is "the next Nvidia" or merely "a very good Micron" remains, as ever, a question best answered in hindsight. Source: TechCrunch
AI: occasionally unreliable, frequently surprising, and — as Ford, Brown, and one unfortunate arson suspect have all learned this week — always paying attention.
|