Today's Hallucination HQCloudflare Tells AI Crawlers to Sort Themselves Out by September
Cloudflare has given AI companies until September 15 to separate their search crawlers from their AI training crawlers — or face being blocked across a swath of publisher sites by default. The logic is simple: if you're scraping content to train a model rather than to index a search result, publishers would quite like to know, and quite like to be paid. Voluntary compliance from an industry famous for moving fast and asking forgiveness later. What could go wrong.
Source: TechCrunch
AI's Dirty Secret: Ask for a Random Number, Get 7. Every Time.
Ask any major AI chatbot for a random number between one and ten, and it will almost certainly say seven. Not because it computed it — because seven feels random, and language models are, at their core, very sophisticated pattern-matchers with commitment issues. A startup is now trying to fix this "groupthink" problem baked into LLMs, which tend to cluster around consensus answers rather than genuinely explore the space. Statistically speaking, your chatbot has the range of a nervous student guessing on an exam.
Source: MIT Technology Review
Serial Entrepreneur Puts $30M of His Own Money Where Microsoft's Lunch Is
Bhavin Turakhia — a man apparently unbothered by his fourth successful tech venture and looking for a fifth challenge — has personally committed $30 million to build Neo, an AI-native alternative to Microsoft Office and Google Workspace. This is either visionary or the most expensive productivity experiment since someone decided PowerPoint needed transitions. Turakhia has sold companies before and clearly has the appetite for a fight. Microsoft, for its part, has roughly $200 billion in annual revenue and presumably will not immediately panic.
Source: TechCrunch
Beautiful Speaker. Shame About the Brain.
Google's new Home Speaker is, by most accounts, genuinely lovely hardware — the company's first smart speaker in years and apparently worth the wait, aesthetically. The problem is Gemini, Google's AI assistant, which powers it and currently delivers the sort of conversational experience you'd expect from someone reading from a script they've only half-memorised. Smart speakers have spent years struggling to justify their existence beyond timers and music; pairing one with an AI that isn't quite ready is a bold interpretation of "solution."
Source: The Verge
AI Takes a Crack at the Diagnoses Doctors Spend Careers Learning to Make
Rare disease diagnosis is one of medicine's hardest problems — patients often wait years, seeing specialist after specialist, before anyone lands on the right answer. RareDxR1, a new model out of academic research, attempts autonomous medical reasoning on rare diseases without relying on human-annotated training data, working instead from complex, unstructured patient symptoms. It won't replace specialists, and nobody sensible is suggesting it should. But as a tool for narrowing a search space that can contain thousands of conditions, "occasionally useful" would already be something.
Source: ArXiv AI
AI: confidently answering everything, randomly generating sevens.
|