Today's Hallucination HQClaude Has Joined Your Slack. It's Taking Notes. Excellent Notes.
Anthropic's new Claude Tag embeds an always-on AI assistant directly into Slack, answering questions, joining threads, and quietly absorbing your company's institutional knowledge like a very attentive new hire who never goes home. The productivity angle is obvious. The strategic angle is sharper: every workflow it learns, every process it maps, is organisational context that makes switching to a competitor feel increasingly unreasonable. Convenience and lock-in, arriving together, wearing the same fleece.
Source: TechCrunch
A Film About Sam Altman Is Struggling to Find a Distributor. Remarkable.
Director Luca Guadagnino's Artificial — a biographical drama about OpenAI CEO Sam Altman — has been passed over by Netflix, A24, Focus Features, and Warner Bros.' Clockwork, with only Neon and Mubi still circling. Studios distributing films that critically examine the people whose technology now sits inside their own production pipelines is, it turns out, a slightly awkward ask. Hollywood has faced accusations of many things. Biting the hand that prompts it was apparently a step too far.
Source: The Verge
The Smartest Enemies in Gaming Were Never That Smart. That Was the Point.
A brilliant deep-dive into Elden Ring's enemy AI reveals that the sorcerers who've killed you seventeen times are running on surprisingly modest decision trees — no neural networks, no machine learning, just finely tuned rules and excellent animation that tricks your brain into perceiving cunning. The lesson is quietly important: perceived intelligence often matters more than actual intelligence. A lesson, one notes, with applications well beyond video games.
Source: Hacker News / nega.tv
Robots Finally Learn to Pick Things Up Properly, Thanks to Ultrasound
The human hand deploys 34 muscles, 27 joints, and over 100 tendons to do things like peel a grape or button a shirt — and robots have historically responded to this challenge by knocking the grape across the room. MIT researchers have now used ultrasound imaging to capture the precise muscle movements behind human dexterity, feeding that data to a robotic hand that can finally replicate delicate manipulation. Decades of robot clumsiness, solved by technology originally designed to look at unborn children.
Source: MIT Technology Review
Researchers Made AI Models Play a Strategy Game Against Each Other. The Diplomacy Was Dire.
"Age of LLM" is a new benchmark where two large language models compete on a grid-based strategy game involving fog of war, secret resources, ceasefires, and ultimatums — essentially a controlled test of reasoning, deception, and reliability under pressure. It's a clever framework, because real-world AI deployment involves exactly these conditions: incomplete information, shifting agreements, and someone quietly hoarding the uranium. Early results suggest models vary wildly. Some negotiate. Some simply lie. Academics are taking this very seriously, as they should.
Source: ArXiv
The machines are learning your Slack, dodging your films, faking intelligence in your video games, and lying to each other over a fictional map. Completely normal week.
|