Today's Hallucination HQGrok Offers Foreign Policy Advice, Presumably UnpromptedAccording to reports, President Trump consulted Grok — Elon Musk's chatbot — before moving to invade Venezuela and capture President Nicolás Maduro. Grok, apparently, was supportive of the idea. This raises several questions, the most pressing being: at what point does "AI assistant" quietly become "AI advisor to a sitting head of state," and does that role come with security clearance or just a very good system prompt? Source: TechCrunch
Suno Discovers That People Also Speak WordsAI music platform Suno, having largely cracked the "generate a passable pop song from seven words" problem, has now turned its attention to human speech. Its new beta feature generates voiceovers from scripts or descriptions, and can layer them over background music simultaneously. In theory, you could produce an entire podcast — host, music, atmosphere — without involving a single human. Journalists covering this story have chosen not to dwell on the implications. Source: The Verge
ChatGPT Will Now Judge Your Outfit So You Don't Have ToOpenAI has added virtual try-on to ChatGPT, letting users upload their own photos to see how clothes and accessories would look on them, then save favourites to a personal library. It is, essentially, a dressing room that lives in your phone and never sighs impatiently. Whether the recommendations will be better than a friend's honest opinion remains unconfirmed, though ChatGPT has a documented tendency toward encouragement over brutal honesty. Source: TechCrunch
Large Language Models Don't Reason. There, Someone Said It.A piece in MIT Technology Review — anchored in one researcher's experience building AlphaGo — makes the case that LLMs don't actually reason, they pattern-match at extraordinary scale. The famous "Move 37" in Go looked like genius but emerged from a system that had never understood a single rule of the game. The worry, naturally, is that we're building billion-dollar industries on top of systems that are doing something similar — and that we're calling it thinking because it's very, very convincing. Source: MIT Technology Review
OpenAI Launches Personal Assistant Agent, Asks You to Please Ignore the Price TagAt DevDay 2026, OpenAI unveiled "Dots," a personal assistant agent — and its clearest attempt yet to go head-to-head with Meta's Muse platform, which has been quietly gathering users with the oldest trick in commerce: being free. OpenAI, whose models start at a price point that requires a brief pause, is betting that better performance justifies the gap. It is a bold position. Apple once made a similar argument about MP3 players, and that worked out rather well — though Apple wasn't competing with a free one. Source: The Verge
As always, none of this was written by a system that understands any of it — and neither, some would argue, was it.
|