When AI Gets It Wrong With Total Confidence

Three different AI screwups this week, and they all involve the same problem: the systems are confidently wrong.

OpenAI's Big Launch Was a Mess Before It Started

OpenAI dropped GPT-6 Astra yesterday, calling it a "generational leap in capability" in Altman's announcement post — and within hours, Sam Altman was publicly apologizing because paying subscribers couldn't actually access it. This isn't the first time OpenAI has botched a rollout (the GPT-4o voice launch, the Sora waitlist), but the pattern is becoming its own story. If you're a business that has built workflows or client promises around OpenAI's paid tiers, this is worth taking seriously: they keep announcing products their servers can't actually handle yet. This has happened at least three times now. You're paying $20/month and you still couldn't log in. At what point do you stop building on their infrastructure?

https://www.theverge.com/ai-artificial-intelligence/990060/altman-apologizes-messy-astra-rollout

Instagram Is Slapping "AI Content" Labels on Real Photos

Meta's system for flagging photos generated by AI, not taken by a photographer, has started mislabeling real human photographs as AI-made. For creators, small business owners, or anyone using Instagram to build credibility, this is genuinely damaging: an "AI Content" label on your authentic product photo or portfolio shot tells your audience you faked it. The deeper problem is that these detection systems are statistical guesses, not forensic facts, and Meta is treating them as ground truth. Appealing a wrong label through a platform this size is about as fun as disputing a credit card charge with an automated phone tree.

https://www.theverge.com/ai-artificial-intelligence/989617/instagram-ai-content-label-confusion

Why AI Food Photography Looks Like a Fever Dream

Restaurants and food brands are increasingly replacing real food photography with AI-generated images, and the results — including what one reviewer called "donut shrimp" and "wormlike noodles" — are becoming their own aesthetic disaster. The reason AI struggles specifically with food is that current image generators learn from patterns, and food's appeal is deeply tied to texture, physics, and cultural specificity that the training data doesn't capture cleanly. A customer who sees an AI-mangled photo of your sandwich before ordering may not order. The cost savings on a photo shoot are real. So is the customer you just lost.

https://www.theverge.com/ai-artificial-intelligence/989376/ai-generated-food-why-does-it-look-like-that

AI in Job Applications Is Making Everyone Lose

Wired reports that job seekers are using AI to mass-generate applications and tailor resumes, employers are using AI to screen them out, and almost no one is getting what they want from the process. The failure isn't just technical — it's that both sides are now optimizing for each other's AI rather than for an actual hire. What this means practically: if you're hiring, volume no longer signals interest, and a polished application tells you almost nothing. If you're job searching, AI-assisted applications may be actively hurting you by making your materials indistinguishable from thousands of others running the same tools.

https://www.wired.com/story/ai-job-market-infinite-doom-loop/

A Storage Company Is Gunning for Your Smart Home

Ugreen, a brand you probably know from cheap-but-good phone chargers, announced a home platform at IFA this week that combines local AI processing, security camera storage, and smart home controls in one box that runs in your house rather than a company's cloud. "Local AI" means the processing happens on hardware you own, not on a server somewhere that could change its terms, raise its price, or get acquired. This is a niche product today, but it points toward a real consumer shift: after years of privacy concerns about smart home devices phoning home constantly, there's a growing market for systems that simply don't.

https://www.theverge.com/tech/990006/this-nas-company-wants-to-run-your-local-smart-home

At some point the cost of AI getting things confidently wrong — your photo, your job application, your lunch order — starts to exceed whatever efficiency it bought you.

← More daily briefs