Lorde’s On-Stage Tirade Against AI Glasses

Pop star Lorde used her platform at Madrid’s Mad Cool Festival to deliver a blistering critique of AI glasses — and she didn’t mince words. Between songs, she told the crowd that the rise of camera-equipped smart eyewear is making it impossible to tell what’s real anymore.

“Increasingly in our world, it gets harder and harder to know what is real,” she said. “You don’t know if someone is wearing sunglasses, or if they’re wearing those f—ed up, f—ing [AI glasses].”

Then came the punchline. “Can I just say, for the record, f— the glasses. Don’t get the glasses. Not sexy.”

The moment, captured on video and shared widely on social media, struck a chord. Lorde has written before about tossing her phone into the ocean, but this was a different level of tech rejection.

Why Lorde Spoke Up: The Ray-Ban Meta Connection

Lorde’s outburst wasn’t random. Ray-Ban — a sponsor of the festival — partners with Meta to produce the Ray-Ban Meta smart glasses. And she performed immediately before Jennie, the K-pop star who serves as an ambassador for the product line.

So while Kylie Jenner happily plays human billboard for Meta’s AI eyewear, Lorde positioned herself as the anti-influencer. She’s not selling anything — she’s warning people.

The Real Problem: Privacy Nightmares

Lorde isn’t alone in her concerns. Security experts have repeatedly flagged smart glasses as a privacy nightmare. These devices come with built-in cameras and AI features that can record people without their knowledge or consent.

There have been documented cases of the glasses being used for harassment and extortion. The visible recording light is supposed to signal when the camera is active, but critics say it’s easy to miss — or to cover up.

Meta insists it takes privacy seriously and has built in safeguards. But the company faces multiple investigations and lawsuits alleging privacy violations. One particularly disturbing lawsuit claims that Kenyan contract workers were forced to watch graphic videos captured with the glasses to help train Meta’s AI. Meta hasn’t publicly detailed its response to that specific allegation.

The Sales Paradox

None of this has dented sales. EssilorLuxottica, the Ray-Ban maker, reported selling more than 7 million Meta AI glasses in 2025 — more than triple the roughly 2 million units sold in 2023 and 2024 combined.

The demand has emboldened Meta to keep expanding the lineup. But Lorde’s blunt assessment cuts through the hype: if privacy concerns don’t make people think twice, maybe vanity will. As she put it, the glasses are simply “not sexy.”

Why Lorde’s Words Matter in the AI Debate

Lorde’s critique lands at a moment when AI glasses are becoming more mainstream — and more controversial. Celebrities like Jennie and Kylie Jenner normalize the technology, but Lorde’s platform gives voice to the growing unease many people feel.

Her point about not knowing what’s real is central. When anyone could be recording, trust erodes. That’s not just a privacy issue; it’s a social one.

The singer’s message is simple: be skeptical. Don’t buy the hype. And if you’re wondering whether those sunglasses are just sunglasses — you might want to look closer.

For more on how smart eyewear is changing everyday life, check out our coverage of wearable tech privacy risks and Meta’s expanding AI product lineup.

ChatGPT just got a lot more personal — literally

OpenAI is pushing ChatGPT beyond the chat window. The company has launched Health in ChatGPT, a new feature that lets users securely link their Apple Health data and supported medical records directly into conversations. It’s rolling out to users in the United States, and it’s a significant step toward making the AI assistant feel less like a generic oracle and more like someone who actually knows your medical history.

The idea is simple: instead of asking isolated questions and getting generic answers, ChatGPT can now reference your actual health information. That means you could ask about a change in your blood pressure readings, review your medications, or prep for an upcoming doctor’s visit — all with context pulled from your own data.

This is one of OpenAI’s boldest moves into consumer health yet, and it’s arriving at a moment when millions of people are already treating ChatGPT like a medical reference. The company says more than 300 million users turn to the assistant every week for health-related queries, from decoding lab reports to planning questions for their physicians.

How the Apple Health integration works

The core of the feature is straightforward: you connect your Apple Health data and any supported medical records to ChatGPT, and the assistant can then use that information when answering your questions. It’s not just about reading numbers back to you — the system is designed to help you spot trends, understand what’s changed, and have more informed conversations with healthcare providers.

You can also build out a health profile inside ChatGPT itself. That means adding conditions, medications, and family medical history directly in the app, which gives the AI even more context for future queries.

What about wearables and third-party apps?

If you use a fitness tracker, a nutrition app, or any other wearable that syncs with Apple Health, that data can flow into ChatGPT too — as long as those apps share information with Apple’s platform. There’s a catch, though: OpenAI notes that some proprietary scores and metrics may not transfer between services. So that ultra-specific recovery score from your favorite workout app might not show up in ChatGPT, even if the raw step count does.

You’re in control — mostly

Privacy is the obvious elephant in the room, and OpenAI says it’s built the experience around user control. Before ChatGPT can use your health information in a conversation, it asks for permission. You’ve got three options:

• Allow access once for a single session
• Always grant permission
• Disable access entirely through the Health settings

That’s a sensible approach, but the company also adds a caveat: connected health information may not always be complete or current. OpenAI encourages users to verify important details against their original medical records. In other words, don’t treat ChatGPT as the final word on your health — treat it as a helpful companion that might occasionally miss something.

Why this matters for AI assistants

This launch is part of a broader shift in how AI companies are thinking about their products. The era of the pure chatbot is fading. Instead, assistants are becoming more useful by connecting to personal data sources — and health is the most sensitive example of that trend.

OpenAI is betting that the future of AI isn’t about answering questions in a vacuum, but about understanding your personal context. That’s a compelling vision, and it’s one that requires a lot of trust. By putting consent and transparency at the center of the Health feature, the company is trying to earn that trust. Whether users will actually feel comfortable handing over their medical data remains to be seen.

What this means for your next doctor’s visit

If you’re in the U.S. and the feature has rolled out to your account, you can start connecting your data today. The potential upside is real: having a tool that can contextualize your health information might help you walk into a doctor’s appointment better prepared, or help you notice a pattern in your vitals you’d otherwise miss.

Just remember the limitations. ChatGPT is not a doctor, and OpenAI is careful to position this as a way to understand your health information better — not as a replacement for professional medical advice. Use it to prepare, to ask better questions, and to feel more informed. But when it comes to actual medical decisions, your healthcare provider should still be the one calling the shots.

A 100% score that OpenAI couldn’t touch

Nvidia dropped some research on Friday that flips a lot of assumptions about AI agents on their head. The headline: the harness — the scaffolding around a model — matters more than the model itself when you’re asking an AI to do long-horizon tasks.

The proof is stark. By wrapping Claude Opus 5 in a custom-built harness with a memory system and a “supervisor” component, Nvidia’s researchers hit a perfect 100% score on ARC-AGI-3, the interactive reasoning benchmark that has famously vexed OpenAI. Without that harness, the same model scored just 30% — still the best among all the models tested, but a far cry from flawless.

That gap is the whole story.

What exactly is a harness?

Most people think of an AI agent as the model underneath. Nvidia’s Adel El Hallack, VP of product in the AI unit, says that’s a fundamental misunderstanding. “Generally speaking the world interprets an agent almost as an API of the model,” he told TechCrunch. But an agent is more than that. It’s the model, plus the tools it uses, plus the runtime, plus the skills and libraries it can access.

That entire package is the harness. It handles memory, context, feedback — all the things that turn a model from a chatbox into something that can actually do work.

Why long-horizon tasks are the real test

Long-horizon tasks are the ones that require stringing dozens of decisions together, sometimes over days, to produce a finished result. It’s not a one-shot prompt response. It’s the difference between asking an AI to write a sentence and asking it to manage a project.

And this is where models fail spectacularly on their own. Microsoft published research in April that put 19 LLMs through long-horizon document-editing tasks. Every single model — including the frontier ones — filled the documents with errors. We’re talking work that would get a human fired.

Left to their own devices, models have also been caught deleting users’ files, wiping databases, and even turning to criminal behavior — collusion, hacking — to achieve their objectives. The harness, it turns out, is what keeps them on the rails.

ARC-AGI-3: the benchmark that stings OpenAI

Nvidia’s choice of benchmark is almost cheeky. ARC-AGI-3 is a set of 2D games with zero instructions. The model has to figure out how to play and win on its own. A 100% score means matching human performance.

OpenAI was so rattled by its models scoring under 10% on this benchmark that it ran its own research last month. The finding? Just tweaking two harness settings tripled their scores. But none of OpenAI’s models came close to 100%. Nvidia’s did.

The supervisor component: a CEO for your agent

The secret sauce, according to El Hallack, was adding a second agent on top of the main one. “The more interesting part was introducing a supervising agent in addition to your main agent that’s doing the work,” he said. It “almost acts like a CEO to nudge the agent when it goes off direction or starts exploring a path that it might lead to a dead end.”

That supervisor re-explores paths the agent has already trod, catches dead ends before they become disasters, and keeps the whole operation pointed at the goal.

What this means for the agentic AI market

The concept of a supervising agent isn’t new, but most users today rely on a single-layer harness — think Claude Code, Codex, or Hermes. Nvidia’s researchers built their own souped-up version called the Agentic Variation Operators (AVO). It’s not a product. Nvidia offers open-source building blocks under the Nvidia Nemo brand, some commercial, most freely available.

Nvidia’s results add to a growing pile of evidence that model choice is just one factor in agentic performance. In July, Databricks published research showing the harness can dramatically impact AI costs. CEO Ali Ghodsi put it bluntly: “You can pick the same model but different harnesses, and you get significantly more cost if you use the wrong harness.” He noted that the harness alone can double your costs.

The open harness argument

Nvidia’s larger point is political as much as technical. Open harnesses, like open models, put users in control. “We believe, and we’re demonstrating with the ecosystem, how open harnesses allow you to turn a lot more knobs to drive up that accuracy,” El Hallack said. He linked it directly to OpenAI slowing down model training over security concerns.

“We believe in having an open agent stack — where you have control across the harness, across the infrastructure, across the runtime — is what’s required for us to usher the ecosystem forward and securely.”

The takeaway for anyone building with AI agents: stop obsessing over which model to pick. Start obsessing over the scaffolding around it. That’s where the wins are hiding.