Jensen Huang and Trump Push Back Against Calls to Slow AI

Morning, {{first name | folks}}! AI had a busy stretch. Trump joined the AI doom debate, Claude topped a new real-world leaderboard, Bolt offered 50x more usage for build data, Google launched personalized AI stories, and Meta’s smart glasses raised new privacy questions.

Jensen Huang used the All-In Podcast to directly rebut Dario Amodei's call to slow AI development, and mid-interview, President Trump called in live to back him up: "The robots will not be taking over... the whole thing is a hoax." Huang went further, saying doomer rhetoric has "done a lot of damage" and questioning whether some CEOs sounding the alarm have "deeply conflicted" motives tied to regulatory capture rather than real concern.

Worth knowing plainly, Nvidia has a massive financial stake in AI's unrestricted growth continuing, which matters when weighing this against Amodei's essay from the same week. A separate voice entered too, Trump's AI advisor David Sacks reportedly said CEOs like Amodei should "step aside" if they can't control what they're building.

Artificial Analysis rebuilt its industry leaderboard around actual job tasks, using the same occupational data the US government uses to define real jobs. Claude Fable 5.1 ranked first across all six categories, covering areas like law, medicine, finance, and engineering. GPT-6 Astra came second in four categories, while Kimi K3, DeepSeek, and GLM-5.3 also made the top 10.

The bigger takeaway is how far open models still have to go on these benchmarks. None made the top five in any category, although several reached the top 10. That could still be enough for teams where lower costs or self-hosting matter more than getting the highest score.

Bolt.new launched Forge, a new agent running only open-source models, GLM 5.3, Kimi K3, DeepSeek v4 Pro, and it's giving every Pro user up to 50x more usage at no extra cost through October 14. The catch, and it's stated plainly upfront, opt in and your anonymized build sessions help train a new open-weight model with Arcee AI, secrets and personal data stripped out first. The performance gap is smaller than you'd expect too, these open models score 91% of what Claude Opus 5 hits on Bolt's own benchmark.

The pitch is simple: brainstorming and rough drafts shouldn't burn the same expensive credits as production work. Whether the trade is worth it depends entirely on how much you actually mind your prototyping sessions helping train someone else's model.

Google Labs has launched Dreambeans, an experimental AI app that connects to Gmail, Calendar, Photos, YouTube, and Search to create personalized stories and recommendations. It works overnight and delivers new stories each day, with AI-generated artwork based on your connected data.

Worth knowing, Dreambeans is testing a more proactive kind of AI. Instead of waiting for a prompt, it uses your activity and context to decide what might be relevant and creates something for you to discover. Google says the app is still experimental and can get things wrong.

Meta’s Ray Ban smart glasses can use their cameras for AI features without turning on the LED that signals recording. They can capture what the wearer is looking at to answer questions or identify objects, even when the light is off. Meta says its consumer AI features do not currently use facial recognition, and images used for AI responses are not stored in Meta AI conversation history.

The privacy light no longer tells the full story. Bystanders may know the glasses are not recording video, but the camera can still be active and analyzing what is around the wearer. As AI becomes more capable on smart glasses, it is getting harder to know when a camera is simply present and when AI is actually using it.