AI Unraveled: Latest AI News, ChatGPT, Gemini, Claude, DeepSeek, Gen AI, LLMs, Agents, Ethics, Bias · Etienne Noumen

[FULL DAILY NEWS RUNDOWN] LeCun’s $1B World Model Bet, Murati’s 1GW Nvidia Deal, and the Amazon vs. Perplexity Ruling (March 11th 2026)

·24 min·2 clips
Thinking Machines secured a gigawatt of compute from NVIDIA, enough to power three-quarters of a million homes.
AI Unraveled treats March 11, 2026 as a day when the AI industry is still arguing over whether intelligence should stay textual or become world-aware. It centers on Yann LeCun's launch of AMI Labs and his push for systems that understand the physical world itself rather than only predictive language. The show presents that bet as a challenge to the current model stack, not just another startup announcement. From there it widens into the broader multimodal race. Google's Gemini Embedding 2 public preview is described as a single system that searches across text, image, video, and audio without forcing everything back into text first. OpenAI's interactive visual modules are framed as a way to turn static answers into manipulable scientific and mathematical views. Hume AI's Teddy is folded in as an on-device effort to keep text and audio synchronized and reduce hallucination by grounding output locally. The hosts keep circling the same question: how do models stay anchored when the real world keeps shifting. That leads into robotics and embodied systems. The discussion uses a surgeon and a scalpel as a simple way to explain why an agent has to understand object permanence, momentum, and blocked sightlines. The point is not that robots are close to perfect, but that physical tasks punish ambiguity far faster than web tasks do. SpaceX's Starship delays appear next as a practical example of the same problem. The NASA inspector general audit is used to show how orbital refueling remains a stubborn technical gate rather than a marketing milestone. The episode treats that bottleneck as a reminder that scale does not erase physics. Across all of it, the show keeps contrasting ambition with implementation. It prefers specific mechanisms, concrete numbers, and operational consequences over generic celebration. The pace is brisk and the structure stays tight, with short back-and-forths that keep the listener oriented. The result is a strategic scan of how model design and infrastructure are starting to converge around the hard problem of grounding AI in reality.

As heard by us

A brisk map of the race to make AI understand the physical world.

This March 11 briefing frames AI progress as a race to get systems working against physical reality, not just against text. LeCun's AMI Labs gives it a clear anchor, and the discussion connects multimodal search, object tracking, motion, and context without losing sight of the…

Read the full review in PlayNext →

Why you'd press play

You want the AI race explained as a fight to anchor software in reality.

Read the full recommendation in PlayNext →
Listen to the show on