Yann LeCun Warns of a "Bubble Explosion" in the AI Industry

Yann LeCun, head of Meta's AI division, has suggested that AI labs such as OpenAI and Anthropic are heading for a "big bubble explosion." According to LeCun, these companies' operations are effectively subsidized by investors, and operating costs are not coming down fast enough.

That said, it's worth noting that LeCun's criticism is not entirely neutral: he himself has launched AMI Labs, a startup pursuing an alternative approach, and has raised $1 billion in funding.

Midjourney Announces a Full-Body Ultrasound Scanner and Its Own Spa

Midjourney, best known for AI image generation, has announced a move in an unexpected direction. The company has developed a full-body ultrasound scanner and will open its own spa facility in San Francisco to house it.

Rumors about Midjourney's hardware ambitions have been circulating for years, but almost no one predicted a move into medical devices. How the company plans to apply its AI image generation technology to medical diagnostics remains to be seen — further details are eagerly awaited.

AI Systems Match Physicians in Diagnostic Ability — Two Nature Papers

Two new studies published in Nature show that specialized AI systems performed on par with physicians — and in some cases better — at diagnosing diseases and determining treatment plans. However, both systems run on base models that are already a generation old, underscoring the challenge of technological obsolescence.

Combined with Eric Topol's article "Agentic AI Comes to Medicine," which also gained traction on Hacker News, it's clear that practical AI adoption in medicine is accelerating rapidly.

Manus's Former Investors Plan to Buy the Company Back from Meta — Roughly $2 Billion

Former investors in Manus, the China-born AI agent company, are reportedly planning to buy it back from Meta for around $2 billion. Manus drew massive attention for a time before being acquired by Meta; the former investors apparently aim to run it as an independent business once again.

OSS Models Overtake Proprietary Models in Share — OpenRouter's Last 3 Months of Data

An analysis based on OpenRouter's data from the past three months shows that open source (OSS) models have decisively overtaken proprietary models in share. The data, which became a talking point in the LocalLLaMA community, suggests that local inference and community-driven development are going mainstream.

Relatedly, unsloth has released a GGUF quantized version of GLM-5.2. At 238GB for the 2-bit version it's massive, but it's a significant option for users looking to run large-scale models in local environments.

---