Nobel Laureate John Jumper Leaves DeepMind for Anthropic

John Jumper, who won the 2024 Nobel Prize in Chemistry for his work on AlphaFold, has reportedly left Google DeepMind to join rival Anthropic. As team lead for AlphaFold, Jumper solved protein structure prediction — a long-standing grand challenge in biology — and revolutionized drug discovery research. According to TechCrunch, several other prominent researchers are also leaving DeepMind beyond Jumper.

The move underscores the intensifying talent war in the AI industry. DeepMind has historically attracted top talent backed by Google's resources, but rivals such as Anthropic and OpenAI are drawing in leading researchers by emphasizing safety and research freedom. What Jumper will work on at Anthropic has not yet been disclosed.

EXL3 Quantized Models Can Now Be Converted on Apple Silicon Macs

The Reddit LocalLLaMA community has reported that EXL3, a high-quality quantization format that previously could only be converted in NVIDIA CUDA environments, is now capable of both conversion and inference on Apple Silicon (M-series) Macs as well.

According to the developer, conversions succeeded for MiniCPM5 and Qwen3.6-27B, and the conversion quality (average KLD) for MiniCPM5 was on par with models converted on RTX-equipped machines. EXL3 boasts a top-tier quality-to-size ratio among quantization formats available on consumer hardware, reportedly outperforming MLX quantization by an average of roughly 0.5 bits per weight.

High-fidelity quantization that previously required RTX setups with 96GB to 128GB of VRAM becoming available on Macs with 64GB or more of unified memory is a significant step forward for the democratization of local LLMs.

SoftBank Rolls Out AI-Driven "Patching as a Service" in Japan

According to Light Reading, SoftBank has begun offering an AI-powered security patching service in Japan. The service reportedly uses AI to determine patch deployment priorities and automatically carries out deployments.

This is one real-world example of the "AI for IT operations" trend, and it is drawing attention as an effort by a Japanese company to put AI to practical use in the security domain. However, only limited information has been disclosed so far regarding detailed technical specifications and scope.

Granta Magazine Halts Publication of Short Story Prize Winners Amid AI Controversy

The Guardian reported that Granta, the literary magazine, has dropped its publication of Commonwealth Short Story Prize-winning works due to a controversy over AI. Amid suspicions of AI-generated text involvement, the magazine decided to withhold publication.

AI's impact on the literary world is deepening, throwing into sharp relief the challenge of how to determine the authenticity of creative works. Similar controversies could arise at other literary prizes and publications, and the boundaries between AI-generated content detection, copyright, and ethics are likely to remain a subject of ongoing debate.

The Reality of AI Coding: Faster MVPs, Slower Reviews, and Security Challenges

An article analyzing how AI-assisted coding actually affects real-world development is drawing attention. The piece, which gained traction on Hacker News, argues that while AI coding accelerates the speed of building MVPs (minimum viable products), code review time actually increases, and security concerns tend to be overlooked.

AI-generated code may appear to work on the surface while containing security vulnerabilities or inefficient designs, and the human review burden required to catch these issues is rising. As AI coding tools continue to spread, managing "the cost of speed" has become a critical challenge.

---