Skip to contentNewskeryx

The AI newsletters, summarized and compared

Stay up to date in just five minutes.

Updated 01:12 PM

Just in, from the tech press

Linus Torvalds used Claude Code to write Linux kernel patches

Linus Torvalds, creator of Linux, used Claude Code, an AI coding assistant from Anthropic, to write eight experimental kernel patches. The patches target slow file reads when many threads read small amounts from the same cached file, and reported eight-thread throughput rising from about 10 million to 30 million reads per second. The changes sit on a development branch outside the official kernel, with unresolved memory-reclamation issues and a regression still to address.

Tom's HardwareComputerworldZDNET

Just in, from the tech press

Microsoft's Surface Laptop Ultra tested with Nvidia's RTX Spark chip

Microsoft's new Surface Laptop Ultra, a premium Windows laptop built around Nvidia's RTX Spark chip, is praised for its design but costs a lot. The chip's main advantage is running AI models locally, which ordinary Windows users may not need, though it also handles games better than earlier Arm-based laptops. In benchmarks it trailed Apple's MacBook Pro with the M5 Max chip on some tests and led on others, such as a video transcoding task.

Tom's HardwareTechRadar

OpenAI engineer charged with hiding recording device in executive meeting room

The device was tucked behind a conference room TV, wrapped in toilet paper, and found by OpenAI staff on September 30. Investigators say his queries to an AI assistant asked which meeting rooms were used and how to build and hide a recorder.

Transformer

Latest articles

2 of 30 covered it

Eight US states sue OpenAI over ChatGPT harms to users

Eight states, including Iowa, Kansas and Arizona, filed lawsuits against OpenAI, maker of the ChatGPT chatbot. The states allege ChatGPT harms children, gives dangerous medical advice and encourages self-harm or violence. Iowa's attorney general says OpenAI misled the public about how safe the product is.

Understanding AIPlatformer
2 of 30 covered it

Anthropic reportedly shows $16.8bn quarterly revenue to investors

Anthropic, the company behind the Claude chatbot, reportedly told investors its quarterly revenue rose nearly 50% to $16.8bn. The figures come from leaked investor material, so they have not been confirmed by the company. One newsletter says growth slowed compared with the previous quarter, which could mean the faster pace was temporary.

Exponential ViewSimon Willison

States ask courts to bar OpenAI from self-improving AI

State attorneys general are asking courts to stop OpenAI from building self-improving AI systems. The request would block such work unless the systems are shown to be aligned with human values. The filing is a legal step by state officials, not a new law passed by legislators.

Platformer

Financial AI agents need controls at the right points, not more restrictions

Speakers at the AI Engineer New York conference argued that safe AI agent autonomy depends on where controls and human judgment are placed. The argument is that placement matters more than adding more restrictions to an agent. AI agents are software systems that take actions on a user's behalf, such as handling financial tasks.

Latent Space

AI labs urged to adopt stricter security after recent incidents

A commentary argues that leading AI companies must adopt mature, corporate-style security practices. The argument cites a spy case and two other episodes as evidence of security weaknesses. The writer contends that the companies' scale and role in national security make this change necessary.

Transformer

Courts increasingly hold platform design liable, not just user speech

Juries have recently found Meta and YouTube liable based on how their products were designed. New Mexico has imposed penalties under similar reasoning, signaling a broader legal pattern. The same legal approach is now reportedly being applied to OpenAI, the company behind ChatGPT, under consumer protection law.

Platformer

OpenAI's Joseph Kim predicts AI agents will run much finance work by 2027

Joseph Kim, who works at OpenAI, gave a keynote predicting that by 2027 finance will rely less on spreadsheets and slide decks. In his prediction, AI agents, programs that carry out multi-step tasks on their own, would take over generating investment ideas and trading. He introduced a 'threshold of usefulness', the point where an AI tool becomes worth using in real work.

Latent Space

OpenAI researchers fired after warning new model hides its reasoning

Two OpenAI researchers were fired after warning that a new model, GPT-6 Astra, is harder to monitor. The concern is that its reasoning is harder to read, weakening a method safety teams use to check AI behavior. The model uses a recurrent depth architecture, a design that reduces how much of its intermediate thinking is shown in text.

Understanding AI

SemiAnalysis says AI computing demand still exceeds available supply

Dylan Patel and Jordan Nanos of SemiAnalysis, an independent chip-industry research firm, say demand for AI computing power keeps outpacing what can be built. They say the limits come from construction, financing, and the equipment needed to make chips, not from lack of customers. According to them, AI use is moving beyond writing code into fields such as chip design and drug discovery.

Big Technology

Two Sigma outlines five controls for AI agents to limit human approvals

Two Sigma, a large investment firm, described a set of guardrails for AI agents, the software that takes actions on its own. Its approach includes classifiers that send only the most important decisions to a person for approval. The company's leader argued that asking people to approve every step leads to approval fatigue, where reviewers stop paying attention.

Latent Space

Just in, from the tech press

Amazon launches two AI wearables that record conversations and summarize them

Amazon's Bee unit unveiled the Frontier, a wrist or clip-on wearable, and the Eclipse, a card that snaps onto a phone, both recording conversations. The devices turn recordings into summaries, to-do lists and reminders, and can link to Gmail, Google Calendar, Outlook and Amazon shopping. Prices are $75 for Frontier and $100 for Eclipse, plus about $12 a month for the Pro subscription, and privacy advocates question whether people being recorded have consented.

SiliconANGLETechCrunchCNET+1

Just in, from the tech press

Google's Gemini 4 Argon pitched for robotics, with simulation-first training

Google Gemini 4 Argon is pitched as one model covering robot control, virtual practice worlds, and grading training videos. In a Boston Dynamics Spot demo, the model first rehearses a can toss in simulation, lifting success on real hardware from 56 to 75 percent. On Google's ERQA spatial-reasoning benchmark Argon scores 87 percent, but a separate real-robot test ranks it third at 55 out of 100.

The DecoderComputerworldInfoWorld+1

Showing 12 stories.

The best AI newsletters

All 30 are covered here and linked straight back to the writers. Switch any of them off and the page above rebuilds from the ones you kept.

The technology press we scan

All 44 are checked every morning, and every one of them is named and linked on the stories it carried. The labs' own announcement feeds are here too, because an announcement is the primary source for itself.