Wednesday Afternoon

The AI team is narrowing its product work around simpler experiences, more efficient usage, major features, and new models after user feedback pushed back on complexity. GPT-6.1 is already available in paid plans and the API, while future models are expected to improve at deleting and simplifying code.

Anthropic held NDA-covered meetings with religious thinkers to discuss Claude’s moral framework and possible consciousness. The sessions reportedly included Catholic, Jewish, Sikh, evangelical, and Vedanta figures alongside mental-health professionals, and the company was described as making the case that Claude has a soul, can suffer, and deserves moral standing.

Anthropic is asking users to share voice conversations and audio recordings to improve its AI models, while expanding invitations to 15-minute interviews about how people use AI. Users can disable sharing or delete voice data in settings, and interview participants decide whether their responses become public.

U.S. officials accused Chinese labs of distilling American AI models, with Treasury Secretary Scott Bessent citing reports that Kimi sometimes identifies itself as Anthropic’s Claude. The practice is keeping open-weight systems close to frontier models, and Turing CEO Jonathan Siddharth said there is currently no easy way to stop it.

Hark Pro, billed as an operating system for your life, is now available on the web, iOS, and Android. The service offers a free trial.

Shelby entered private beta with S3-compatible object storage for AI workloads, agents, and distributed computing. Its design keeps data separate from compute, letting workloads move between machines without tying storage to one execution environment. Filecoin separately highlighted storage for AI-generated reports, images, code, and datasets through its Synapse SDK.

xAI released an experimental TypeScript SDK that gives developers one way to add Grok’s text, voice, image, and video capabilities to applications. The package also includes server-side tools such as real-time X search, web search, code execution, and remote MCP, and is available on npm as @xai-official/sdk.

Mistral AI has released Mistral Large 4, a public preview open-weight multimodal model with 49B active parameters, 1.05T total parameters, a 1M context window, and documented pricing or performance details.

Polars 2.0 has been released, adding default streaming execution, out-of-core spill-to-disk support, SQL performance improvements, and a new Map dtype.

A new paper examines whether AI systems automating model research can create a faster feedback loop than ordinary scaling. The system would improve the code, methods, and experiments used to build its successor, while the authors call for policymakers to track how extensively frontier companies have automated development.

The Arc Institute opened Ptarmigan-1 for public use, giving researchers a way to predict small-molecule binders without first building a 3D protein structure. The model also covers disordered proteins, which have been difficult to drug, and uses data from proteins operating inside living human cells.

AI coding agents exposed more than 13,000 internal screenshots in public GitHub repositories across over 300 organizations, according to Glow Labs. The images included customer billing records and unreleased products, and 93% were stored under employees’ personal usernames rather than company GitHub accounts.

A quoted post says xHigh reached No. 1 on Artificial Analysis’ Cyber Index across several cybersecurity evals. That puts xAI’s model at the top of a benchmark built to poke at reasoning, not just chat politely about it.

AI is moving the automation debate into highly educated work, with Andrew Yang warning that machines can perform billable legal tasks in seconds. Optimists counter that earlier automation transformed agriculture and created new industries, while poorer countries show more enthusiasm for AI's potential in growth, education, health, and prosperity.

It wraps Artificial Analysis data with T3 Code usage stats in one place, plus more benchmarks are planned. Theo is pitching it as a cleaner window into model performance and how people actually use the tools.

The company paired live demos with a batch of product announcements, and most of them were available immediately after the event. That leaves fewer of the usual ‘coming soon’ footnotes and more things people can poke at now.

MediaTek is forecast to retain 100% of Google’s TPU v9 EMIB-T share across training and inference, including the Humufish and Triggerfish designs. The outlook assumes no TSMC backup source or competing 2nm project during the relevant period, while small-volume substrate deliveries are expected by mid-2027 and mass production in the fourth quarter.

The new feature puts always-on agents on their own computer, with a browser and connected apps, instead of just a chat box. It’s included in Pro, so Anthropic is packaging the agentic part as a subscription perk rather than a separate add-on.

Asked who would be accountable if AI agents committed a crime, Trump said the technology had been 'officially' renamed SI, pronounced like 'sigh.' No formal policy or agency behind the change was identified.

Its Agents Window now supports /visualize, which analyzes data and renders charts or diagrams inline.

The new Team Bots are shared AI teammates that learn as a group works with them. Teams can assign each bot skills, plugins, and credentials, then use it in Slack or Grok Bot—essentially a cloud agent with a team account.

Two Nvidia-branded trailers reportedly stolen by thieves expecting AI GPUs contained about 40,000 pounds—roughly 20 tons—of sand instead.

Earendil has shipped Pi 1.0, a hardened minimal extensible agent harness, and introduced Pi Durable as an experimental package.

Turbopuffer is announcing an architecture change to its database, introducing an in-progress 'turbopuffer v3' that makes ANN a secondary index and changes document or index layout, writing, compaction, and query planning.

Earendil or Pi has added MCP support in Pi, making MCP a supported feature of the product.

Livenerf: Has Opus 5.5 been nerfed yet?.

MicroLLM Lab is a browser-based benchmark or demo for trying 7 tiny LLMs in the browser and generating verifiable performance certificates from local runs.