The AI Wire ●

A conjecture Harold Grad made in 1967 — that certain three-dimensional plasma equilibria for magnetic confinement cannot exist without special symmetry — has fallen to AI-assisted work. University of Maryland physicist Matt Landreman published explicit counterexamples on September 22, crediting GPT-6 Astra Pro with discovering the solutions, and says he manually checked every equation.

Landreman's preprint landed one day after a separate team published its own counterexamples with Lean 4 certification, so AI does not get sole priority. A follow-on paper submitted October 5 then gave Landreman's results and the remaining open questions to a team of Google DeepMind Gemini 4 Argon agents, which produced broader solutions that remove an additional symmetry restriction; the researchers checked those derivations independently, and they held up.

The result is mathematics, not a reactor: it supplies exact solutions useful for theory and numerical testing of stellarator designs, and it neither proves better plasma confinement nor changes any engineering timeline. It is, though, another data point that frontier models from two rival labs are now producing checkable research results on the same problem within weeks of each other.

DeepSeek released DeepSeek-V4.1-Flash-1017 on October 17, about five weeks after the previous Flash checkpoint, describing it as a Pareto improvement in performance and cost. API pricing is unchanged and developers reach it through the existing deepseek-flash endpoint; weights are also available for download.

On DeepSeek's own DeepSWE v1.1 coding evaluation the new model scores 77.0%, up from 74.2% for the previous version and above the 74.0% the company lists for Claude Opus 5. DeepSeek says shorter outputs lower the effective cost further, with per-task costs often under $0.20-$0.25 in its figures — company-reported numbers that have not been independently verified.

Alongside the model, DeepSeek opened its search and web-page retrieval tools to developers at $0.75 per 1,000 calls, free for a limited time when connected as a server-side feature to designated development tools. The release keeps pressure on the majors on cost per coding task, even as the headline gains are incremental.

Anthropic announced on October 17, through the ClaudeDevs account, that first-token latency in Claude Code has fallen sharply between August and October. For local sessions the p90 figure dropped from 4.07 seconds to 1.33 seconds, a gain of nearly 3x.

Cloud sessions improved from 16.99 seconds to 6 seconds, Claude's web and Cowork experiences improved about 1.5x from 3.36 to 2.17 seconds, and API first-token time fell from 1.53 to 1.01 seconds. Anthropic attributes the gains to gzip-compressing per-turn conversation payloads and to starting cloud sessions as soon as a user begins typing, rather than after they submit.

No new model is involved; this is an engineering speedup across the Claude lineup. For developers working in rapid ask-and-tweak loops, the cloud-session cut is the most noticeable change, though a 6-second p90 wait remains substantial.

The UK's Competition and Markets Authority proposed on October 16 to designate Microsoft with strategic market status (SMS) across its business software ecosystem, under the Digital Markets, Competition and Consumers Act 2024. The proposal spans five activities — operating systems, productivity software, security software, server operating systems and database software — reaching more than a million UK businesses and public bodies and roughly 20 million monthly users.

SMS is not a finding of wrongdoing, but it would give the CMA power to impose tailored conduct requirements later. Its indicative roadmap names four areas for further evidence: cloud licensing, AI features in productivity software — including whether customers can freely choose AI tools beyond Copilot — communications software such as Teams, and security software bundling. The CMA provisionally points to strong positions in each area, plus switching barriers and bundling advantages rivals struggle to overcome.

A consultation runs until November 13, with a final decision due by February 2027. Microsoft says all its products, including Copilot, are open and interoperable and can be used alongside competing services.

SpaceXAI gave sworn testimony to the New York City Council's Committee on Technology on October 16, at a follow-up AI safety hearing. The Council said it had issued its first subpoena on record against a private actor after seeking the company's testimony voluntarily and then threatening judicial enforcement when SpaceXAI missed an October 5 hearing.

Senior Manager of Global Government Affairs Gabriela Castillo Madrid appeared for the company. Councillors questioned her about a sexualised image Council Speaker Julie Menin said her team had produced using Grok; Castillo Madrid said a safety team could examine it and that the company prohibits sexualising real people.

A former employee alleged at the hearing that only three of SpaceXAI's 200 technical employees worked on AI safety training during his time there. Councillors are considering restrictions on unauthorised AI depictions of elected officials and a financial incentive for whistleblowers.

SpaceXAI announced a bundle of Grok Bot improvements on October 16. Computer tasks now finish 3x sooner on average on the company's own figures, replies arrive in less than half the previous time, and the app gets a refreshed design, better voice calls and notifications that go only to the device in use.

Grok Bot can now read a connected user's X Chat messages and can be set a task by tagging @bot in a post or reply on X, with the work and results staying in the Bot's own chat. Tagging is not yet available for X accounts based in the UK, the European Union, Australia, Iceland, Liechtenstein or Norway.

New connectors include Robinhood, Coinbase, Trello, PostHog, beehiiv and eToro, and teams can now submit their own connectors to the Grok Bot marketplace. Grok Bot's own email address also gains plus-alias support, so routines can run on mail sent to a tagged address.

Replit CEO Amjad Masad said onstage at TechCrunch Disrupt on October 16 that the company is no longer working with Anthropic. He cited cost, partnership and performance, saying OpenAI's Sol and Astra models consistently beat Claude at building vibe-coded apps from scratch on Replit's internal and public benchmarks.

Masad said Replit had been one of Anthropic's top five customers, and that the move followed Replit's August partnership announcement with OpenAI. There had been no public indication before the interview that the Anthropic relationship had ended.

The account is disputed: a person familiar with the relationship told TechCrunch the partnership is still in place, and Replit's own September production notes still showed Claude models in its agent environment. Replit has not clarified the scope of the break.

October 25 update: The Information has now reported the financial shape behind the shift — OpenAI offered more than 30 percent off a contract worth over $100 million a year — and Masad has since clarified that Replit 'remains model agnostic' and that 'OpenAI is now our preferred partner for powering Replit Agent.' Replit's documentation still lists seven Claude models, so the October 16 'no longer working with Anthropic' framing overstates the current state; the relationship has shrunk rather than ended. The wider discount campaign is covered in the October 25 price-war entry.

Reflection has released the weights for Beam, its 501-billion-parameter mixture-of-experts model, on Hugging Face under an Apache 2.0 licence. Baseten offers day-zero support through its Model APIs and dedicated inference.

Beam was first introduced on October 5 while it finished red-teaming and evaluations, with early access by waitlist. The release includes a model card with final benchmarks and a safety technical report, and Reflection frames the launch as an open ecosystem of more than 40 partners and integrations for agents.

A public beta of Reflection's own API at platform.reflection.ai is free to start. Named launch platforms span the major clouds and inference providers, with Dell and HPE offering on-premises deployment for sovereign and air-gapped environments.

OpenAI Codex lead Thibault (Tibo) Sottiaux said on October 16 that Codex and ChatGPT Work now give subscribers more effective usage from the same plan. The changes reuse context across subagents, keep context cached longer, and route requests to raise cache hit rates.

Tibo did not quantify how much extra usage the changes deliver. The post is Day 12 of his 28-day ship-or-reset series, following Day 11's cuts to 'model is at capacity' errors on October 15.

For subscribers, the practical effect is that long, subagent-heavy sessions should burn less of a plan's allowance on repeated context. How large the saving is in practice will only show up in users' own usage meters over the coming days.

OpenAI said on October 16 that GPT-6 Pro can now split a task among multiple agents working in parallel inside ChatGPT, with the agents' progress visible to the user. The feature is available to Pro users in the Chat tab, according to the company's post.

OpenAI also claims GPT-6 Pro responds faster and uses plugins and tools better in this mode. The post gives no latency figures or worked examples to back those claims, and it does not explain how users control or approve the parallel agents' tool actions.

For Pro subscribers this brings the delegation pattern previously associated with the Work tab and Codex into ordinary Chat conversations: one request can fan out into several simultaneous workstreams, with the user watching each one rather than waiting on a single sequential reply.

Late Friday, October 16, OpenAI updated the ChatGPT app on iOS and iPadOS with a home-screen widget for your Dot and new Shortcuts actions to control it. The small widget simply shows your Dot's avatar and tapping it opens the Dot's message thread — or, if you toggle on "Show shortcut button" when editing it, it adds messaging and voice buttons beneath the avatar.

The update also adds a trio of Shortcuts actions to message, dictate to, or call your Dot, each taking two parameters: the name of your Dot and how you want to communicate with it. Users can currently have only one Dot, so a parameter that opens a list of Dots strongly suggests multiple Dots are on the horizon, something OpenAI telegraphed at DevDay, MacStories notes.

The widget gives Dots — OpenAI's always-on agents, launched in September for a limited set of paid plans — a place on the iPhone home screen, shortening the gap between an ongoing task and checking in on it. It adds no new task capability, and users must initiate the voice calls themselves.

In release notes published October 16, OpenAI introduced ChatGPT Slides for creating and editing presentations inside Space. The feature is rolling out to eligible Pro, Business and Enterprise users with Space access on the web and in the desktop app.

Users can start a presentation from a conversation, files, an outline or an example deck, then edit text, images, shapes and charts directly and ask ChatGPT to make changes. Teammates can edit in real time and leave shared comments, and presentations can run directly in ChatGPT.

Web users can export decks to PowerPoint, PDF or Google Slides, while desktop users can open them in PowerPoint or Keynote where those apps are installed. Slides were first previewed as part of the Space and Pages announcement at DevDay on September 29; this is the feature reaching users.

OpenAI announced a round of product updates to ChatGPT Ads late on Friday, October 16. The new visual formats under test include wider ad cards, larger images on desktop and more prominent, full-width images on mobile, running on advertisers' existing campaigns and creative.

Advertisers with active campaigns now receive weekly performance emails comparing results with the previous week, and can report bugs or suggest improvements from the Ads Manager account menu. A redesigned Conversions page pulls together conversion activity, event trends, optimization status and diagnostic warnings.

The update builds on OpenAI's October 5 announcement of a visual ad format during image generation, and adds new measurement partners. Early partner numbers remain just that: DV Rockerbox measured WeightWatchers' cost per acquisition at 15.3% below its paid-search benchmark, and Triple Whale found 93% of Portland Leather's ChatGPT Ads visitors were new.

Microsoft has told partners that from December 1, 2026, new Microsoft 365 Copilot offers can be bought without a Microsoft 365 base subscription, priced the same as the current offers. The change was reported in coverage of October 16 pricing news built from Microsoft's partner communications.

The same coverage says usage-based billing will be on by default for new Copilot Business licenses sold through Microsoft's CSP channel. That shifts new buyers toward paying for agent work as it is used, rather than only per seat.

Until now Copilot has generally required a qualifying base plan underneath it, so unbundling changes the entry cost for organisations that want Copilot seats without the full suite. The December 1 date, not any new feature, is the deadline budget holders need to plan around.

Update, October 19: the usage-based side of that change now has dates and a cap in partner and admin reporting. For new Copilot Business licences bought directly from Microsoft, pay-as-you-go billing is the default from October 19, with the CSP/reseller channel following on December 1; Microsoft creates the Azure billing setup automatically. The preset spending limit is 4,000 Copilot Credits per user per month — at the $0.01 list rate, up to about $40 per person per month on top of the seat — and admins can raise, lower or zero it. Existing licences are not switched over; until now the metered features stayed locked until an admin connected Azure and set a policy themselves.

Microsoft has told Microsoft 365 administrators, in Message Center notice MC1511056 pushed on October 16, that it will retire the Copilot Studio app inside Microsoft Teams by February 2027. Agent creation in the Teams app is already no longer supported, editing agents there stops in January 2027, and the app itself disappears the following month.

Anyone who builds, edits, publishes or manages classic-style chatbot agents from inside Teams will need to move to the Copilot Studio web app instead. Microsoft is advising admins to warn affected users, and to update internal documentation, training, bookmarks and helpdesk scripts that point people at the Teams app.

It is a small, practical change, but a telling one: Microsoft is consolidating agent-building into its dedicated web tooling rather than keeping a second, embedded editor alive inside Teams.