the week in AI, briefly. then briefly again.
Artificial Intelligence — briefly, then briefly again · tbb.ceo
listen tothe daily 0:00 –:––
/daily ·30 SEPT 2026 ·WEDNESDAY ·2 MIN READ ·6 STORIES

OpenAI DevDay; DeepSeek goes Huawei

OpenAI held its developer conference, shipping always-on agents and a cheaper frontier model; policy moved from voluntary dinners to a voluntary accord; and DeepSeek formalised its chip-independence project with Huawei.

01 / The Day

WEDNESDAY 30 SEPT 2026, ranked

06

OpenAI ships Dots agents and GPT-6.1 Sol at DevDay

OpenAI's annual developer conference produced two products: Dots, described as always-on AI agents that proactively complete work inside ChatGPT, and GPT-6.1 Sol, which OpenAI claims delivers near-Astra intelligence at one-fifth the price. A premium Ultrafast speed tier was also announced.

  • Dots agents run persistently and initiate tasks; users receive updates in a chat-like interface inside ChatGPT
  • GPT-6.1 Sol targets the mid-market gap between commodity models and frontier Astra pricing
  • DevDay follows OpenAI cancelling GPT-6.1 Astra last week over alignment concerns — a pointed release calendar
Why it mattersThe two-product DevDay is OpenAI's clearest statement on how it plans to monetise the agent layer — and at what price point.

White House releases voluntary AI accord with tech

The Trump administration published a voluntary accord asking technology companies to partner with external auditors and establish internal controls to monitor AI alignment — the first formal governance framework the administration has put to the industry since taking office.

  • Participation is voluntary; no enforcement mechanism or liability provision is specified
  • Companies asked to set up internal alignment monitoring and external audit partnerships
  • Follows weeks of frontier-lab agent safety incidents — the timing is unlikely to be coincidental
Why it mattersVoluntary or not, a White House governance framework gives companies political cover to invest in safety infrastructure — and sets a baseline others will be judged against.

DeepSeek joins Huawei on open-source CUDA alternative TileLang

DeepSeek announced a partnership with Huawei to develop TileLang — an open-source programming tool for Huawei's Ascend AI chips — formalising the collaboration between China's two leading AI organisations and releasing the toolchain publicly.

  • TileLang is an open-source CUDA alternative targeting Ascend's NPU architecture
  • DeepSeek had previously self-optimised on Ascend without a formal Huawei arrangement
  • A public open-source release widens Ascend's developer base beyond DeepSeek and Huawei's internal teams
Why it mattersIf TileLang achieves adoption, it closes the software-toolchain gap that has kept Ascend chips off-limits for most inference workloads — the last practical CUDA dependency in China's AI stack.

Goldman: $500B in AI-linked financing raised in 2026 so far

Goldman Sachs reported that investors have provided approximately $500 billion in financing to AI-linked groups in 2026 to date, with hyperscaler debt issuance now spreading across euro and Canadian dollar markets — the financing side of the infrastructure build-out Goldman projected at $1.2 trillion by 2027.

  • $500B in a single calendar year is the largest technology capital-raise cycle on record
  • Spread to euro and CAD debt markets signals the investor base is no longer US-only
  • Q4 2026 has not yet been reported; the annual total will exceed this figure
Why it mattersThe financing figure is the most concrete single-year measure of how much capital the AI buildout has absorbed — and it still has a quarter to run.

RFK Jr.: AI offers a medical second opinion better than any doctor

Health and Human Services Secretary Robert F. Kennedy Jr. stated publicly that AI can offer a second opinion much better informed than any doctor in the country, and described the technology as capable of freeing people from what he called medical tyranny — without citing any regulatory or clinical-validation framework.

  • HHS Secretary frames AI as a check on the medical establishment, not a supplementary clinical tool
  • No reference to FDA oversight, clinical validation standards, or liability was included
  • FDA and CMS are actively reviewing AI-assisted diagnostics rules; this statement reaches those rule-writers directly
Why it mattersAn HHS Secretary publicly endorsing AI medical advice without regulatory caveats is a new kind of policy signal — one with direct implications for FDA rulemaking already in progress.

Alibaba ModelScope and MoArk race to become China's Hugging Face

Alibaba's ModelScope, hosting over 170,000 open models, and OSChina's MoArk, with around 20,000, are competing to become the default distribution platform for China's open AI ecosystem — building the infrastructure layer that Hugging Face provides in the West.

  • 170K+ models on ModelScope covers the majority of Chinese-originated open-weight releases
  • A domestic model hub reduces dependence on a US-controlled distribution platform
  • Chinese providers already drive most of Vercel's open-model traffic shift — a hub accelerates adoption
Why it mattersModel distribution infrastructure is as important as model quality; whoever wins this race controls the default channel for China's growing open-model ecosystem.
Be subscriber #014 the weekly ten, every friday by email · no spam · unsubscribe anytime
Past days

The Daily Archive