01
OpenAI is cutting Cursor off because SpaceX bought it
OpenAI is ending Cursor's direct model access on November 12 after SpaceX acquired it - the first time a model provider has cut a top coding tool for who owns it.
What they showed / shipped
- OpenAI announced it is ending its partnership with Cursor following the SpaceX acquisition; under its proposal Cursor's direct access to OpenAI models ends November 12.
- OpenAI's own post names the affected party out loud: "the people most affected by this decision are the developers who rely on OpenAI models" inside Cursor.
- The post is the single most-engaged item in today's pull at 6,256 likes - roughly 2.5x the next lab announcement.
Why it matters
- Your editor's model access is now a function of your editor's cap table. If you have Cursor workflows pinned to GPT models, November 12 is a real migration date, not a rumor.
Sources
02
Claude spent 48 hours aligning other AIs and beat the human researchers
Anthropic gave Claude 48 hours and one GPU to improve the alignment of small models, and it out-performed the human baseline.
What they showed / shipped
Why it matters
- 48 hours, 1 GPU. This is the scale that says a solo person with a rented A100 can now run a real research loop, not just fine-tune.
Sources
03
A federal judge ruled the Pentagon's Anthropic blacklist illegal
A US judge blocked the Pentagon's designation of Anthropic as a supply-chain risk, calling the measures illegal and baseless.
What they showed / shipped
Why it matters
- Federal procurement was the lever being used to pick model winners. A court just said it can't be used that way on ideological grounds.
Sources
04
a16z raised $1.1B for the physical layer under AI
The Machine Age Fund is $1.1 billion aimed at chips, memory, networking, power and the machines that put AI into the physical world - not apps.
What they showed / shipped
- a16z announced the Machine Age Fund: $1.1B for founders "rebuilding what intelligence runs on: chips, memory, networking, systems software, power."
- Ben Horowitz on the thesis: "we have a whole new technology that's the most important technology ever. You need a whole new infrastructure."
- He also called out the oldest startup law as dead - that a two-year lead can't be closed by hiring 1,000 engineers.
- Adjacent money: neocloud Lambda secured $1B in debt purely to buy more chips.
Why it matters
- The smart money's bet is that the app layer is commoditized and the margin moved down the stack to power and silicon.
Sources
05
Grok Bot became shareable and people are handing it their corporate card
Grok Bot shipped template sharing and online purchasing in the same week, and the demos are people giving an agent real spending power.
What they showed / shipped
Why it matters
- Shareable agent templates is the app-store moment for agents. The distribution problem for an agent you build just got solved by someone else.
Sources
06
Perplexity took the top three spots on the search leaderboard
Perplexity's Search API swept the top three places on the Artificial Analysis Search Index at about $0.091 per task.
What they showed / shipped
Why it matters
- $0.091 per search task is the number to hold. If your agent does retrieval, that's the price you're benchmarking your own stack against.
Sources
07
Midjourney's V8.2 edit model is in testing and already patched
V8.2 brings instruction editing, up to 4 reference images, inpainting and outpainting - and Midjourney shipped a quality fix within a day.
What they showed / shipped
Why it matters
- Four reference images plus instruction editing is the feature set that makes character consistency across a series actually tractable.
Sources
08
An open-source game got pulled from Google Play by an AI copyright bot
Luanti was removed from Google Play over a DMCA notice generated by an AI tracer with no valid claim behind it.
What they showed / shipped
Why it matters
- An automated claim system with no human in the loop can delist your app. There is no appeal fast enough to matter for a launch week.
Sources
09
Harvard and MIT simulated the world with 8.3 billion AI personas
Researchers built one AI persona per human on Earth to run population-scale social simulations.
What they showed / shipped
Why it matters
- Population-scale persona simulation is a testing surface. You could run a product decision against a demographic slice before shipping it.
Sources
10
The local-AI corner shipped a microcontroller image model and a frontier TTS
Someone ran an image generator on a $1 microcontroller, Breeze-TTS-2 hit frontier quality locally, and ROCm 10.0 landed for agentic workloads.
What they showed / shipped
- A latent flow transformer running on an RP2350 microcontroller generates 128x128 face images.
- Breeze-TTS-2 is drawing "genuinely frontier" first impressions from the local crowd.
- ROCm 10.0 shipped, explicitly positioned for agentic AI - AMD's bid for the local inference stack.
- StemDeck is a free, open-source, fully local AI stem separator (34 points on HN).
Why it matters
- An RP2350 costs about a dollar. The floor for "device that can generate" just fell through.
Sources
11
South Korea is giving every citizen free AI access
South Korea's 'AI for All' program hands free AI access to every citizen - national distribution as industrial policy.
What they showed / shipped
Why it matters
- A country-wide free tier creates a 50-million-person market with zero acquisition cost for whoever builds on top of it.
Sources
12
Open-weight labs became the hottest acquisition target
TechCrunch says open-weight AI companies are now the Valley's most-wanted acquisitions - the open model as an asset class.
What they showed / shipped
Why it matters
- If the open-weight lab you depend on gets bought, the weights you already have stay - but the next release might not exist. Pin versions you rely on.
Sources
