Description

Provenance and trust in the inference stack dominated the day: Microsoft was caught stamping server-issued GUIDs into locally generated images, and a widely-read essay argued the inference engine itself is the unguarded attack surface between a model and its host. On the silicon and capital side, Xiaomi's 3nm Xring O3 matched Apple single-threaded while TSMC's CoWoS bottleneck kept handing Intel Foundry an opening — all of it financed by an AI build-out that analysts now describe as too big to fail.

1. MS Paint and Photos invisibly watermark even locally generated output with a GUID —
2. Xiaomi's Xring O3 matches Apple single-threaded, and a die shot maps the blocks —
3. LLMs could control their host machines by exploiting inference engines —
4. TSMC's HBM-packaging yield issues hand Intel a foundry opening, analysts say —
5. What's new in Emacs 31.1 —
6. The debt-fueled AI build-out may already be too big to fail —
7. Neocloud Lambda in talks to raise up to $3B in a pre-IPO round —
8. Thomson 1.0: continual learning turns an open-weight model into a frontier one for SovereignAI —
9. JPMorgan: US manufacturing will turn to humanoids at $10/hour against $30/hour labor —
10. Bartholomew: a 2.82B "vintage" LLM trained only on English written before 1931 —
11. Apple reverses course: iCloud+ Hide My Email stays on icloud.com —
12. How Europe's packaging rules are squeezing makers and micro-entrepreneurs —
13. Walgit: a Git server that is one binary in front of an object store —
14. Jabber/XMPP: 25 years of digital independence —
15. The entire city of San Francisco as a browser video game —
16. llm-anthropic 0.27 lands the httpx2 migration —