Jensen Huang resigns as Nvidia CEO effective immediately; CFO Colette Kress takes over as interim chief
Nvidia announced on November 12 that Jensen Huang resigned as president, CEO and board member effective immediately, with CFO Colette Kress interim CEO and earnings still set for November 17.
A forensic report says six Chinese frontier models started life on export-controlled Blackwell GPUs
A single-author report published November 12 reads the random starting weights of DeepSeek, Moonshot, MiniMax and Ant models and says six match the layout of Nvidia Blackwell chips, evidence that supports but does not prove Blackwell training.
Gemini 4 Hydrogen ties the leaders on one index, but its token bill and a changing Terminal-Bench version complicate Google's speed pitch
Google released Gemini 4 Hydrogen on November 12 at $1.50 and $7.50 per million tokens; Artificial Analysis scores it 54, near the top, while calling it very verbose and slower than average per task.
Edge0-145B's '29 GB' laptop headline holds only for GPU memory; the card measures about 61 GiB in total and names Qwen as the base
Edge0 released a 145-billion-parameter open model on November 12 with a 29 GB peak-memory headline, but its own model card measures 23.25 GiB of VRAM plus 38.27 GiB of host RAM, and its release file identifies Qwen3.8-Flash-Next as the base.
Microsoft 365's 18-hour outage knocked out Claude's and ChatGPT's connectors while the models themselves stayed up
Microsoft's outage beginning November 12 hit Exchange, Teams, SharePoint and Copilot for at least 18 hours, and Claude's Microsoft 365 connector failed for about five hours in Anthropic's feed with no resolution entry.
'Leaderboard Hacking' paper: votes, prompts and decoy models can lift an arena ranking by up to 45 places
A November 10 preprint by Fleisig, Baumann, Hadfield-Menell and Hovy shows that arena-style leaderboards built on the Bradley-Terry model can be manipulated, with each attack moving a model 15 to 20 places and all three combined up to 45.
Subliminal learning crosses into vision-language models: a student fine-tuned on a teacher's numbers inherits its animal preference
A WACV 2027 paper finds that students fine-tuned only on number outputs from a biased vision-language teacher pick that teacher's favorite animal more often, by up to about 10 percentage points.
llama.cpp reaches version 1.0.0: FP8 KV cache, a host-memory expert cache and server hardening
llama.cpp tagged v1.0.0 on November 13 after a three-file version-bump pull request, with a changelog that includes FP8 KV-cache support, a host-memory MoE expert cache and SSRF and DNS-rebinding fixes.
A voltage-glitching hack that blows GPU fuses turns an 8 GB mining card into an 80 GB A100-class card
A researcher who goes by Kate published a fault-injection attack on the GA100 die that blows arbitrary fuses, converting Nvidia CMP 170HX mining cards into A100-80G equivalents; the early software-only unlock is a separate, earlier hack.
Brave's Search API now accepts x402 payments: an agent pays $0.005 per request in USDC and needs no account or key
Brave announced on November 12 that its Search API accepts x402 agentic payments, where a signed USDC payment on Base is the credential, priced at $0.005 per request.