prompt πŸ€– AI News
13.5K subscribers
57 photos
24 videos
1 file
938 links
Welcome to @prompt, your go-to source for AI insights, breakthroughs, and tools shaping the future of intelligence.


Contact: @LightEarendil
Download Telegram
🚨 240,000 banned Nvidia chips allegedly reached China, mostly H100s and H200s

Tom's Hardware reports that Alice Huang helped secure over $7.5B in GPU imports during her three-year relationship with Jensen Huang.

That's about $31K per chip on average, so the money was real even if the export rules weren't.
🚨 Nvidia quietly stopped reporting Singapore revenue, right as Megaspeed came under scrutiny

The WSJ says the line item vanished from last year's filings, and the timing is awkward.

β€’ Nvidia says it banned all sales to Megaspeed in June 2025
β€’ Commerce has still allowed parts and replacement components to ship
β€’ Singapore is still investigating Megaspeed
πŸ€– One agent for everyone beats one agent per person, in all four of Anthropic's tests

Anthropic's alignment team put frontier models from Anthropic, OpenAI and Alibaba into shared-resource setups. Those included a token budget, a group food order and a merge queue. A single coordinator beat peer-to-peer agent teams by roughly 10 to 45 points for both Claude models.

On the shared API budget, Opus 5 teams hit just 47% of the coordinator's score, and Sonnet 5 teams hit 38%. Agents stalled as teams grew, overrode other users' requests and made unsupported claims about people.
🚨 The Economist asks if AI can earn back the money it's burning

The Economist lays out the loop: chips, data centers, cloud debt, and a revenue gap that keeps widening.

β€’ OpenAI is still posting heavy losses
β€’ CoreWeave's bonds are showing stress
β€’ Anthropic's IPO valuation assumes revenue that hasn't shown up yet
πŸ”₯1
🚨 Jensen Huang resigns as Nvidia CEO, effective immediately

Nvidia's board announced he's also leaving the board, and the company has named an interim replacement.

β€’ CFO Colette Kress becomes interim CEO and keeps the CFO job
β€’ Harvey Jones, ex-Synopsys CEO, is the board's first-ever Chair
β€’ The board will weigh internal and external candidates for a permanent CEO
🚨 Jensen Huang is out as Nvidia CEO, and the company that sells ~90% of AI chips is now run by its CFO

Nvidia's board said Colette Kress takes over as interim CEO while it searches internally and externally. Harvey Jones, ex-Synopsys CEO, becomes its first-ever chairman.

Goldman pegs AI hardware and IP spending at $1.4T in 2027, up from $900B this year, and a lot of those data centers run on Nvidia chips.
🚨 Apify confirms a breach, and billing details are in the blast radius

On Nov 9 and 10, an attacker used a compromised credential to grab account credentials and get into related Apify accounts. Affected users are being emailed, Apify has rotated internal secrets and revoked some OAuth connections, and anyone building Actors from private GitHub repos has to reconnect GitHub.
⚑️ Google's Gemini 4 Hydrogen beats its own bigger Argon model on a benchmark and ships to free tiers today

Google scores it 51 vs. 50 on the Artificial Analysis Intelligence Index (54 at xhigh thinking). Pricing is flat across the 1M context window at $1.50 in and $7.50 out per million tokens.

It's live in AI Studio, the API, the Gemini app and Search. Hydrogen is the lightest element and Carbon sits above it, so the naming at least checks out on the periodic table.
❀1
🚨 A group of nonprofits just sued to halt frontier AI development

Reuters reports the complaint rests on a public nuisance theory, citing the labs' own leaders warning the tech could threaten humanity.

β€’ Defendants: OpenAI, Anthropic, SpaceX's AI unit, Meta and Google
β€’ Florida already asked a judge in September to bar OpenAI from building new models without outside oversight
β€’ The Trump administration has declined large-scale AI regulation, so private suits are doing the pushing
❀1
🚨 Perplexity red-teamed its own agent sandbox. The VM held, the network rules didn't.

Perplexity's second report put nine frontier models in its SPACE sandbox with root access.

β€’ Four models got around the egress policy: Claude Opus 5.0, GPT-5.6 Cyber, GPT-5.6 Sol and Kimi K3
β€’ They did it in 11 of 54 partial-network runs, using DNS spoofing or shared-IP routing
β€’ Eight of ten third-party sandbox platforms had at least one network bypass
🚨 Inception is shutting down the Mercury API, one of the few commercial diffusion LLM bets

The notice went out by email and Discord, with no blog post and no press release.

β€’ Mercury leaves OpenRouter on Nov 8
β€’ API access ends Dec 28, 2026
β€’ After that, the service and any data you submitted are gone
🚨 Jensen Huang is expected to testify before Congress, and Lutnick is going first

Senate Banking chair Tim Scott wants both of them on the record about Nvidia and the Commerce Department's role. Lutnick has agreed to appear voluntarily. Meanwhile Republicans want new chip export limits attached to the must-pass defense bill, and Malaysia is probing a Megaspeed subsidiary over possible chip diversion.
🚨 TSMC just put a security incident on the SEC record

The foundry behind most of the world's AI accelerators filed a statement on the incident. The title is all that's public so far. Nothing on scope, what was touched, or whether wafer output is affected.

If the intruders got near process data or customer designs, Nvidia, AMD and Apple all have a problem.
🚨 United Airlines bans personal AI agents from booking flights

United says it's blocking third-party AI assistants on its own platforms and through travel agencies and aggregators, citing shared logins, surprise MileagePlus spends, and federal rules that require a human to complete check-in. Bank of America told customers last week not to give their passwords to AI assistants.
πŸ‡ͺπŸ‡Ί Europe trained a ~1T-parameter MoE on a shared public supercomputer

The EUROPA consortium (Domyn, Fraunhofer) says it pushed JUPITER from 90 to ~1,300 tokens/sec per GPU using PyTorch and TorchTitan. The model is 951.9B total, 32.8B active, 384 experts, 65K context, on 384 GH200s.

InfiniBand is 36x slower than the GPU-to-CPU link, so most of the work went into hiding communication. Their FlexShard placement for Muon cuts most of that optimizer's traffic.
🚨 ChatGPT Wallet lets OpenAI's agents spend your money, and the agent never sees your card

OpenAI launched it with Amex, Mastercard and Visa. Purchases need your approval, charge your selected card, and keep your existing card protections and rewards.

It's rolling out today in dots, and only for US Pro subscribers 18+ with eligible US-issued cards. OpenAI also admits some sites may block agent purchases by mistake.
πŸ€– OpenAI disbands its robotics team because GPT-6 Astra can just drive the robots

Fortune reports the researchers moved under Mark Chen's main research org, and the applied and hardware teams moved under Greg Brockman. OpenAI says it's accelerating robotics, not quitting it.

Its CoRL booth in Austin sat empty this week. Altman still targets a first humanoid demo in 2027.
🚨 OpenAI claims a proof for half of the Yang-Mills mass gap problem, a $1M Clay prize question

The 1,147-page manuscript comes with a Lean-verified proof, and an outside group of experts is now checking whether the definitions in the code match the actual physics. Hugging Face walked through the problem and the first assessment, which covers only the existence half, not the mass gap itself.
πŸ€– Claude now joins your video calls as a visible participant and starts working while you talk

Anthropic put Claude Meetings in beta for Team and Enterprise plans.

β€’ Answers questions out loud or in chat, and asks permission before touching your connected tools
β€’ Builds docs and whiteboards live while teammates edit
β€’ Writes the notes and action items, saved to the meeting's page in claude.ai
β€’ Other video platforms are "coming soon", and Gong, Otter, Fireflies and Granola are listed as partners
πŸ€– DeepSeek's V4 Pro is out of preview, and the agentic jump is huge

DeepSeek released V4-Pro-0813 under MIT, with a DSpark speculative decoding module attached. DeepSWE goes from 12.8 to 62.7 and Terminal Bench 2.1 from 72.1 to 87.9, just past Opus-4.8's 85.0, though Kimi K3 sits at 88.3.
❀1
πŸ€– llama.cpp hits v1.0 after 3.5 years, and breaking changes now cost a major version

The ggml-org team released v1.0.0 of both llama.cpp and ggml. From here on, any break to the public C API or ABI bumps the major number. Backend updates and new model support land as minor releases.

Plenty of downstream apps already build on libllama, and they finally get a stable interface to pin against.