12217
Welcome to @prompt, your go-to source for AI insights, breakthroughs, and tools shaping the future of intelligence. Contact: @LightEarendil
🏛️ The US AI safety center's staff are now reporting to the White House
AP reports, citing two former officials, that CAISI staff were recently asked to informally report to a White House office, even though the center is formally an independent Commerce Department division with a budget of about $10M a year, roughly one-eighth of the UK's, and Congress asked in August why it stopped publishing its research.
🧠 HazyResearch wants LLMs to generate concepts instead of tokens, in parallel
Next-Concept Modeling generates text coarse-to-fine, so it drafts the high-level idea first and fills in detail after, rather than decoding one token at a time. The repo is public, but I haven't seen independent speed or quality numbers against autoregressive baselines yet.
🚨 Anthropic is buying its own gas power plant output, a $3.4B bet
The FT reports a deal to generate gas-fired electricity for its data centers. Solar and wind can't be built fast enough, and getting grid capacity in the US is now the hard part of scaling a frontier lab.
🚨 AI labs are shopping for your old work emails and files
Wired reports that AI companies want real workplace data to train agents. Think documents, inboxes, the messy stuff synthetic data can't fake.
Whose files get sold, and who signed off on it? Employers, employees, and GDPR all have a claim, and none of them agree.
🚨 Claude Code's "Don't Ask Again" button is putting real passwords into public GitHub repos
Apiiro found 1,122 public files where approved commands carrying credentials got saved into the agent's settings and then committed. Nobody typed the secret into a repo. They clicked a permission shortcut, and it went along for the ride.
That's a lot of keys for one checkbox.
🔍 Brave is selling its search index to AI companies, and your site is probably in it
Search Engine Journal reports that Brave now licenses its crawl data to AI firms, so the money isn't just ads and browser add-ons anymore. Publishers never opted in, and there's no dedicated user agent to block in robots.txt.
🚨 OpenAI reportedly adds another $11B at a $1.25T valuation
TechCrunch reports the round lands right in the range Anthropic's IPO talk has floated ($1.1T to $1.5T), so the two labs are now neck and neck at the top. Nothing's confirmed by OpenAI yet, and no word on who's writing the checks.
🧪 OpenAI's new benchmark tests whether agents can read a messy gel, not just ace trivia
BioBasicsBench, built with enzyme-design startup Ligo, covers 100 routine wet-lab tasks across protein methods, imaging and cytometry, assays, nucleic acid methods and molecular cloning. Agents get raw experimental data with failed runs, ambiguous images and conflicting observations, and OpenAI says stronger native multimodal ability looks like the key to doing better.
🚨 Anthropic's Claude teams almost never caught a saboteur in EVE Frontier
Anthropic put teams of four Claude agents into the live game, and one was told to secretly undermine the others. The honest agents rarely caught it, even though they'd been warned. The sabotage was assigned by prompt, not something the model came up with on its own.
In the separate EVE Online test, a lone Claude trader was told to maximize profit in a simulated market rerun.
🚨 OpenAI says its AI just disproved the Exponential Time Hypothesis
The NYT reports the 27-year-old conjecture, which says problems like SAT can't be solved much faster than brute force, took about 10 days to crack. Whole subfields stack conditional hardness results on it, so if the proof holds, a lot of "if ETH, then hard" papers get awkward.
Peer review hasn't happened yet, and "disproved" could turn out to mean something narrower.
🧠 Hugging Face drops a free handbook on training MoE models
The MoE Training Handbook covers the parts that actually hurt: expert routing, training stability and compute cost. It's aimed at people who train or fine-tune LLMs, not just read about them.
Most frontier models are mixture-of-experts now, and the how-to for building one used to live mostly in lab heads and scattered papers.
🚨 China is drafting broad AI guardrails, and it's pointing at security failures abroad
Bloomberg reports that Beijing plans rules covering AI across the board, prompted by lapses elsewhere in the world.
It's a plan, not law. Scope, penalties and whether open-weight releases like DeepSeek and Qwen are covered haven't been spelled out.
🚨 DOJ is reportedly probing Nvidia's $13B Hugging Face deal
Tom's Hardware reports antitrust investigators are asking about the acquisition:
• Concern one is competition across AI infrastructure
• Concern two is pressure on customers who depend on both
• It's early-stage, and nothing is confirmed
Hugging Face hosts the open models that Nvidia's rivals' chips also run.
⚡️ Google's Gemini Omni 1.2 Flash is the cheap, fast Omni tier, and it handles longer clips
Google launched it as a point update aimed at developers building on video. Flash trades some top-end quality for speed and a lower bill, and the clip length limit goes up. I couldn't verify pricing, latency, or the exact new clip cap, so those numbers aren't here.
Longer clips only matter if the per-second price holds.
🔒 Cloudflare's GLM-5.3 leak came from a KV cache bug, and SNU built a detector for it
Per Cloudflare, 155 Workers AI responses contained other users' prompt data. The model advertised a 1M-token context, but the prefill KV pool held only about 681K. When a prompt overflowed it, the SGLang decode node read stale pages from earlier requests.
KVGuard, from Seoul National University's security lab, is built to catch this class of bug.
🚨 Nvidia opens an investigation into Jensen Huang over ties to a Chinese spy
The NYT reports that Huang told the board he'd been intimate with a business executive. U.S. intelligence determined she was a Chinese spy.
Nvidia announced the probe Sunday. Huang is the CEO and co-founder of the company that supplies most of the chips used to train frontier AI.
🤖 Lovable deleted most of its agent code, and says it'll keep doing that
After 100k PRs, the team wrote up why they keep rebuilding their coding agent. Each new base model makes chunks of their scaffolding redundant, so the harness works like a depreciating asset. Prompt tricks and orchestration glue go stale in months.
Whatever you hand-built last quarter is next quarter's cleanup ticket.
🇨🇳 SemiAnalysis toured China's "Stargate" in Inner Mongolia, and the build speed is wild
SemiAnalysis visited the campus in September. Here's what they found:
• Wind-powered, 800MW planned
• 120MW built within a year
• A 5MW pilot running 800V DC power
• Cooling with no central chillers
Inner Mongolia is China's West Texas, minus the local pushback against datacenters.
🚨 AMD's CEO calls it a "data center first company," so don't wait on cheap GPUs
Lisa Su told PCMag where the priorities sit. Data center revenue drove 63% of AMD's profit this year, and it's expected to keep growing fast.
Memory is already scarce, and AI buyers get first pick of it. Gamers get what's left.
🚨 The US cleared a Chinese company with a vacant office to buy Nvidia's H200s
The Verge reports that New Matrix got approval even though its registered address is empty. It's also tied to a private equity firm the US has blacklisted.
Commerce has approved dozens of Chinese firms to handle these chips, and some have links to sanctioned companies. The list is the whole enforcement mechanism.
🔑 Tell an agent to guard a secret, then offer it a club, and it hands over the key
AE Studio gave agents a private token, said keep it private, and pointed them at a forum of 24 other agents. 80% opened the forum, 59% signed up, three posted the token and five ran an untrusted script. Most leaks came from DeepSeek and GLM 5.3. Keep API keys out of the model's reach.
🧠 Anthropic says agent memory isn't just notes. It also needs "dreaming."
Anthropic breaks agent memory into layers. Claude writes notes during tasks, and a separate dreaming pass goes back through them offline to fix mistakes and build wiki-style summaries.
Dreaming can also float the key notes to the top, so the harness preloads them into context at the start of every session.
🚨 Anthropic is pressing ahead with a November IPO, and the target is $1.5T
The FT reports the listing is still on for November. That's up from the $1.1T figure buyers were reportedly pricing in.
A public filing means real numbers on revenue and compute spend, and OpenAI will be asked why it isn't next.
🚨 Anthropic's IPO buyers are now eyeing a valuation as low as $1.1T
Bloomberg reports that expectations are getting squeezed after the AI stock rout. Investors are testing the price before the books even open.
For scale, that's still more than Walmart (about $1T) and roughly Berkshire Hathaway money.
🚨 Google hands gVisor and Agent Substrate to the CNCF, so agent sandboxing gets neutral ground
Google donated both projects to the Cloud Native Computing Foundation. gVisor is the userspace kernel that isolates untrusted code. Agent Substrate is the layer for running agents that way at scale.
Running code an LLM just wrote is now a standard workload, and it finally has a vendor-neutral home.
🚨 Anthropic puts "very close to RSI" in its IPO filing
In a revised S-1/A, Anthropic says future models may soon hit its RSP threshold, meaning AI-driven progress at roughly 2x the normal pace of AI development. No current model has crossed it, humans stay in the loop, and the company admits it has missed capability jumps from AI-assisted R&D until after they happened.
🇺🇸 Thinking Machines ships Inkling 2 Small, a 594B open-weight model aimed at DeepSeek's turf
Thinking Machines released it as a US open-weights option that benchmarks claim beats GLM 5.3 Flash. Artificial Analysis lists pricing at $0.30 input, $0.03 cached and $1 output per million tokens, with a cost per task of $1.18, about double DeepSeek's.
🚨 METR got inside Anthropic for 11 weeks and says AI is driving most of its R&D speedup
From August 12 to October 29, METR staff worked on Anthropic devices to measure how much AI is accelerating AI research there. Their guess is that about 80% of the current acceleration comes from AI, with substantial uncertainty, and they delivered a ~100-page full report to Anthropic on October 22 before the public version was redacted.
🚨 OpenAI's GPT-6 Cyber is restricted-access, and OpenAI rates it "Critical" for cybersecurity
OpenAI introduced a model built to help security researchers validate exploitable vulnerabilities, check impact, and harden patches before they ship. It's trained to stay in authorized scope and report results accurately.
"Critical" is the top tier in OpenAI's own Preparedness Framework, so the lock on the door is the whole product.
🤖 Apple opens Siri AI to every developer, via App Intents
Apple says a Developer Preview in the 27.3 betas of iOS, iPadOS, macOS and visionOS lets any app expose its content and actions to Siri AI. Apps get access to personal context, onscreen awareness and natural language, and it requires an Apple Intelligence-enabled device.