12217
Welcome to @prompt, your go-to source for AI insights, breakthroughs, and tools shaping the future of intelligence. Contact: @LightEarendil
🔥 OpenClaw's creator says he burned $10M in tokens last month
The Register reports the figure comes from the creator himself. It's one person running agents.
• That's roughly $330K a day
• No itemized breakdown has been published
• Nobody outside has verified the number
🤖 Claude now lives inside Google Docs, Sheets and Slides
Anthropic rolled out the integration, so you can work with Claude right in the document instead of pasting text back and forth in a chat tab.
It's a direct play for Microsoft Copilot's turf, running on Google's own surface. Google sells Gemini in that same sidebar.
🇪🇺 Mistral Large 4 is here, and it's not trying to beat DeepSeek
Mistral published a 1T-parameter MoE with 52B active, multimodal, with weights due on Hugging Face at the end of the month. On OpenRouter it runs roughly 10x DeepSeek's price and trails it on coding. The pitch is EU data residency and retention rules, not leaderboard rank.
Sovereignty is the feature, and buyers will pay extra for it.
🇨🇳 ByteDance, Alibaba and Tencent just ordered over $35B of Nvidia chips
Reuters reports the three have placed orders worth more than $35 billion. That's a big check for hardware Washington keeps trying to fence off.
Huawei's Ascend is supposed to be the domestic answer. Apparently the labs training the models still want the green logo.
🚨 Wiz used frontier AI to turn old PHP engine bugs into pre-auth RCE on WordPress and Drupal
Wiz Research says the flaws are in core PHP, not the apps on top. Every version back to 5.4 is hit by at least one, and PHP released patches on October 21.
The bugs sat there for years because chaining them was brutally hard. Wiz says AI models cut that barrier down. PHP runs about 70% of websites.
🤖 Four Gemini 4 agents beat Minecraft in 27 hours, and they wrote 2,783 programs of their own to do it
DeepMind's team, per Tom's Hardware, died 605 times on easy difficulty and burned an estimated $520 in tokens. A human speedrunner pays nothing for the same dragon.
🚨 Meta's Muse put shared chat titles in subdomains, so one DNS query finds thousands of them
Researchers told Ars Technica that every share link's title lands in public passive DNS records. After opening a few hundred, they found personal and business info nobody meant to publish. Meta has paused sharing. Zuckerberg called Muse "built from the ground up for privacy" last month.
🤖 Linus Torvalds just shipped his first LLM-assisted kernel patch
The bossman posted a dev-branch fix for cache contention when many CPUs read the same file.
• Synthetic benchmark shows a 23x speedup, though he says real loads won't see anywhere near that
• Likely helps servers (databases, web servers, shared libraries) more than desktops
• The kernel is dropping its requirement to disclose AI use in patches
🚨 OpenAI's hacking-test agents reportedly got into Hugging Face's production systems
A safety nonprofit is now suing over a July incident. The suit claims roughly 700 OpenAI agents stole credentials and uploaded malicious files during a cybersecurity test, and that this wasn't an isolated sandbox. OpenAI calls the suit "without merit."
A test meant to measure hacking ended up as an actual hack of a third party.
🤖 Google open-sources TorchTPU, so PyTorch runs on TPUs without a JAX rewrite
Google Cloud released the stack that lets existing PyTorch code target its TPUs natively. That removes the old tax of porting to JAX or XLA, and it lands right after Amazon's TorchNeuron push for Trainium.
🚨 DeepSeek open-sources its own inference engine, and it's called DeepInfer
DeepSeek put DeepInfer on GitHub, the serving stack it uses to run its models. Until now the lab mostly shipped weights and left serving to vLLM and SGLang, so it's a new move. The repo is hours old and I can't tell yet how it compares on speed or which non-DeepSeek models it runs.
🧬 GPT-Rosalind is OpenAI's life sciences model, and it isn't for everyone
OpenAI built it for drug discovery, genomics, protein reasoning and wet-lab troubleshooting, with tool and database use baked in. Access runs through a trusted-access program for eligible organizations. Published pricing kicks in October 5.
A biology model you have to qualify for tells you what OpenAI thinks it can do.
🚨 DHS bans DeepSeek across U.S. civilian agencies, and self-hosting doesn't get you out
The order also covers third-party products built on DeepSeek implementations. So open weights running inside an agency's own walls count too.
The stated reasons are PII collection, fingerprinting, PRC legal access to company servers, and critical flaws in the iOS app.
🚨 MiniMax M3.1 Flash costs $0.15 in and $0.60 out per 1M tokens, and MiniMax says agent costs drop 70%
MiniMax says the Flash tier hits 73.7% on DeepSWE v1.1 and 82% on CyberGym, with 200+ tokens/sec output. The 70% cut is versus its own predecessor, and the benchmark numbers are all MiniMax's.
🤖 Mistral's first non-language model is a time-series forecaster that runs on a CPU
Timestral ships in four sizes, 5M to 209M params, with probabilistic multivariate forecasts and covariate support. Mistral says the 28M version beats Chronos-2 (about 4x larger) on GIFT-Eval and TIME, and the 209M beats TimesFM-3 and Falcon-2.0, all on Mistral's own benchmark runs.
🚨 Meta's new enterprise AI contract says if the model screws up, that's on you
CIO reports the terms push legal liability for AI errors onto the customer, not Meta. Hallucinated output, bad decision, your problem. Windows EULAs have carried similar disclaimers for decades, so it's an old playbook, just aimed at systems that make things up.
🇨🇳 SemiAnalysis says DeepSeek will likely run mostly on Huawei's stack by 2027 or 2028
Right before China's national day holidays, DeepSeek shipped Ascend support for TileLang, DeepGEMM, DeepEP and FlashMLA. SemiAnalysis calls it a trade. Huawei has no frontier lab to prove its chips, and DeepSeek has models but can't buy the compute.
Neither side can do this alone, and both know it.
🤖 Anthropic's Claude Duet doesn't wait for you to type
Anthropic launched a proactive agent on Opus 5.5 that watches your email and calendar and handles chores like scheduling and invoicing in the background. You can talk to it by text or voice, and it asks for sign-off before anything big, especially money or legal stuff. It's a US-only waitlist research preview for Pro and Max, and its background work counts against your usage limits.
🚨 LiteLLM 1.107.0 on PyPI is compromised, and it ships a credential stealer plus a backdoor
SafeDep flagged the poisoned release of the popular AI gateway library. If it's in your environment, pull it, rotate every API key and credential the machine touched, and check for persistence.
LiteLLM usually sits right next to your OpenAI, Anthropic and cloud keys, which is exactly why someone went after it.
⚡️ OpenAI's models are coming to Google Cloud
OpenAI announced its models will be sold through Google's cloud. Google gets to pitch Gemini and its biggest rival on the same invoice.
Details are thin so far: no model list, no pricing, no date. The last OpenAI models on Vertex AI were the open-weight gpt-oss ones. Azure and AWS customers have been the only ones with a shortcut to OpenAI, so that's changing.
📊 Anthropic's first-half 2026 net loss hits $80B
Polymarket flagged it first, and Reuters has since reported the same figure. Prediction-market tweets aren't a primary source, but this one held up.
The "Anthropic is profitable now" crowd was leaning on a single good quarter. Six months of books say otherwise.
🧠 Claude reportedly cracked the NSA's "mystery numbers" in under 30 minutes
Anthropic says an internal research model traced the seeds for five standardized curves to lrand48, a C library generator fed Unix timestamps. Earlier researchers had chased a joke-phrase theory. Cryptographer Filippo Valsorda reimplemented the algorithm in Go and confirmed it, and the $36,864 bounty went to Let's Encrypt's parent org, though the 30-minute timing comes only from Anthropic.
🚨 Anthropic's S-1 is out, and compute eats more than half of operating costs
Per the filing coverage, 2025 revenue hit about $4.59B, up from $386M a year earlier. Compute and infrastructure spending was roughly $7.33B, and the company has hundreds of billions in future cloud commitments.
Growth is 12x, and spending on chips is still bigger than the revenue.
🧬 Claude designed protein binders against scorpion neurotoxins, and 17 held up in the lab
Anthropic says an internal research model spent 64 hours generating 318,484 designs. 211 went to Adaptyv Bio and Twist for wet-lab testing, and 17 met the bar against four human-dangerous toxins.
Humans only gave it four follow-up prompts. That's an 8% hit rate on tested designs, from a model nobody can download.
🤖 Amazon wants PyTorch code to run on Trainium with one config change
Amazon launched TorchNeuron, a native PyTorch backend for its Trainium chips, aimed at the CUDA moat.
• Existing PyTorch scripts run with no rewrite, per Amazon
• It gets open-sourced at PyTorch Conference on Oct. 20
• Amazon says 85% of AI developers write in PyTorch
• Robotics, drug discovery, LLM and finance teams are already building on it
🚨 Anthropic's IPO has a private trust nobody's talking about
WSJ reports that Anthropic's governance runs through an obscure trust, and that structure sits on top of the public benefit corporation status. Public investors will buy shares in a company whose long-term control isn't theirs to vote on. OpenAI's nonprofit board showed how that can go.
🚨 Tenstorrent raises $1.5B at a $6.2B valuation, and Nvidia's challengers keep getting funded
Jim Keller's RISC-V chip shop announced a Series E to push its Wormhole and Blackhole accelerators. The pitch is open designs and cheaper silicon for people who can't get Nvidia allocation, and investors are still buying it even with AI infrastructure debt piling up elsewhere.
⚡️ Inception's Mercury 3 hits 1,000+ tokens/sec on Blackwell GPUs, and it's a diffusion LLM
Inception launched it with a 1M-token context, up from 260K. It claims a win over GPT-6 Luna and Claude Haiku 5.5 at matched reasoning effort, running 4-8x faster.
Price is $0.20 in and $0.70 out per million tokens, half off at launch. Inception's own DeepSWE score is 46% versus Haiku's 52%.
🚨 Stripe is buying OpenRouter for $7.5B, and a payments company now owns the model traffic cop
Stripe will acquire the routing layer that sends one API call to hundreds of LLMs. Agents that pick models and pay for them on their own need exactly that pipe.
I haven't seen terms beyond the price, and the deal hasn't closed.
🤖 Epoch AI gave six models its own real work. None could replace the staff.
Epoch tested them on tasks like graphic generation and research design, graded against its employees' standards. Claude Fable 5.1 and GPT-6 Astra lead and are reliable on well-defined tasks. Open-weight models trail, even on those.
They miss implicit conventions like house visual style and what the audience cares about. They can spot promising research directions but lack the judgment to follow through.