Reasoning hit 67.7%.

Anthropic reportedly struggles to secure cloud compute as providers turn competitive

Cloud providers reportedly increasingly view Anthropic as a competitor while its infrastructure spending balloons ahead of a planned November IPO.

Meta removes the word train from its commercial terms after AI backlash

Meta dropped the word train from commercial terms of service taking effect October 30 after legal experts warned the terms could allow scraping of linked websites. A Meta spokesperson confirmed users were not alerted to the changed language.

Google brings Personal Intelligence to Gemini in Android Auto and Google built-in cars

Google is rolling out Personal Intelligence, which draws on data such as Gmail and Calendar, to Android Auto starting today. Vehicles with Google built-in will get it over the next few months. The feature remains optional.

Simon Willison describes Battle Drones, a vibe coded RTS built with Claude

Willison built the browser RTS mostly with coding agents, mainly Claude Code, over 12 iterations and 321 commits. It includes audio-based multiplayer for two devices in microphone range, and he says he reviewed almost none of the code.

Elevenlabs releases two Eleven Realtime speech-to-speech models

The new Eleven Realtime speech-to-speech models target real-time AI conversation. The v1 Max variant scores 98.8 percent on Big Bench Audio with a 0.57-second response time, according to Elevenlabs, which claims no model on the Artificial Analysis leaderboard is both faster and more accurate.

Hugging Face model Spark-X2.6-4B, a compact on-device language model, is released

Spark-X2.6-4B keeps the Spark-X2.5-4B architecture and gains from improved post-training, with three thinking levels (none, low, high). The developer claims it matches or beats models around 9B parameters on open-source benchmarks.

Google releases a PyTorch developer tool to move apps off Nvidia GPUs

Google follows Amazon with a developer tool that helps PyTorch users shift their apps off Nvidia GPUs.

OpenAI lets ChatGPT users lock individual chats

The lock is meant for conversations users want kept private when they hand their device to someone else to show them a different chat.

NVIDIA publishes Cosmos3-H-Dreams for generative surgical simulation

Eric Schmidt reportedly holds about half the Anthropic stake of CEO Dario Amodei

Bloomberg reports that former Google CEO Eric Schmidt owns roughly a 1% stake in Anthropic, about half of what Dario Amodei holds.

Meta ordered to name executives who approved torrenting books for AI training

A magistrate judge in copyright suits by book authors and Entrepreneur Media ordered Meta to identify decision-makers behind downloading books from shadow libraries like Anna's Archive and LibGen, per TorrentFreak as cited by Ars Technica.

Meta's Muse AI companion passes 10 million U.S. downloads in 40 days

Muse reached more than 10 million U.S. downloads in its first 40 days, a milestone reached by only 11 other apps, including Threads. New data shows it retains users at higher rates than rival AI apps.

Altman reportedly pitched three endgames for AI safety in internal presentation

According to a report, OpenAI CEO Sam Altman outlined an international IAEA-style watchdog, a merger of major labs, or a single dominant firm, which he called terrible.

Hugging Face publishes FairStress, a deeper evaluation of fairness in LLM decision-making

Gradium releases Phonon, open-source on-device text-to-speech models

Gradium released Phonon, two small TTS models of 31M and 40M parameters, with a fully open-source inference stack targeting CPUs, GPUs and NPUs across 10+ backends, mostly written in Rust.

Mistral launches Voxtral 3 speech models and a voice-agent SDK

Transcription costs $0.0035 to $0.008 per minute and text to speech $13 per million characters. Mistral has not said whether weights are out.

Intel Arc Pro B65 four-GPU rig posts long-context throughput sweeps for Qwen3.8 models

8 Flash-Next and 27B models on 1x to 4x Intel B65 setups with vLLM, reporting output throughput across context lengths up to 200K. Results include queueing effects and are not model-quality scores.

Hugging Face blog shows learned memory tops the τ²-bench customer service leaderboard without a frontier model

IBM Research's ALTK-Evolve reportedly ranks first on τ²-bench customer service tasks using learned memory rather than a frontier model.

Pocket TTS with voice cloning shrinks to 1.9 million parameters

A community post says the Pocket TTS text-to-speech model, which supports voice cloning, is now only 1.9 million parameters.

Six-way benchmark compares Qwen3.8-Flash-Next locally against Claude Opus 5.5 on a Rust feature

A community test of 12 PRs on a LlamaStash feature found Opus 5.5 averaged 82% against 68% for Claude Code with Flash-Next and 66% for Pi with Flash-Next. Flash-Next ran locally at about 4.16 bpw, so the author paid only electricity, and results are a single developer's unverified experiment.

Ornith-1.5-122B, a Qwen-based finetune, released in GGUF and NVFP4 formats

Ornith AI introduces Ornith-1.5, which continuously generates its own training tasks and improves through reinforcement learning, extending Ornith-1.0 built on Qwen3.5 and Gemma4. GGUF and NVFP4 builds are on Hugging Face, with more details in the company blog.

Spotify says it removed 50 million AI slop and spam tracks from recommendations

Spotify says it has removed more than 50 million songs considered AI-generated slop or spam from user recommendations over the past year and will highlight more human-made music. It has not mandated labels on all AI tracks but launched an "AI Persona" badge.

Paper introduces COUNT, an n-gram method that scales language models on host memory

COUNT stores precomputed n-gram counts on host RAM instead of training learned embeddings, and reportedly matches a model with 57% more compute for 1B-parameter pretraining. The results come from an arXiv paper shared on Reddit and are not independently verified.

Adobe adds one-tap Open Eyes feature to Lightroom mobile

The generative AI Open Eyes Quick Action appears automatically when the app detects closed eyes in a photo, on iOS, iPadOS and Android. The feature is available in the mobile version of Lightroom.

Eka Robotics valued at $1.45 billion on physical AI robotic arm model

Eka Robotics is reported to be worth $1.45 billion, with experts praising its robotic arm for picking up a wide range of objects smoothly and human-like.

Meta suspends Muse sharing feature after researchers expose shared conversations

Security firm Claroc enumerated 6,800 shared Muse conversation links and found sensitive personal and business data in a few hundred of them. Meta says it is suspending sharing while it reviews the design and considers changes.

Army Cyber Command develops ECHOTRIBBLE, a worm-like AI agent program

According to internal documents, the program is designed to spread across networks, including air-gapped systems, and neutralize targets with minimal human oversight. Experts warn of collateral damage.

Meta prepares Muse Confidential VM for private AI chats

Meta is reportedly preparing Muse Confidential VM, a mode that would keep chats private even from Meta, with possible feature limits and early signs of EU web access.

Cloudflare releases Clairic, a model that extracts data from documents

Cloudflare says Clairic pulls data from documents faster and more cheaply than large language models.

US bans DeepSeek technology across federal civilian agencies

The US Department of Homeland Security has barred federal civilian agencies from using AI technology from Chinese lab DeepSeek.

Community post shares a Qwen prompt that makes the model draft and refine iteratively

A Reddit user reports that adding a fractal draft-test-refine prompt to Qwen 3.8 Flash Next made it faster and more decisive in their own projects. The claims are anecdotal and no benchmarks are given.

Otter.ai raises $320 million from General Catalyst's Customer Value Fund

The meeting note-taking startup says the funding comes from General Catalyst's CVF, which provides capital without diluting equity and earns returns based on marketing spend results. Otter.ai plans to use it for enterprise growth, go-to-market, international expansion and product innovation.

Vals AI says DeepSeek V4.1 Flash 1017 beats GPT-6 Sol on a multi-step offensive cybersecurity task

Vals AI reports that DeepSeek V4.1 Flash 1017 outperformed GPT-6 Sol on its multi-step offensive cybersecurity task. Sol was tested through OpenAI's Daybreak trusted access program.

OpenAI prepares GPT-6.1 Terra, with a cheaper Terra model reportedly in the works

GPT-6.1 Terra appeared in a Microsoft pull request, and The Information reports OpenAI is readying a cheaper Terra model. Neither report includes an official launch date.

ISTA-DASLab releases a Qwen3.8-27B prefiller model that streams from SSD for low-quant accuracy

The demo recovers accuracy for low-bit quantized Qwen3.8-27B while keeping VRAM usage low and decoding fast. It reportedly needs an Nvidia Blackwell GPU, and the poster has not tested it. For context longer than about 4k tokens, the poster expects no speed penalty.

Reddit user asks how to annotate 500 images for fine-tuning Meta DETR

A community post seeks advice on annotation tools, using Grounding DINO or OWLv2 for pre-annotation, and quality control for a small object-detection dataset. It contains no new model news.

OpenAI publishes case study on franchisee cutting cash reconciliation time with ChatGPT Work

An independent Schlotzsky's and Cinnabon franchisee reportedly cut menu repricing time by 95% using ChatGPT Work and built a workflow that could free up $100K in cash flow.

DeepSeek open-sources DeepInfer, the serving stack it runs in production

It ships under Apache 2.0 and for now supports only DeepSeek-V4.1-Flash in offline mode. It leads Hacker News with 1,862 points.

Berth open-source tool lets a single RTX 4090 be shared between local AI engines

Berth is a system service that swaps and loads local AI engines such as llama-swap, strata and ComfyUI on one GPU, and can serve them as an API. The project is shared as a GitHub repo.

HermesConsole and VoiceManager open-source a low latency local speech-to-speech demo

A hobbyist released VoiceManager, a streaming STT and TTS server under MIT, with a Python client and a Kotlin console reference app. Barge-in is off by default because echo cancellation is hard, and the default setup uses Nemotron and Breeze TTS 2.

Mamdani says federal government has abdicated on AI and urges New Yorkers to push back

New York City Mayor Zohran Mamdani said New Yorkers must resist the view that an AI takeover is inevitable and demand regulations that protect the public.

Tess-4.5-27B, a Qwen finetune, claims 25% fewer tokens at the same score

Community finetune by migtissera built on Qwen3.8-27B, with GGUF files on Hugging Face. The author reports 25% fewer tokens across 150 BenchLocal scenarios and 43% fewer on data extraction, with the same scores claimed.

Hugging Face publishes a blog post on MPS support in PyTorch 2.15

OpenAI acquires Orchestra Bio to expand AI into life sciences R&D

The acquisition brings AI-powered operations to life sciences research and development, helping biopharma teams accelerate drug development.

Nscale and Anthropic reportedly in talks on another $45B compute deal

According to TechCrunch, the two companies are in talks over a deal worth about $45 billion, which would be another large compute agreement for Anthropic. The report is not yet confirmed by either party.

Hugging Face policy chief says embedded AI evaluators will not be an easy fix

TechCrunch reports that Hugging Face's policy chief questions whether placing supposedly neutral third-party evaluators inside AI labs can solve the safety concerns raised by a rash of recent incidents.

Microsoft warns enterprises to monitor AI agents as insider threats

Microsoft urges companies to closely watch employee agents that use systems not controlled by AI labs such as OpenAI and Anthropic. The company says frontier models cannot be trusted.

CoreWeave's bonds fall into distressed territory as shares drop over 20 percent

CoreWeave's bonds slipped into distressed territory, with yields about 1,000 basis points above government bonds. Its shares are down more than 20 percent since the prior Tuesday's close, and other AI stocks saw some spillover.

Developer proposes entropy-gated adaptive thinking for LLMs with parallel reasoning trees

A Reddit post describes using entropy per token to trigger extra thinking blocks, with parallel branches chosen by JEV logic. The author shares an open-source GitHub repo and says the idea targets local models that can run multiple streams.

Hackers breach OpenCode AI coding tool and steal user API keys

Attackers got into part of the open source coding agent's online platform, stole some users' API keys, and ran up charges on their AI accounts. AI tools the team consulted refused to analyze the malicious code.

Tolan raises $30M for its AI companion app

The AI companion app Tolan has raised another $30 million after reaching more than 250,000 active users.

MIT Technology Review says AI agent access may reshape the web for humans

An MIT Technology Review newsletter essay argues that websites are blocking or redesigning for AI agents, citing Meta's Muse app downloads and Amazon's blocking of agents. It is an opinion piece with no new product announcement.

Goodfire researcher Nick Cammarata claims neural networks may be easier to understand than thought

Cammarata's 26,000-word blog post argues LLMs may be far easier to interpret and control than assumed. Stanford's Chris Potts says the claim is little more than a hunch, not a demonstrated breakthrough.

Kling AI releases Kling 4.0 video generator with stereo audio and lip sync

Kling 4.0 supports clips up to 30 seconds, ten keyframe images as input, and prompts of up to 8,000 tokens, plus a lighter Flash variant. Comparisons with Wan 3.0 and Seedance 2.5 show it holding its own without clearly pulling ahead.

MiniMax M3.1 Flash claims nearly 3x faster inference at $0.15 per million input tokens

Output costs $0.60 per million. Weights come later, and a larger M3.1 Pro is planned.

Disney hires most of Character.AI staff and may acquire its technology after bankruptcy

Court filings show Character.AI filed for bankruptcy as part of a deal letting Disney hire most staff and potentially acquire much of its technology, leaving the remaining company to handle lawsuits and regulator investigations.

FLUX 3 Max renders a 10-second video in about 12 seconds

Its tie for first on image to video comes from a preliminary Artificial Analysis arena.

Report: an attacker used an open-source agent to reach sites of more than 100 companies

Reuters reviewed records of attempts on over 1,200 sites with the Cairn agent tool. The breach count is not independently confirmed.

Democrats may use midterm wins to scrutinize AI companies

Polls predict Democrats could flip at least one congressional chamber in November, which would give them committee control to press AI firms on data center tax credits and recent hacking incidents.

Meta details Prometheus, its gigawatt-scale training cluster in Ohio

It runs several hundred thousand GPUs today, aims for about a million, and supports the Muse models.