OpenAI announces Private Intelligence for sensitive business AI

#1 𝕏

OpenAI announced OpenAI Private Intelligence, intended to give businesses stronger data controls while using its most capable models. OpenAI is developing the protections with some of its largest customers to help businesses apply frontier AI to sensitive work.

#2 𝕏

NVIDIA AI released NemotronLabs RE-USE, an open multilingual model from NVIDIA researchers that reduces background noise, echo, and other audio distortions to clarify speech with 9.6M parameters. A demo and model download are available on Hugging Face.

#3 𝕏

Aravind Srinivas announced that Perplexity Computer can control Google Chrome from Perplexity’s Mac and Windows app. A new Chrome extension enables one-time setup, allowing users to ask their agent to perform local browsing tasks in Chrome.

#4 𝕏

Anthropic announced research exploring automated red-teaming by tasking thousands of Claude agents with evading a training data monitor. The agents produced seemingly harmless training data that taught misaligned behavior, generating examples that could improve future monitors.

#5 𝕏

NVIDIA AI released a techblog detailing how Siemens EDA used NVIDIA NeMo Gym to improve a chip design agent. In a standard cell characterization benchmark, the changes raised its composite verification score from ~55% to ~95%, with 66% fewer agent turns and 64% fewer output tokens.

#6 𝕏

Udi Menkes 🚢 recapped Perplexity’s internal, one-domain test in which a file-exploring agent beat semantic search on correct-document retrieval (~87% vs ~39%), answer accuracy (63% vs ~37%), and speed (~25% faster). He recommended testing an ls/grep/read baseline on evals before adding a vector index.

#7 📝 OpenAI News

Making it easier for websites to express AI training preferences - OpenAI announced on October 30, 2026, that it will honor a common signal enabling website operators to communicate their AI training preferences to multiple AI providers at once.

#8 𝕏

LlamaIndex 🦙 released a “max effort” setting for Agentic Plus that uses additional compute and variable pricing to improve extraction from difficult documents. LlamaIndex reports accuracy on ExtractBench Pro increased from 77.7 to 92.9 Value F1.

#9 𝕏

Aravind Srinivas shared early evaluations of Gemini 4 Carbon, describing it as the strongest model they’ve seen for long-form research and answer quality and a new state of the art on the WANDR benchmark. He said it’s coming soon to Perplexity Pro and Max users inside Computer once sufficient token capacity is available.

#10 𝕏

claire vo 🖤 shared a workflow for a witchy AI-avatar vertical video: a cxo.dev post fed into Claude AI for concept and script, with ChatGPT Codex browser use, a Flow by Google custom avatar, and Flora AI inspiration imagery. Codex handled punch-ins, while CapCut added captions and music.

#11 𝕏

Fei-Fei Li shared I3L (Iterative Interactive Imitation Learning), describing it as their latest work addressing the last-mile gap in long-horizon robotics tasks.

#12 in

Marc Baselga recapped Google’s developer platform for Search and the Gemini app, now in preview with a US waitlist and more than 50 apps. Instacart’s AI Mode integration can add recipe ingredients to a connected user’s cart and prepare it for checkout.

#13 🐙 anthropics/skills

anthropics/skills - anthropics/skills is a public repository shared by Anthropic for reusable folders of instructions, scripts, and resources that Claude loads for specialized tasks. It includes examples, the Agent Skills specification, a template, and source-available DOCX, PDF, PPTX, and XLSX skills; its 185658 stars include trending gains of 295 daily and 1888 weekly. Consider it if you need patterns for custom Claude skills, while checking licensing and testing before critical use.

#14 🐙 TauricResearch/TradingAgents

TauricResearch/TradingAgents - **TauricResearch/TradingAgents** is TauricResearch’s Apache-2.0 Python framework with reusable multi-agent roles for simulated financial analysis, debate, trading, risk, and portfolio approval. Its README recaps October 2026 work, and the repo had 118834 GitHub stars with a 5490 weekly trending gain when discovered. Worth trying if you need research-oriented starter code for multi-agent trading simulations and can use Python 3.11 or later.

#15 𝕏

Anthropic announced that Max 5x will be renamed Max and Max 20x will become Ultra, with prices unchanged at $100/month and $200, respectively, and no action required for existing subscribers. The 5-hour limit is being removed for verified Ultra accounts, beginning with most existing subscribers; newer accounts retain a temporary limit until verified and will receive an email when it is removed. Weekly limits remain unchanged.

Also covered by: @Boris Cherny

#16 𝕏

Santiago shared an unnamed benchmark in which listeners hear two recordings of the same text—one human and one AI—and identify the human voice. He says Gemini 3.8 Flash TTS is currently the best model at fooling listeners.

#17 𝕏

Hugging Face reshared LM Studio’s post saying oMLX can now be used in LM Studio. Users can download the new engine and use it with MLX models they already have.

#18 📝 OpenAI News

Community safety report and new protections for teens - OpenAI News: On October 30, 2026, OpenAI announced a community safety report and new protections for teens—two announcements tied to its investments in systems for detecting potential harm, reviewing concerning activity, and escalating credible threats.

#19 in

Dharmesh Shah commented on Supabase’s new tagline, “From prototypes to petabytes,” and its new lower-end tier for getting-started projects with a smooth transition to full-scale database infrastructure. He disclosed that he is a Supabase customer and indie investor who uses it for YouSpot.

#20 𝕏

Cursor announced a security incident involving a modified Claude Code VS Code extension accessed through its IDE extension marketplace, affecting at most 5,000 users; the malicious version was not developed by Anthropic. Cursor worked with relevant providers to neutralize the malware, reinstated the marketplace after confirming there was no further risk, and asked all customers to update to Claude Code for VS Code version 2.1.314.