November 19, 2026 Understand what changes. Find what to use.

Updated today at 7:08 AM

Your daily AI edition

What deserves your attention.

What's new, the context, and what you can do with it.

1000 stories · Updated today at 7:08 AM

Friday, October 23

SecurityNews1 source

Ars Technica: Google policy may let Gmail data inform ads

According to Ars Technica, an update to Google’s privacy policy may let Gmail and Google Photos data inform ads when shared with other services. Google says ads will not be based directly on that data.

Full summary

ModelsNews2 sources

Block announces Dessa, a voice model trained only on human conversation

Block announced Dessa, a full-duplex voice model trained, according to the company, solely on human speech. It can run locally on a MacBook Pro M4 Max or better, but is not publicly available yet; safety researchers will receive early access.

Full summary

ModelsNews1 source

IBM CEO says open-weight models reach 90% of performance at 10% to 25% of the cost

IBM CEO Arvind Krishna said open-weight models reach about 90% of the performance of the most advanced models and cost 10% to 25% as much per token. He did not cite specific tests to substantiate the comparison.

Full summary

SecurityNews1 source

Anthropic cybersecurity deputy says the situation “will get much worse”

Jason Clinton, Anthropic’s deputy chief of cybersecurity, wrote on X that the situation will worsen before it improves and urged security leaders to reinforce basic measures. The warning came amid the disclosure of critical PHP vulnerabilities, but he did not explicitly link the events.

Full summary

RegulationNews1 source

German court provisionally orders Google to block false AI Mode accusations

A court in Aachen ordered Google to prevent AI Mode from falsely attributing a sexual-assault offense and employment consequences to a doctor. The ruling treats responses without supporting links as statements by the company itself, but it can still be appealed.

Full summary

SecurityNews1 source

Google updates policy and may use Gmail and Google Photos data for targeted ads

A Google policy update may allow targeted ads based on Gmail, Google Photos and Drive data when those services are connected to personalized AI features. Advertisers will not get direct access to the data, and users can disable personalization.

Full summary

ResearchNews1 source

Hugging Face says LitPivot helped researchers ground ideas in the literature

Hugging Face and University of Pennsylvania collaborators developed LitPivot, a system that organizes papers around parts of a research proposal and suggests revisions. In a study with 17 researchers, experts rated ideas developed with the tool as better grounded in the literature.

Full summary

AgentsNews1 source

Unitree wants to use AI to help its robots learn to improve themselves

Unitree Robotics presented a system that uses AI agents to research, write and test control code, aiming to speed up humanoid robots’ acquisition of skills. The company says the method still needs to be validated in the real world.

Full summary

BusinessNews1 source

Meta delays new business terms after publisher pushback over AI scraping

Meta delayed enforcement of new business terms from October 30 to November 20 after publishers questioned whether they would allow news sites to be scraped for AI training. The company says it will use the extra time to explain the changes and how its crawlers work.

Full summary

CodeNews1 source

The Verge: AI speeds up retro game ports and strains preservation communities

AI tools have made reverse-engineering projects and retro game ports move much faster, The Verge reports. The acceleration is also putting pressure on preservation communities, with hostile forks and little contribution to original projects.

Full summary

ResearchNews1 source

Nick Cammarata says LLMs may be easier to interpret and control

In a 26,000-word post, Goodfire’s Nick Cammarata says large language models may be easier to understand, monitor and control than previously thought. Stanford researcher Chris Potts considers the idea little more than an intuition.

Full summary

SocietyNews1 source

Indian workers are paid to film activities used to train physical AI

Workers in India and other countries are being paid to record their activities in videos used by technology companies and research groups to train physical AI. Demand for more detailed data already includes equipment with multiple cameras.

Full summary

BusinessNews1 source

inDrive plans to test Pony.ai robotaxis in Kazakhstan in 2027

inDrive plans to begin testing Pony.ai robotaxis in Astana, Kazakhstan, in the first half of 2027. The project would use Nvidia technology and initially include a safety operator onboard.

Full summary

Thursday, October 22

HardwareNews1 source

NVIDIA says Kazakhstan is expanding AI computing capacity

NVIDIA says Kazakhstan is moving from AI infrastructure plans to deployment. During Jensen Huang’s visit to Astana, the company announced an investment in Higgsfield AI and highlighted a Firebird project involving 45,000 NVIDIA Grace Blackwell GPUs.

Full summary

ResearchNews1 source

Hugging Face presents catalog for checking models on AWS Trainium

Hugging Face presented a Space showing which popular Hub models run unmodified on AWS Trainium. The tool also reports the resources used, tested versions and provides a script to repeat the tests.

Full summary

ResearchNews1 source

Microsoft and Hugging Face say ThinkingBox improved Qwen3.8-27B success

In a joint post, Microsoft and Hugging Face say ThinkingBox raised Qwen3.8-27B’s average success from 51.70% to 60.78% after 50 GRPO steps. Tasks succeeding on all 20 attempts increased from 38 to 107.

Full summary

BusinessNews1 source

OpenAI reportedly negotiating a $30 billion round at a $1.25 trillion valuation

OpenAI is reportedly negotiating a private round of up to $30 billion at a $1.25 trillion pre-money valuation, according to sources cited by CNBC. The terms could still change, and the funding would come ahead of a potential IPO.

Full summary

ResearchNews1 source

Microsoft says Flint-interactive can make AI-generated charts interactive

Microsoft presented Flint-interactive, a language that lets AI agents create explorable visualizations with zooming, filtering, highlighting and cross-chart interactions. The proposal uses the Flint compiler to reduce the complexity and reliability problems involved in building such visualizations.

Full summary

HardwareNews1 source

MIT Technology Review: enterprise AI’s edge depends on infrastructure around the model

MIT Technology Review argues that enterprise AI’s competitive advantage increasingly depends less on model choice and more on the serving infrastructure around it. Architecture, memory, scheduling and optimization determine whether that intelligence can be delivered reliably, economically and at scale.

Full summary

AI in practiceNews1 source

Microsoft and Public Library Association expand free AI literacy training

Microsoft announced an initiative with the Public Library Association to expand free AI literacy and skills resources and training for public library workers and patrons in the United States. The program includes webinars, guidance and a community of practice.

Full summary

One story, many sources

Coverage of the same story gathered in one item, with a link to each original.

Summary and interpretation apart

What the sources say stays in the summary; editorial context is labeled separately.

Always with the source

Every story links to the original articles, and the full content belongs to them.