2026-08-13
AI Briefings Daily Report.

Apple in talks to pay for publisher content, providing real-time news for Siri AI ↗

The tech giant has considered a nine-figure budget for the payments, according to the WSJ.

DeepSeek Harness tops Hacker News hot list ↗

DeepSeek Harness: Everything is a Plugin. Contribute to deepseek-ai/deepseek-harness development by creating an account on GitHub.

10% of this article was generated by AI. ↗

Pangram, Authorship, and French Theory

Some Claude models will add"Text Watermark", Anthropic responds to concerns from various sectors ↗

Some Claude models directly embed an "imperceptible watermark" in their output text. This watermark remains with the text after copying and pasting, and may persist even after certain modifications, helping to determine whether a piece of text was generated by AI or written by a human.

Claude Tag now reads even more of the room ↗

Claude has more context to decide when to proactively collaborate in Slack (and when not to)

DeepSeek Harness enters public beta, benchmarks against Claude Cowork and opens plugin ecosystem ↗

The supporting plugin ecosystem is opened simultaneously; the newly released V4-Pro model has 1.6 trillion parameters, and its weights have been opened under the MIT license, adaptable to domestic GPUs to lower the private deployment barrier.

a16z invests in Vals: building a trust layer of independent evaluation for AI ↗


DeepSeek-V4-Pro official version launched: natively supports OpenAI Responses API, specifically adapted for Codex ↗

DeepSeek V4 Pro official version has been simultaneously updated and launched on App, web end and API. Users can select "Expert Mode" in the App or web end to use the all-new V4 Pro official version model.

DeepSeek-V4-Pro official version launched ↗

DeepSeek-V4-Pro official version launched, Agent capability greatly enhanced ↗

DeepSeek API pricing adjustment plan announced, effective August 17 ↗

DeepSeek announced that from August 17, DeepSeek API will adopt peak-valley pricing, with off-peak prices at only half of peak times. This price adjustment covers multiple models in the V4 series, aimed at optimizing resource utilization.

AsiaInfo × Volcano Engine: 6000+ seats run enterprise-grade AI R&D with TRAE ↗

Claude's"Red text"Watermark: currently invisible, but about to cover all processed content ↗

Missing the mark? The mark flags anything Claude processed, even human writing it only edited. 

Researchers at top AI labs previously warned about automated AI research; several predicted milestones have now been achieved ↗

IAPS fellow Severin Field interviewed 25 researchers from OpenAI, Anthropic, Google Deepmind, Meta, and US universities about recursive self-improvement. In a new blog post, he takes stock. Several of the milestones those researchers named have already been hit.

Inside story of Google DeepMind leadership change: After Gemini delayed two months, coding ability still lags rivals, Brin personally oversees and demands"All-in" ↗

After Gemini's two-month delay, its coding ability still lags behind competitors, so Google DeepMind underwent a high-level reshuffle, with co-founder Brin personally overseeing, and the adjustment also raised concerns about DeepMind's independence.

Anthropic introduces Claude Cowork to Chrome extension, supporting skills and plugins in the browser ↗

Claude Cowork now runs directly in the side panel of Anthropic's Chrome extension.

Y Combinator CEO Chen Jiaxing: Entrepreneurs who make good use of AI agents will live in 2028 ahead of time ↗

Speaking of AI agents, Chen Jiaxing said: 'You have to crank the intensity all the way up. For example, give it 1 million tokens at a time, or 800,000 tokens. I feel that at that level, you're basically living in 2028 ahead of time.'

Okta uses MCP scopes to cut AI agent token costs ↗

Checking the site connection security

Cambricon has completed adaptation of five major domestic large models, sixth-generation chip architecture is under development ↗

Company discloses semi-annual report, H1 revenue and net profit doubled year-on-year, has completed inference adaptation for five domestic mainstream large models, sixth-generation chip architecture still in R&D, continuously advancing AI technology R&D.

Claude Code v2.1.231 released ↗

Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands. - Release v2.1.231 · anthropics/claude-code