Welcome back! The AI landscape is getting increasingly self-referential. A model is now helping build its own successors. Tech giants are clashing over who controls the AI stack. And your personal knowledge base is becoming directly usable by intelligent agents.
In today’s Generative AI Newsletter:
MiniMax M2.7: The Chinese AI model that helped build itself.
Microsoft weighs legal action over the Amazon–OpenAI deal.
Readwise launches MCP server + CLI for AI access to your library.
Latest Developments
MiniMax M2.7: The Chinese AI Model That Helped Build Itself
MiniMax launched M2.7 marking a significant milestone in autonomous AI development. The company describes it as their first model to "deeply participate in its own evolution." Early versions of M2.7 were used to write their own training routines and tune how the model learns from human feedback.
Key features:
Self-Improvement Loops: M2.7 ran over 100 cycles of analyzing its own mistakes, rewriting code, and testing fixes.
Accuracy Boost: These autonomous loops resulted in a 30% accuracy increase on internal benchmarks compared to previous versions.
Competitive Coding: M2.7 scored 56.2% on SWE-Pro, placing it near flagship models like Opus 4.6 and GPT-5.3-Codex.
Agentic Logic: The model hit 55.6% on VIBE-Pro, proving its capability for complex, real-world engineering tasks.
The launch of M2.7 is a "shot across the bow" for Western labs that have been more secretive about their self-training methods. By automating reinforcement learning (RL) research, MiniMax is solving the biggest bottleneck in AI: the scarcity of high-level human researchers.
Special highlight from our network
Drop rough notes or an outline onto a flexible canvas. Chronicle turns that structure into a clear visual presentation.
Less time aligning boxes and fixing slides. More time shaping the narrative.
It’s a small shift. But it changes how presentations get built. And that’s what matters.
AI won’t replace creative thinking. What it can do is remove the formatting work between ideas and communication.
A simple way to test a new workflow and see how story-first presentations actually feel.
Get 1 month of free Pro access with code JOINGENAI
Special highlight from our network
What if your AI could search the entire web in real time and pull back exactly what you need?
Bright Data's Deep Research toolkit gives you everything to build this:
Real-time search across every major engine, browser automation for sites that fight back
Access to historical web archives so you can track how pages changed over time
Cross-source validation to make sure nothing slips through
Pre-built templates for competitor intel and market research
It's the full stack for building research agents that actually work at scale. Bright Data have an excellent free guide, give it a try.
Microsoft Weighs Legal Action Over Amazon-OpenAI Deal
Microsoft is considering legal action against Amazon and OpenAI. The dispute involves a new $50 billion cloud deal. Microsoft claims the agreement violates its exclusive contract to host OpenAI models on Azure. The conflict centers on Frontier, which is OpenAI’s new enterprise agent platform.
The standoff:
Hosting Dispute: Microsoft dropped its exclusive hosting lock last year. However, it kept a clause forcing all developer API access through Azure.
AWS Workaround: Amazon and OpenAI are building a "Stateful Runtime Environment" on AWS. This system handles memory and context locally.
Technical Clash: Microsoft executives believe this workaround is not feasible. They claim it still violates the spirit of their agreement.
Pentagon Access: The AWS deal allows OpenAI to support the Pentagon. This fills a gap left after the military labeled Anthropic a risk.
Microsoft sources told the Financial Times they will sue if a breach occurs. OpenAI maintains that the deal does not provide a backdoor to its models. Both companies are currently in talks to resolve the issue before Frontier launches. This tension grows while OpenAI prepares for a potential IPO later this year.
Tool of the Day: Readwise MCP Server & CLI
Readwise has launched an official Model Context Protocol (MCP) server and Command Line Interface (CLI), providing AI agents with direct access to your entire digital library. This integration allows tools like Claude Code, Cursor, and Codex to search, analyze, and reference every highlight, article, and document you have ever saved.
Core Functions:
Unified Knowledge Access: Connects your AI assistant to both Readwise (highlights from books and ebooks) and Reader (saved articles, RSS feeds, and PDFs).
Agentic Workflows: Enables complex tasks like triaging your reading inbox, generating weekly digests, and creating quizzes based on recently archived content.
Seamless CLI Integration: The Readwise CLI allows terminal-based agents like Claude Code to interact with your data without a complex manual setup.
Personalized Research: Agents can now build reading profiles from your history to offer tailored recommendations or synthesize themes across multiple sources.
Try this yourself:
To connect your favorite AI app, use the official MCP URL: https://mcp2.readwise.io/mcp. If you use a terminal agent, install the CLI by running npm install -g @readwise/cli and then readwise login to authenticate your account.
