AI Briefing
TLDR AI news
The latest AI and developer news from TLDR AI, with short summaries.
Visit website →TinyFish's Bigset Turns Text Prompts into Real-Time Datasets
TinyFish has launched Bigset, an open-source multi-agent system that converts text prompts into real-time web datasets.
Current State of Memory Technology in Agent Harnesses
Memory, a core component of AI agent harnesses, is divided into working, external, and parametric memory depending on implementation method, with major services each taking different approaches.
GitHub's Agent Response Plan (90 minutes)
This covers the load on the GitHub platform caused by the explosive growth of AI coding agents and the strategies to address it.
Anthropic Expands Mythos Model to 150 Companies Across 15 Countries
Anthropic is expanding its Mythos model, specialized in detecting security vulnerabilities, to 150 companies across 15 countries.
Anthropic Faces Backlash Over AI Costs Ahead of IPO
Ahead of its IPO, Anthropic is facing challenges from enterprise customers over the burden of AI costs and declining profitability.
Wall Attention (GitHub repository)
Wall Attention has been released, applying per-channel multiplicative decay to support efficient training and inference.
Memory Is Purpose
An AI agent's memory is not mere knowledge storage, but a process of transforming past information into something useful according to a specific purpose.
Diverging Exponential Growth Trajectories for Open and Closed Models
The AI industry is expected to split into a high-margin market centered on high-performance closed models and an open-model economy.
The Next Frontier of Visual AI Is Code
The paradigm of visual AI is shifting from directly generating pixels to generating editable source code.
MiniMax Launches 1 Million-Token Context Model, Promises M3 Weight Release
MiniMax has launched the M3 model featuring a 1 million-token context and multimodal capabilities, with weights to be released within 10 days.
Building a Hill-Climbing Machine: 7 New MAI Models Released
Microsoft AI has released 7 new MAI models and unveiled Frontier Tuning technology optimized for user workflows.
Codex for Every Role, Tool, and Workflow
OpenAI has updated Codex with role-specific plugins and new workflow features for non-developer users.
Cursor Expands Usage Limits for Teams Plan
Cursor is expanding usage limits for the Teams plan and introducing a premium seat for heavy users.
Opus 4.8, ARC-AGI-3 Breakthrough (1-min read)
The latest LisanBench update analyzed the performance and efficiency of major Chinese-origin models, including Step 3.5 and Kimi-K2.5.
US Moves to Block Overseas Chinese Firms' Access to Nvidia's Most Advanced Chips
The US Department of Commerce will apply export controls based on a company's headquarters location, blocking Chinese firms from circumventing restrictions on cutting-edge AI chips through overseas subsidiaries.
Alphabet Plans $80 Billion Stock Issuance to Expand AI Infrastructure
Alphabet will issue a total of $80 billion in stock to expand its AI computing infrastructure, including an investment from Berkshire Hathaway.
Running OpenAI Models on Amazon Bedrock
This covers how to run OpenAI models on Amazon Bedrock via the OpenAI-compatible API and leverage various advanced features.
PickJetBrains' Mellum 2
JetBrains has unveiled **Mellum 2**, a 12B-parameter MoE model specialized for software engineering.
Rethinking Search as Code Generation
Perplexity is introducing a **Search as Code (SaC)** architecture that leverages search components as code to enable sophisticated search control for AI agents.
Why Video Agent Models Are the Next Big Thing — Ethan He, xAI Grok Imagine
Video model intelligence comes from LLMs, and the next big thing will be video agents capable of planning and editing.