AI Briefing
TLDR AI news
The latest AI and developer news from TLDR AI, with short summaries.
Visit website →Introducing Ternary Bonsai: State-of-the-Art Intelligence at 1.58 Bits
PrismML has released Ternary Bonsai, a family of 8B, 4B, and 1.7B LLMs using 1.58-bit ternary weights.
Qwen 3.6 and Agentic Coding (GitHub Repository)
Qwen3.6 is the latest Qwen model family with strengthened agentic coding and thinking-context preservation.
Migrating Codebases with Sandboxed Agents
Agents SDK and sandbox safely automate legacy code migration.
What We Learned This Week (20 min read)
Frontier models are harder to prevent **distillation** of than expected, and large-scale pretraining can easily collapse from small **bias** and **causality** violations.
What Jensen Huang Said About Anthropic, OpenAI, China, and Inference Token Demand
Nvidia is solidifying its AI infrastructure dominance through supply chain, financial, and China strategies.
The Computer is Personal
JavaScript was disabled on X, so the article body could not be loaded.
A New Programming Model for Durable Execution
Vercel Workflows moves to GA, handling long-running agents and backends inside your code.
A PR I would have opened myself
Released a Skill and verification harness that helps port transformers to mlx-lm.
Introducing GPT-Rosalind for Life Sciences Research
GPT-Rosalind has been released as a reasoning model to help with life sciences research and drug development.
Codex for Almost Everything
Codex now handles computer use, web browsing, and even image generation.
A new way to browse the web with Chrome's AI Mode
Chrome AI Mode lets you browse the web and search results side by side without switching tabs.
Google Testing Agentic Shopping and Native Checkout Features in Gemini
Google is building a shopping cart and direct checkout functionality right into Gemini.
Jane Street Commits $6 Billion to CoreWeave and Acquires $1 Billion Stake
Jane Street committed $6 billion to CoreWeave and invested $1 billion.
Claude Was Probably Not Secretly Nerfed. Anthropic Just Made the Black Box Too Dark
The core of the Claude Code performance-degradation controversy is less about model nerfing and more about hidden operational variables.
Jensen Huang - TPU Competition, Why Chips Must Be Sold to China, and Nvidia's Supply Chain Moat
Nvidia's moat lies not in the chip itself but in its power to control the supply chain and ecosystem together.
Multi-Tier Instruction Hierarchy for LLM Agents
The authors propose **ManyIH**, which resolves agent instruction conflicts using up to **12 tiers** of authority.
NVIDIA's Lyra 2: Explorable Generative 3D Worlds
Lyra 2.0 reduces spatial forgetting and temporal drifting to create longer, more consistent 3D worlds.
Parcae, a stable looped model that achieves higher performance with fewer parameters
Parcae scales up recurrence to reach 1.3B Transformer-level performance with 770M parameters.
Why dLLMs Are Prone to Collapse Under RL
This piece examines why dLLMs are prone to collapse when RL is applied to them.
Humwork A2P Marketplace Connects AI Agents with Human Experts
Humwork has launched an A2P marketplace that connects AI agents to human experts in real time whenever the agents get stuck.