AI · Video
Theo Says Multi-Model AI Code Review Beats Hunting for One Perfect Model
Theo - t3.gg ·
At a glance
Theo explains how he uses several AI model families together to review and fix code, instead of picking one model per task. He shares his tstack skills, which let a 'council' of models check each code change until it meets high standards.
Key points
- Theo argues against building complex routers that send each task to a single model.
- He built tstack after seeing potato's pstack project, adding cross-model skills like council and sol-loop.
- He says models from the same provider share biases, so reviews across different providers work better.
- The council skill has three models from different families review, test, and score code from 1 to 10 until it hits a target such as 9.5.
Original source
Stop Searching For the Perfect Model
Theo - t3.gg · Nov 16, 09:09 KST
Watch on YouTube ↗AI summarized the video content. Check the original for accuracy. Copyright belongs to Theo - t3.gg.
More from this field
- AI-Related Child Abuse Material Reports Surge 8,400% in Two Years
A new report from Childlight, a child safety research group based in Edinburgh, found that reports of AI-related child sexual abuse material rose sharply over two years. The number went from about 4,700 in 2023 to more than 400,000 in 2025, and the study says AI-made fake images, including deepfakes made with nudify apps, are used to blackmail children.
The Guardian · Nov 16, 09:01 KST - Claude Expands Into a Work Operating System, Mythos Access Explained
Wes Roth looks at Anthropic's plan to turn Claude into a full work platform with Sheets, Whiteboards, and Claude Meetings. He also covers reports on how businesses use AI and how eligible researchers can apply for access to Anthropic's Mythos models.
Wes Roth · Nov 16, 08:27 KST - David Ondrej Shows How to Build an Open-Source Grok Bot From Scratch
David Ondrej shows how to build an open-source version of Grok Bot, the AI chatbot. He reverse engineers it and uses Claude Code and Cloudroom to build it from scratch in Rust. The bot is then self-hosted on your own virtual private server (VPS, a rented remote computer).
David Ondrej · Nov 16, 08:15 KST - LangChain Releases langchain-mistralai 1.2.0 With Streaming Fix for Reasoning Models
LangChain released version 1.2.0 of its langchain-mistralai package, which connects to Mistral AI models. The update fixes how streaming works for reasoning models and routes ChatMistralAI through the LangSmith gateway.
langchain-ai/langchain 릴리스 · Nov 16, 07:38 KST