profile

The LLM Edge

πŸ—πŸš’πŸš€ DeepSeek-R1

Published 2 months agoΒ β€’Β 2 min read

​

Hey, AIM community!

​

Next Wednesday, join us as we look into COCONUT: Chain of Continuous Thought. Following up on our recent LRMs event on DeepSeek-R1, we’ll continue exploring Chains of Thought (CoTs), but this time in latent space! We might even go beyond COCONUT to talk a bit about latent recurrence as well.

​

Last week, we explored Deepseek-R1! We covered a brief history of models from DeepSeek, then tied together important ideas ranging from CoT, to test-time compute, to process and outcome reward modeling, to RL versus RLHF, and more!

We learned that DeepSeek-R1 and DeepSeekMath use RL for real, in a way that lets the LLM "play" during training to figure out CoTs. Then, we can apply these directly to domains beyond math and code, in contrast to typical RLHF approaches that fix outcomes based on specific human feedback.

PS ... s/o to the guys at Unsloth for their "Train your own reasoning model" blog!

🧰 Resources

πŸ”­ Coming Up!

Cursor: An AI Engineer’s Guide

Join us live for a 1-hour dive into how to set up a proper 2025 development environment that the best AI Engineers use, and build, ship, and share your very first LLM application with these new tools! If you're not sure where to start with AIE, start here.

PydanticAI: From Data Validation to Agents

Question: "Is it true that just because my team is good at using Pydantic that I should consider using PydanticAI for agents?" It is built by the same team, after all. We've been asked this enough by our community and in our course - let's find out together, live!

🌐 Around the Community!

πŸ’‘ Transformation Spotlight: Allan Tan! A serial techno-preneur and founder of Predictive Systems, Inc., learn why he still loves to code and what he's working on at the LLM Edge.​

video preview​

πŸ€“ See what the community is building, shipping, and sharing this week. Join us in the Lounge every Monday at 9 AM PT for some accountability!

​

Want to join the AIM community? Hop into Discord and share your intro!

​

πŸ–ΌοΈ Meme of the Week

🌟 Want to start building, shipping, and sharing, but not sure how? Check out our LLM Foundations - a 5-day email-based course.

​

Keep building πŸ—οΈ shipping 🚒 and sharing πŸš€,

​

​Dr. Greg, The Wiz, Seraacha, and Lusk​
​AI Makerspace​

​
​Unsubscribe Β· Preferences​

The LLM Edge

Weekly Concepts, Code, and Community! Every Saturday you'll receive a detailed overview of one the latest tools or techniques from the open-source LLM edge.

Read more from The LLM Edge
The Llama 4 Herd

Hey, AI Makerspace community! This Wednesday, we'll cover the new release from OpenAI: Codex CLI, a lightweight coding agent that runs in your terminal. Carrying the same name as the original Codex models released in 2021, this new open-source tooling is a CLI-based coding agent that bears little resemblance to the model that originally powered GitHub Copilot. Join us to understand coding agents and best practices for their use in 2025 πŸ” Coming Up: What is an Agent? OpenAI just released their...

6 days agoΒ β€’Β 1 min read
RAG: The 2025 Best Practice Stack

Hey, AIM community! Tomorrow, we'll cover Enterprise Agents with OpenAI! What does the agents SDK look like from OpenAI? How does it build on previous work they've done? Are they officially in the end-to-end platform game competing with orchestration frameworks like LangChain, LlamaIndex, CrewAI, and others? Join us live to find out! Last week, we discussed RAG: The 2025 Best-Practice Stack This is the year of Practical RAG, and we kicked it off by unpacking the Minimum Viable...

26 days agoΒ β€’Β 2 min read
DeepSeek Week

Hey, AIM community! On Wednesday, we'll cover the infra stack that we recommend for RAG in 2025. Then, we'll build, ship, and share a best-practice RAG app. We'll also discuss important production tradeoffs and implications that you should consider before and after deployment when going from zero to production RAG! Last week, we discussed the latest open-source repo drops from DeepSeek Week, and we covered how they're being used as a new best-practice way to do inference on MoE models via...

about 1 month agoΒ β€’Β 2 min read