Sam Witteveen
LLMs, Agentic Frameworks, and RAG pipelines with a focus on practical implementation and local deployment.
Nutrition Label
Sam Witteveen specializes in technical breakdowns of new LLMs and agentic frameworks, often providing runnable code for developers. While his hands-on tutorials offer excellent practical value, his news-focused videos sometimes rely on theoretical analysis or vendor claims rather than independent testing.
Strengths
- +
- +
- +
Notes
- !Hands-on coding videos offer deep practical value, while news updates are often theoretical summaries.
- !Performance claims often cite vendor leaderboards rather than independent, standardized benchmarks.
Rating Breakdown
Breakdown across the key dimensions we rate. Methodology →
Recent Videos
Google's Agent Upgrade
Nano Banana 2 - Smaller, Faster, Cheaper
Caught Distilling from Claude?
Tiny Aya - Cohere's Mini Multilingual Models
KittenTTS - The Nano TTS
Introducing Gemini 3.1 Pro
The "Token Muncher" Problem: Is Sonnet 4.6 Actually Cheaper?
Qwen 3.5 - The next NEXT model
OpenAI Just Bought OpenClaw!!
Minimax M2.5 - What Makes This Different!
The Rise of WebMCP
Kimi K2.5- The Agent Swarm
Ollama Launch + Claude Code + GLM Flash
Clone ANY Voice for Free — Qwen Just Changed Everything
Beating Cowork with Open Source Cowork
Why this rating
Evidence receipts showing why each dimension is rated the way it is.
“This goes back to the 'Let's Verify Step by Step' paper... basically what they're doing is they're training a reward model not just on the outcome... but actually on the steps to get there.”[13:01] →
Demonstrates deep technical understanding by connecting the product's 'Forge' framework to specific seminal research papers (Process Reward Models).
“I'm going to run this... and you can see it's basically acting as a proxy now.”[03:15] →
The creator demonstrates the live execution of the 'ollama launch' command in the terminal, showing the actual server output rather than static slides.
“The idea here is that we have an agentic loop... the model can emit thoughts, it can emit calls to tools.”[05:27] →
Uses precise domain terminology to explain the architectural flow of the new standard.
“Beating Cowork with Open Source Cowork”[00:00] →
The title claims to 'beat' a product called 'Cowork' (likely referring to Claude Projects/Artifacts, as 'Claude Co-Work' is not a standard product name), but the video is an architectural overview without comparative benchmarks.