[LIVE] Anthropic Distillation & How Models Cheat (SWE-Bench Dead) | Nathan Lambert & Sebastian Raschka

February 26, 2026

AI Summary

5 min read

🎙️ The Voices & The Context

The Format: Casual live discussion on Substack Live #6, blending insider AI analysis, screen shares, and real-time chat reactions—unscripted and spontaneous like a virtual happy hour for AI enthusiasts.
The Key Players:
- Nathan: Host from the Substack AI media coalition, sharp on geopolitics and benchmarks; welcomes Swix (Sean) as the newest writer.
- Swix: AI researcher/podcaster, expert on benchmarks and distillation; adds technical depth with live demos and recent podcast tie-ins.
The Vibe: Fun and educational—geeky excitement over AI drama, laced with banter on "attacks," privacy weirdness, and benchmark "bullshit," feeling like pals geeking out live.

🗝️ Key Themes & Topics

The discussion dives into AI frontier drama: Chinese labs "stealing" via distillation, benchmark saturation, and the messy reality of model training/evaluation. Four core topics emerge amid laughs and screen shares.

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

Listen to Audio Summary Open in App

What you'll learn

1 (00:00) **🎙️ Introduction: Sean Swix**
2 (01:44) **Anthropic's Distillation "Attacks" by Chinese Labs**
3 (06:33) **Detection Challenges and Implications**
4 (19:36) **Teacher-Student Dynamics in Distillation**
5 (28:50) **SWE-Bench and Benchmark Issues**
6 (43:47) **Future of Benchmarks and Evals**

+ Full timestamped outline available in the app

Show Notes

Swyx joined SAIL! Thank you SAIL Media, Prof. Tom Yeh, 8Lee, Hamid Bagheri, c9n, and many others for tuning into SAIL Live #6 with Nathan Lambert and Sebastian Raschka, PhD. Sharing here for the LS paid subscribers.

We covered:

This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.latent.space/subscribe