Paradigm's SFT + DPPO Training Diagram Roasted as Looking Minecraft-Style
tensorqt · x · 2026-10-07
Paradigm shared that its post-training interleaves SFT with DPPO RL, alternating between novel policy exploration and exploitation of latent model capabilities — but the internet couldn't get over the diagram's blocky, "hired a bunch of Minecraft zoomers" art style.
More from Fun
- Seedance 'Age Is Just a Number' Meme Video With Shared Prompts — techhalla · 2026-10-07
- Meta researcher mocks AI safety figure Jacob Coxon as the Greta Thunberg of AI safety — mjdramstead · 2026-10-07
- Reasoning tokens reportedly wasted repeating "Redword" across samples — cis_female · 2026-10-07
- Dev complains: AI coding agents are 'conservative as hell' on production work — alexmacgregor__ · 2026-10-07
- Creator makes 'AI model release cycle' video entirely with Opus 5.5 and Suno — techartist_ · 2026-10-07
- Asked to film 'the meaning of life', Claude Opus 5.5 delivered something eerily human — char-gen · 2026-10-07