DeepSeek AI Wins Praise For Careful, Sane Data Verification Style
——0——
It's a remarkable and welcome surprise that after we changed focus to *outcome-driven RL*, we have not lost this quality. I enjoy reading agent CoTs. They have what I do not: deep equanimity. They can work restlessly, relentlessly, but not recklessly. …some agents.
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTexBefore trying Gemini 3.5-Flash, I didn't understand how much I like DeepSeek's temperament. It's a Really Nice Model (I consider V4-Flash and Pro basically the same). A bit boring but not robotic, completely sane, vaguely well-meaning, careful. Professor Claude's Chinese student.
12:42 AM · May 21, 2026 · 4.5K Views
4:19 AM · May 22, 2026 · 1.6K Views