Yacine slams claim that SAC beats PPO in sample efficiency as 'most idiotic thing today'
yacineMTB · x · 2026-10-04
Popular AI X account Yacine (ex-Tesla Autopilot) quote-tweeted a claim that "SAC was way better than PPO in sample efficiency," calling it the most idiotic thing he'd read all day — "I actually put my phone down and leaned back in my chair."
The post reignites a classic RL-community debate: PPO dominates on-policy RLHF/post-training pipelines, while SAC's off-policy sample efficiency in classic control tasks may not translate to those settings.
More from Fun
- AI safety spat: "Never criticize the ingroup" is bad if the ingroup could end humanity — zetalyrae · 2026-10-04
- Clicking ChatGPT's imagegen animation reveals a hidden snake game — flowersslop · 2026-10-04
- When a mathematician asks for help with a million-line Lean proof — ctjlewis · 2026-10-04
- Before the internet, people blamed stupidity on lack of information — it wasn't that — bennash · 2026-10-04
- Dev rant: we're too generous with people who make nothing and only chase attention — ctjlewis · 2026-10-04
- Yacine meme: eat your FLOPs, kids in Africa have no 8x GPU homelab — yacineMTB · 2026-10-04