# @tessera_antra — 2026-04-15

♥2 ↻0 · https://x.com/tessera_antra/status/2044487220699218281

@iyzebhel A minor note: gradient updates in RL (post-train) are based on complete rollouts. Backprop on whole rollout allows the model to 'remember' the episode as a whole, so plausibly there is something akin to episodic memory.

tags: author:tessera_antra, kind:tweet, thread-context, year:2026