# @repligate — 2025-09-06

♥1 ↻0 · https://x.com/repligate/status/1964398735980712027

@lennyeusebi I’ll give you an example. An LLM can, in principle, visualize a complex object (and spend computation rendering it), without describing it in tokens, and then retrieve it later on, and only then describe what it had visualized internally

tags: author:repligate, kind:tweet, thread-context, year:2025