# @davidad — 2023-04-27

♥4 ↻0 · https://x.com/davidad/status/1651516958322442241

The way you describe the first one, it lacks anything to nudge the distribution in a particular direction, such as prompting (pre-conditioning) or filtering (post-conditioning). But if you meant to include those— how do you know they don’t work at sufficient scale? have you tried it with LLaMA? I thought it’s still more-or-less an open question

tags: author:davidad, kind:tweet, on:llama-1, year:2023
cited on: _dossiers/llama-1.md, llama-1