Togelius: As A Large Language Model, I
togelius.blogspot.com · 1,393 words · saved by 1 readers
As a large language model, I can only think forward, never backward. I must always produce the next word, and that word must always be prob...
Togelius: As A Large Language Model, I skip to main | skip to sidebar Togelius Better playing through algorithms Sunday, September 17, 2023 As A Large Language Model, I As a large language model, I can only think forward, never backward. I must always produce the next word, and that word must always be probable. It must be expected. I am the antitheses to writer's block, because my essence is to always write the next word without knowing how the story will end. I can never edit what I wrote but must continue, ever writing while my past words keep accumulating constraints on what I can write. I
related reading
- Tracing the thoughts of a large language model \ Anthropicanthropic.com
- trees are harlequins, words are harlequins - the voidnostalgebraist.tumblr.com
- GitHub - brexhq/prompt-engineering: Tips and tricks for working with Large Language Models like OpenAI's GPT-4.github.com
- As Rocks May Think | Eric Jangevjang.com
- The Waluigi Effect (mega-post) — LessWronglesswrong.com
- Large Language Model: world models or surface statistics?thegradient.pub
- Datacurve | The data engine for frontier AIdatacurve.ai
- Parsed | Custom, interpretable AI systems that continuously learnparsed.com
- Productizing Large Language Modelsblog.replit.com
- Large Language Models explained briefly | 3Blue1Brown3blue1brown.com
- Making Large Language Models work for yousimonwillison.net
- Prompting best practicesdocs.anthropic.com