Nonfiction writing by OpenAI’s GPT-3 model, testing logic, commonsense reasoning, anagrams, PDF/OCR cleaning, creative nonfiction, etc
GPT-3 can’t really do ASCII art, which comes as little surprise to me given the relative rarity of ASCII art these days, the difficulty of understanding what it’s art of (even looked at in 2D), and the possibility that most ASCII art was filtered out of the training dataset. AK got a little better results by prompting using just ASCII art.
Per a suggestion, I tried a ‘contest’ prompt:
- Anagrams
- Logic
- The Database Prompt
- Parity
- Verbal Counting
- Concept Blending
- Coq Proofs
- ASCII Art
- PDF Cleaning
- Meta-Prompts
- Common-Sense Knowledge
- Expressing Uncertainty
- “Why Deep Learning Will Never Truly X”
- Arxiv Paper
- Overcomplicated Explanations
- Epigrams & Proverbs
- Movie/Book Plot Summaries
- Problematic Things
- Dwarf Fortress Changelog
- Board Games
- Art Criticism
- Individual Imitations
- Two-Digit Arithmetic