GPT-3: First attempts

Scott's first GPT-3 experiments via AI Dungeon (July 2020): a transcript of prompts and the model's uneven completions — flubbed arithmetic, a rambling Civil War essay, a half-mangled Shakespeare sonnet, a 'why is the sky blue' non-answer — ending 'These results seem worse than other people are getting; am I doing something wrong?' Historically interesting as real-time early-LLM reaction, thin as writing.

Quality 50 · Solid. Solid, mid: mostly a curated transcript with minimal authorial content; its value is the period-snapshot of early GPT-3 tinkering, not craft or argument. 50.

Claude’s paradigm shift 38 · Slight. Slight/Moderate: a raw experiment log, little original content. 38.

Real-world impact 1 · Negligible. An experiment log; no material reach. RWI 1.

Humor 1/5 · Lizardman’s Constant. Gate at 1 (Lizardman): the AI outputs are funny, but the humor is found, not authored; light comic curation. 1.