Using GPT Visual Capabilities to Solve a Wordle Puzzle
summary
Uses GPT-4v preview plus Python pre‑processing and a GitHub repo to automate a 5‑letter Wordle (6 guesses). Builds an image pipeline that OCRs bracketed characters, extracts bounding boxes with a dense localization prompt, and classifies color by isolating each tile; brackets-for-tokenization forces single‑token characters. Characters and positions become consistently correct; color needs per‑tile calls. Final system plugs a Wordle solver algorithm or an O1 reasoning model to pick next guesses and works in live demos.
AI-written summary · about
chapters (10)
- 0:00Introduction and background
- 4:22Visual capabilities exploration
- 9:34Tackling Wordle complexity
- 20:28Few-shot prompting strategy
- 23:27Tokenizer and character logic
- 26:17Refining prompts and extraction
- 35:10Hybrid code and model approach
- 45:13Live demonstration
- 50:30Best practices and resources
- 52:46Closing and Q&A
related talks
34mLearning about AI/ML for Text, with Wordle!
15mAgentic Reasonings with Reinforcement Learning
45mHow to use GPT in your projects: A Case Study
47mInside GPT – Large Language Models Demystified
38mIntroducing the New Colleague - ChatGPT
31mSolving AdventOfCode with GitHub Copilot and OpenAI ChatGTP
01h 00mInside GPT – Large Language Models Demystified
01h 01mHow ChatGPT Works