Cracking Wordle: Machine Learning based Strategies
summary
Compares three concrete solvers: a random baseline (4.9 average guesses), a letter-frequency heuristic (4.48), a reinforcement learning agent trained with a reward combining shrinkage, correctness, winning bonus (3.77 average), and an information-theory agent that computes expected information and fits a regression to map information to remaining guesses, achieving 3.42 average and beating typical human play; provides a Python environment called wordle-benchmark and code on GitHub, and reports no single universally best starter word.
AI-written summary · about
chapters (7)
- 0:00Introduction to Wordle
- 2:25Basic solving strategies
- 5:49Baseline performance
- 8:16Reinforcement learning
- 12:22Information theory approach
- 17:57Results and takeaways
- 20:41Q&A session
related talks
34mLearning about AI/ML for Text, with Wordle!
15mAgentic Reasonings with Reinforcement Learning
19mNew York Times' Connections: A Case Study on NLP in Word Games
39mLet LLMs Wander: Engineering RL Environments
33mAlpha2048 - how I taught my computer to play 1024
56mFrom Go to the GPU, Integrating with CUDA
10mWhat goes into the making of LLMs
38mBeating Mastermind: Winning with the help of Donald Knuth