The speaker acknowledges a small bug in the Wordle simulation that mishandled the coloring of repeated letters. Fixing the bug changes the optimal opening guess: using a one‑step expected‑information metric, “soar” becomes best; a two‑step look‑ahead pushes “slain” to the top; and a full simulation of all 2,315 possible answers shows the lowest average score is achieved by the obscure word “Salé” (with real‑word alternatives “trace” and “crate” performing almost as well). Despite this shift, the core lessons about information, entropy, and the limits of greedy heuristics remain unchanged, and the speaker advises viewers to focus on those concepts rather than memorizing a supposedly optimal opener.
1. The speaker posted a video about solving Wordle using information theory.
2. The video included an addendum explaining a mistake.
3. There was a slight bug in the code used to recreate Wordle and run the algorithms.
4. The bug affected a very small percentage of cases.
5. The bug had only a slight effect on the results.
6. The bug related to assigning colors to guesses that contain multiple identical letters.
7. For the guess “speed” versus the answer “abide”, the first e is yellow and the second e is gray.
8. According to Wordle conventions, the first matching letter gets yellow; extra copies are gray if they are not present in the answer.
9. If the answer is “erase”, both e’s are yellow.
10. If one e is green and the answer has no second e, the second e is gray; if the answer has a second e elsewhere, the second e is yellow.
11. The bug arose from trying to speed up computations with a trick that was not fully thought through.
12. The optimal way to speed up is to pre‑compute all patterns for lookup.
13. The bug’s effect on the video’s substance is minimal; the main lessons about information and entropy remain unchanged.
14. Occasionally on‑screen distributions might be slightly off due to bucket counts.
15. The bug rarely affected shown words that had multiple letters hitting this edge case.
16. One substantive change caused by fixing the bug is the conclusion about the optimal first guess.
17. Originally the video claimed the best performance came from opening with “crane”.
18. After fixing the bug and rerunning, the theoretically optimal first guess changed.
19. Using one‑step expected information, the best opening guess is “soar”.
20. Using a two‑step expected information metric, the ranking changes: “soar” falls to 14th and “slain” rises to the top.
21. Simulating all 2315 possible Wordle games with the top 250 candidates gave the lowest average score for “Salé”.
22. “Salé” is an alternate spelling for a light medieval helmet.
23. “trace” and “crate” give almost identical performance and are both valid Wordle answers.
24. When sorting by lowest average score, “Salé” moved from third place to first, while “crate” and “trace” were fourth and fifth.
25. A blog post by Jonathan Olson explores optimal following guesses for some starting words.
26. Humans do not memorize the Wordle word list or perform exhaustive searches; they rely on intuition such as vowel placement.