Each guess must be a word (of any length). Unlike the original, only regular, lowercase words are allowed. The game will tell you how semantically similar it thinks your word is to the secret word. Unlike that other word game, it's not about the spelling; it's about the meaning. The similarity value comes from Word2vec. The highest possible similarity is 100 (indicating that the words are identical and you have won). The lowest in theory is -100, but in practice the lowest is something like -34. By "semantically similar", I mean, roughly "used in the context of similar words, in a database of news articles."
Secret words may be any part of speech, but will always be single words. It's tempting to think only of nouns, since that is how normal semantic word-guessing games work. Don't get caught in the trap!
The "Getting close" indicator tells you how close you are --if your word is one of the 1,000 nearest words to the target word, the rank will be given (1000 is the target word itself). If your word is not one of the nearest 1000, you're "cold".
You will need more than six guesses. You will probably need dozens of guesses.
Unlike the original, this version is run without a server. That means that all the computation is done locally on your device. This causes a few notable differences.
First, the game will start by downloading around 250 MB of data, which can take a few minutes. If you're on a metered connection, you may want to leave now.
Second, only regular, lowercase words with no punctuation are included to make the data more manageable. Therefore, proper nouns and short phrases do not work. Your guesses will be automatically convereted to lowercase, so if you guess "Nice", you are actually asking about the regular word "nice" and not the city Nice.
Click Ok to continue.