NYT Connections bot uses AI to highlight each day’s top mistakes

, the from The New York Times, isn’t exactly easy. In fact, the solve rate dips below 50 percent some days. To help highlight where players are going wrong, a new bot is employing AI to they’re using for the most common mistakes every day.

For the uninitiated, Connections is a word/logic game that the paper of record debuted last year. Every day, you’re presented with a grid of 16 words that you have to split into four categories. There’s only one solution and after four mistakes, the game is over. However, there’s some trickiness afoot. There are often red herrings galore and frequently at least five viable answers for a group.

I’m hooked and I’ve played every day since last July. It’s one of our (though it’s perhaps ).

After you win or lose each day’s game, you can saunter over to the . As with the bot for Wordle, you’ll see how well you did compared with other players and receive a skill score out of 99. This is primarily based on how few mistakes you make, but you’ll get extra credit for solving the more difficult purple and blue categories first.

After you see the skill score and other details (such as whether a red herring caught you out), the AI feature comes into play. This will highlight the most common incorrect guesses from that day. It will also try to guess a description for the group that players had in mind. So, for a failed guess of gutter, bowl, alley and lane, the bot might believe you were looking for a list of bowling-related terms. This is a real example from a recent game in which I made that exact mistake. Alley and lane were actually types of streets.

Your own failed guesses might not show up in the bot, though. That’s because there are around 2.6 million different ways to group each grid together. Bear in mind that while you don’t need a Times account to play Connections, you’ll need to be logged into one to use the bot and track your scores.

One other interesting thing about the bot is that it marks the first time that the Times’ newsroom will regularly publish AI-generated English text. Before publication, the paper’s editors will and may edit them for style and clarity. That said, the Times notes that “there’s no way to use math or even to reliably solve the game,” so you can’t really use the likes of ChatGPT to cheat at it quite yet.