Or just play with your friends?
my friends were bad clue givers. i just had to switch to ai.
OpenAI's o1 Playing Codenames
111–120 of 138 posts
Re: OpenAI's o1 Playing Codenames
#112I mean, it's playing against itself, not really a fair comparison to humans in my mind. The fun and hard part of this game is to get into your teammates brains and decipher what they possibly meant with what they played.
yea, didn't mean to take the fun out of the original game.
the idea for this came when we asked chatgpt how to connect the words 'carrot' and 'ray'. maybe you can give a try too!
Re: OpenAI's o1 Playing Codenames
#113Fun quirk about this game: If there aren't too many cards left and your teammate knows their powers of two, you have a winning strategy. You simply lay a mental bitmap over all remaining cards, setting 1 for cards that belong to your team and 0 for all others. You can then just say the number that is represented by this bitmap, e.g. "five" for 0101, and your teammate can decode it in their head. All numbers are, afte…
goated.
Re: OpenAI's o1 Playing Codenames
#114GPT-3 was superhuman at this too
yep, agree. One big part of the experiment was to see how well it does the reasoning by asking it to output the reasoning.
Re: OpenAI's o1 Playing Codenames
#115It would be really interesting to see an LLM watch other players and learn how they think to find the best clues THEY need to hear to find the right words.
definietly an interesting approach. I started writing down gameplays when i play with friends. then eventually stopped to enjoy the moment.
Re: OpenAI's o1 Playing Codenames
#116You can pretty reliably get 2-clues and sometimes good 3-clues just using word2vec embedding similarity
agree. i think getting to have a look at o1's reasoning was pretty fun.
Re: OpenAI's o1 Playing Codenames
#117I cracked myself up with a ridiculous train of thought for fun while playing Codenames once. It went a little something like this Star => Twinkle => Twinkle Khanna => Married to Akshay Kumar => Canadian Citizen => Maple Syrup ( Leaf ? )
haha, i've been in similar situations, but this one's something else.
Re: OpenAI's o1 Playing Codenames
#118I mean, it's playing against itself, not really a fair comparison to humans in my mind. The fun and hard part of this game is to get into your teammates brains and decipher what they possibly meant with what they played.
yea, didn't mean to take the fun out of the original game. the idea for this came when we asked chatgpt how to connect the words 'carrot' and 'ray'. maybe you can give a try too!
I still enjoyed reading your post, it's fun and interesting!
Maybe one could try having two different models play together, to see if they are genuinely good at the game or simply able to infer their own reasoning, if that makes sense.
I'm kinda bad at word games like codenames, even in my native language (french). With carrot and ray, I'd try something like "striation"? But it's really convoluted.
Re: OpenAI's o1 Playing Codenames
#119A small weakness in this test is that one of the keys to strategic Codenames play is understanding your partner. You're not just trying to connect the words, you're trying to connect them in a way that will be obvious to your partner. As a computing analogy: you're trying to serialize a few cards in a way that will be deserializable by the other player. This test pairs o1 with itself, which means the serializer is th…
thanks for the comment. I actually tried explicitly mentioning in the prompt that 'Your guesser follows the same reasoning process'. But this did not make any clear improvements. Maybe I should've done more prompt engineering.
Nah, prompt engineering wouldn't have solved the fundamental issue, which is that the associations between ideas as stored in the weights will be the same between the two AI players, which makes it an easier game for them than for a human equivalent. It'd be like two copies of you playing on a team, having shared all the same experiences right up until the moment the game starts.
And don't get me wrong, it's still a fun experiment! It's just that that 4 would never have worked if a human played against another human—there are simply too many other words that would be equally strongly associated:
* Gum: Gum is often wrapped in paper, so 'GUM' is strongly associated with the word 'PAPER'.
* King: King is a type of face card, which are printed on paper, so 'KING' is strongly associated with the word 'PAPER'. (Repeat for JACK.)
* Light: Paper is a lightweight material.
That's 4 others right there that are at least as closely connected in my head as LAWYER or LOG. The only reason why o1 pulled up the same four when guessing as it did when clueing is that it's the same model.
Again, I didn't mean this as a knock, just a warning about drawing too many conclusions from the test!
Re: OpenAI's o1 Playing Codenames
#120Earlier quoted context omitted.
I don't think this sort of communication from guessers to clue giver is in the spirit of the game (at least in my play group). However, inflating later clues is a reasonable approach! It's just that I don't think you're allowed to communicate the amount of inflation. Guessers must determine whether people 5 has slack to allow additional guesses on previous clues.
You're free to add additional prohibitions on communication as a house rule I guess, but the only prohibition in the rule book I've seen is that the clue giver's speech must consist exclusively of clues (and private consultation with the other clue giver). The clue giver is free to adjust their clue in reaction to anything they hear, and guessers can speak freely. Important: the clue giver cannot acknowledge the inst…
I played in a Codenames tournament at CGE's stand at GenCon, and they forbid guessers from communicating at all. Officially, its supposed to be just the clue and number and nothing else.
Of course, I never play this way in my own games