Forced Unlimited Guessing Codenames
It's codenames, and the clueing is normal.
But your team has to keep guessing until they win, hit the assassin, or hit a
Titan - 7
Nevada - 4
combat - 3
[Strike]protest - unlimited[/strike]
payment - 3
Festival - 5
Religion - 3
Outback - 3
Film - 2
Alien
Cap
Draft
Hospital
Mount
Sock
Straw
satellite
Mammoth
Mercury
Ship
Check
Charge
Grace
Cast
ball
Casino
Bill
Church
Australia
March
Play
Mexico
Match
Knight
Good job on Cast still, I autocorrected to castle
Omg
WHAT THE **** IS HAPPENING
Ngl for got this was me how long its been going
Extraterrestrial - 1
LOL our AI
alien most likely
Titan - 7
Nevada - 4
combat - 3
[Strike]protest - unlimited[/strike]
payment - 3
Festival - 5
Religion - 3
Outback - 3
Film - 2
Alien
Cap
Draft
Hospital
Mount
Sock
Straw
satellite
Mammoth
Mercury
Ship
Check
Charge
Grace
Cast
ball
Casino
Bill
Church
Australia
March
Play
Mexico
Match
Knight
List for new page. Gonna need to think on this for a few days.
Check
Satellite
Cap
Mercury
Charge
Mammoth
Cast
Grace
Ship
Bill
Australia
Play
Casino
March
Church
Alien
Ball
Hospital
Mexico
Knight
Straw
Draft
Mount
Match
Sock
So in previous games I'd work with AI in a potential angly way.
I'd ask for a range of ideas for clues and then pick holes in some terrible ones before asking for a suggestion.
It would still sometimes suggest an awful clue I'd already criticised but usually the worst were avoided.
This game I decided to go more in the spirit of the game and was one and done.
"Here's the codenames list, can you suggest me an aggressive clue" "Titan 7" (I forget all 7 it thought were connected).
"Here's thr codenames list, can you give me a clue for 3 red words" "combat - 3" (Grace, Cap and Charge I think.)
Next payment was check, charge and cap.
Religion 3 was Grace, Cast and Cap for a religious hat.
Film 2 was Cast and Cap. Cap was specifically meant to link to Captain America.
How do others use AI? One direct question or a bit of back and forth before asking for an official suggestion?
How do others use AI? One direct question or a bit of back and forth before asking for an official suggestion?
I just use one direct question, but my AI doesn't seem to be quite as dumb as yours. (Though it's had some moments)
Main reason I went play > alien at the end is because I thought alien was theirs the whole time because of Nevada/Area 51. Turns out I was right about that but our word was Cap? What the hell?
well he pulled a double-reverse fakeout with play
he eliminated it being ours by doing n-1 instead of n-2, so we'd never guess it but you thought it was safe from being ours
I think most likely he forgot he could do n-2 in this variant since we can always guess unlimited in this variant
So in previous games I'd work with AI in a potential angly way.I'd ask for a range of ideas for clues and then pick holes in some terrible ones before asking for a suggestion.It would still sometimes suggest an awful clue I'd already criticised but usually the worst were avoided.This game I decided to go more in the spirit of the game and was one and done. "Here's the codename
I have a huge prompt, but I always only ask for one clue tied to a specific number, which is what I thought everyone else was doing.
Rules:
- You must ask the AI chatbot of your choice/availability for a Codenames clue.
- In your question, you may include any number of your wordsand may also ask to exclude any number of words.
- Your clue must be the first clue the AI suggests (no asking for another clue, or changing the input words to try to get something better), and your number must be the number of words you asked it to give a clue for. You must use it even if the suggestion is illegal.
- Zero clues are allowed, but only if you asked the AI only for a clue that excludes words. Unlimited clues are not allowed.
- Teams must guess at least once, but are not restricted by the number of the current clue.
ok, so we can't do unlimited clues (even if list is correct) so must use a number (including 0 if we want) but the guessers do get unlimited guesses...so we get a bit more info than normal unlimited guessing codenames since we get a number but also get unlimited guesses...we'll see if it's too easy or too hard with a few rounds
Maybe I’m misunderstanding what you’re saying, but “one and done” is the rule. If you’ve been getting multiple clues and picking them apart and then asking AI for the best one, that doesn’t seem legal. If the angle is “well, I’m not technically asking for a suggestion, I’m just asking for possible clues,” then I think that’s way outside the spirit of the game and shouldn’t happen anymore.
yeah; I give it the list of words and have it repeat back to me so I know it knows to avoid assassin and opponent words
I also gave it instructions that it saved to memory:
Now in this game I have to ask you to provide a clue and I MUST go with your first clue option so be very sure that the first clue you give me each turn is solid and doesn't hit the danger words. I'm not as worried about neutrals because they'll make a list and make them up the next turn if they hit a neutral, but hitting assassin is death and hitting a blue word is very harmful.
then I say something like:
"let's start with a 4 clue"
Our team did guess fence first, BUT their list they left is awesome! So a 0 clue can get us 6 words this turn! Here is their list. All are our words except for Port. So let's give a one-word clue that very specifically hits JUST port on this list and no other word. Something obviously meant to be cluing just port as a 0 clue so that they'll guess the other 6 words on this list: File Port Compound Fighter Hospital Millionaire Robot
they got all 6; last word to clue is PIN. Let's give them a good clue that hits Pin very hard to finish this
let's go with 5 word clue connecting Hollywood, watch, laser, fan, light
let's do a 1 clue for paste
ok we just need a 2 clue for rose+deck that doesn't hit trip, kiwi, table, court
I'd like to target to win in 3 turns so they are pressured to win in only 2. So pick a good grouping of red words for a 4 clue and let's start with that.
Let’s do a 5 clue for remaining words so they know suit and date aren’t theirs. Just make sure it’s super safe and won’t have them guess the assassin or the remaining blue words: compound, date, suit, witch.
then I add in extra stuff as feedback to the clues they gave to try to make it perform better in the future; I let it know how we think about connections so it gives better connections or when it gives dangerous clues to help it in the future
and also about capitalization:
remember when giving clues that it should be lower case clue unless capitalization matters (like if capitalized for part of proper noun or reference to movie or company or product or something similar)
I have a huge prompt, but I always only ask for one clue tied to a specific number, which is what I thought everyone else was doing.Maybe Iβm misunderstanding what youβre saying, but βone and doneβ is the rule. If youβve been getting multiple clues and picking them apart and then asking AI for the best one, that doesnβt seem legal. If the angle is βwell, Iβm not technically ask
Yeah, that's kinda been happening. Along the lines of I'd ask for a clue and it'd just talk about connections without giving a clue. Then ask me a question. I'd answer his question and also comment on the connections it made. In hindsight I should have just repeated the request to give me a clue.
Sorry for breaking rules and if that's spoiled anyone's fun.
Why is my AI so bad at codenames? I thought it was all the same AI?
Yeah, that's kinda been happening. Along the lines of I'd ask for a clue and it'd just talk about connections without giving a clue. Then ask me a question. I'd answer his question and also comment on the connections it made. In hindsight I should have just repeated the request to give me a clue.Sorry for breaking rules and if that's spoiled anyone's fun.Why is my AI so bad at
All good! Weβre figuring out this shittastic variant as we go, haha.
then I add in extra stuff as feedback to the clues they gave to try to make it perform better in the future; I let it know how we think about connections so it gives better connections or when it gives dangerous clues to help it in the futureand also about capitalization:remember when giving clues that it should be lower case clue unless capitalization matters (like if capitali
I donβt have an account with ChatGPT, and Gemini is useless (ime), so I just take to insulting it for not knowing how to follow simple prompts and likely losing my team the game. But itβll never learn because Iβm not logged in.
well he pulled a double-reverse fakeout with play
he eliminated it being ours by doing n-1 instead of n-2, so we'd never guess it but you thought it was safe from being ours
I think most likely he forgot he could do n-2 in this variant since we can always guess unlimited in this variant
Can you remind me about this cuz it went right over my head
If you have a backup list and 4 words left to guess then a 3 clue will indicate that the first word in the backup list is likely correct. A 4 clue will indicate that the first word in the backup list is wrong.
Since we can guess more than just the usual 1 more than the clue number then it opens up other possibilities. If there's a 2 clue for those 4 words then perhaps the next 2 on the backup list are correct or perhaps they've just got 2 words to clue and the backup list is irrelevant, but if a 3 clue is given then it's NOT going to be 2 words from the backup list.
Titan - 7
Nevada - 4
combat - 3
[Strike]protest - unlimited[/strike]
payment - 3
Festival - 5
Religion - 3
Alien
Australia
Cap
Cast
Charge
Draft
Grace
Hospital
March
Mount
Play
Sock
Straw
satellite
Mammoth
Mercury
Ship
Check
ball
Casino
Bill
Church
Mexico
Match
Knight
so here we had guessed knight/march/play last turn and knight was neutral so our list was:
march
play
we had 4 words so needed 4 more
if you clue 4, we know march is wrong
if you clue 3, we know march is ours, play isn't, as we can't win by getting just 3 words so we would guess top of list
if you clue 2, we know march+play are likely ours; we can't win by just getting 2 so we would guess 2 then go top 2 of list
so a 2 clue for Australia+Alien (immigrant? etc) gets us to guess those 2 words and then march and play and win one turn sooner
by same logic, you doing a 3 clue confirms march is ours and denies play is ours, so we are not guessing play because you told us it wasn't
Ok but I thought you could only guess 3 if I clued 2?
Not in this game