#Play cards and chat with GPT bots

7 messages · Page 1 of 1 (latest)

green notch
#

Bots are set up to remember all their opponents, although they do not reflect this quite yet, they are already saving histories. Game state is fed to GPT for conversation, but bot game decisions are still made by hardcoding... for now.

#

Only the main public card table gets a GPT bot currently, other tables are for pvp.

finite arrow
#

So do you want to make hardcoded parts for decision making GPT-based?

green notch
# finite arrow So do you want to make hardcoded parts for decision making GPT-based?

possibly.
describe the game rules, give the game state, and get a response. Make an "ingester", something that can translate a range of GPT responses back to my API.
Obviously it's not trained for my heretofore non-existent game, so it's an odd thing to do, and would depend, more than anything, on how accurately my prompts encapsulated it's current state

finite arrow
#

Have you solved the problem>

green notch
# finite arrow Have you solved the problem>

I focused on 'ingesting' normal chats first, deciding when to reply, in what way, etc.
And it works well except now the API is taking minutes to send a single response, so ... no dice.
as for the GPT-powered game decisions.. I'm sure I could do it, but I'm not sure it woulud be that interesting. Because the difference between a "GPT decision" and a normal algorithmic one would be imperceptible to the end user