The post describes a prompt that randomly selects either "dump" or "match" with a 30% probability for "dump" and 70% probability for "match". The comments discuss the use case for this prompt, with community members questioning why the random module in Python couldn't be used instead. The community members also suggest trying different prompts and temperature settings to achieve the desired distribution of "dump" and "match" responses. However, there is no explicitly marked answer in the provided information.
Please respond by randomly selecting either only "dump" or "match" for each interaction, with a 30% probability for "dump" and a 70% probability for "match".
I watched the video, but I’m afraid I still don’t understand the usecase nor why using the random module wouldn’t accomplish your initial ask. Can you please clarify?
I see. I still think you should have them clarify why they want that. If I was going to try to do something like this with LLMs I would try a few different prompts like “generate match most of the time and dump sometimes” and then play with the temp and do some trials to see what temp gives you about the distribution you’re looking for. In any case, I would only do something like this just to see if it’s possible, not to rely on it as a feature.
