I’ve been experimenting with Gpt-3 (and derivatives like Gpt-neox). I’m trying to run them on my home machine which has only 6GB of memory on the GPU. It isn’t going well.
So I switched to the playground at Open ai. This is fine, but I want a model that always gives me an answer, no matter how bad it is. I want a narcissistic model that doesn’t care if it is wrong but always responds with something that looks like a reasonable answer but isn’t burdened by facts. OpenAI will say unknown if I’m using Q&A mode and it didn’t know the correct answer.
Is there a model that is poorly done that always gives back an answer? I’ve got a good use for that one, and great if that doesn’t take much memory since it is really stupid.
It is difficult to search for this because, strangely, no authors seem to want to publish their work if they know it does a bad job