>>512175231
yeah, but let me explain the concept again based on image gen
you can craft a very long, specific, detailed prompt, and test it again different models, and some results are going to be better than others, but in general, you get way too much variation within the model itself, so you would have to run 10 gens each to actually get a good sense of which one is best, and not just that, but there's quality variation within the same model as well, some are much more constant than others, adding even more complexity to the mix
so the best practice, imo, is to run extremely simple prompts, and just let the model reveal its own nature
then you can do two things, use it for what it's good for, and tweak that baseline into doing interesting things
this applies very well to local LLMs too, but it's a bit more difficult with big-dog LLMs because they tweak them constantly, so now I just do vibe-prompting most of the time, and pick the one that's giving me the right mindset, and it's been working really great, I haven't "argued" with an LLM in a very long time
>The Moldovan retard was even using a tool to tell him what posts were human or not. Imagine being that stupid
lmao, that shit was priceless
>>512175687
>>512175722
KEK