The human brain takes 12W to run. The weakest useful Large Language Model takes 400W to run. It may be "better than a markov chain" but it's literally this fucking meme

@sevvie Yeah but human resources doesn’t have free, unlimited power to scramble the human’s brain to ensure it doesn’t say politically inconvenient things, duh

@WashedOutGundamPilot @sevvie Humans also don't have instant perfect recall of every piece of information we've ever encountered

@monsterislandcolonizer @WashedOutGundamPilot This is one of the reasons I want to demystify AI: neither does ChatGPT. It only has perfect recall of the data on which it was trained, in the shape it decided upon at the end of training iterations. In order for AI to have access to information, it must be provided that information as part of its prompt, or as an embedding which is just a pre-vectorized string. It takes a fuckton of real work to get to the point where an LLM can remember accurately and not hallucinate.

@sevvie @WashedOutGundamPilot @monsterislandcolonizer
In my (very limited) experience AI is incredibly inconsistent. You can ask it the same question in a slightly different way and get a completely different output. Of course this is because it doesn't actually understand anything, it just tries to give an output that looks like it should come from the imput based on whatever training data is has. It's just the simplest form of pastern recognition and extrapolation.

@Reluctant_Weeb @sevvie @monsterislandcolonizer @WashedOutGundamPilot I couldn't get it to write a story without remembering the previous parts.