Post #4195704

it’s uhh notable that a large portion of computer science is fundamentally about being able to evaluate what a computer can practically do by knowing the efficiency of an algorithm, and meanwhile LLMs do everything they can to prevent evaluation of their efficiency at every possible level because they’re horrifyingly inefficient and hit a wall in terms of their actual capabilities many years ago if we had a full understanding of the algorithmic efficiency of LLMs, we would be extremely angry, but also extremely unlikely to use them for much of anything at all, as understanding the efficiency of an algorithm is part of how we understand its limitations. a system like an LLM is extremely and fundamentally limited by its inefficiency.