ai content will always be terrible because it is by design low information content. in choosing the most heavily weighted outputs from a prediction machine, even if we crank up the temperature we're going to choose ultimately predictable outputs, and predictable outputs necessarily have lower information content. if i have a coin that lands on heads 90% of the time, and i flip it and tell you it got heads, i haven't given you much information because you were already expecting that. if i told you it landed on tails, however, i would be giving you a lot of information. unlike humans who write by turning reality models we have built up from experience with the world into information-dense language, ml architectures that are ultimately prediction-based like llms will always choose the most generic, most soulless solution to every problem. the technology has gotten and will continue to get better at producing outputs that are at least within spec and not obviously broken, but it cant create like a human, particularly not in artistic pursuits.in other words are ai engineers fucking stupid?
>>109578525if a coin lands on heads 99.9999999999999999%does it matter
>>109578611its an example to demonstrate an information theory concept in an intuitive way.
>>109578525and yet that image is more unpredictable than 99% of art
>>109578733the subject matter is up to the prompt, the model's execution is its own story and the obvious problem with picrel. you can't seriously think that image is passable?and have you never read llm text? is it not obvious how it's just not dense? it leaves you hungry, it's like it's always skirting around saying anything of meaning or substance. it's beyond obnoxious.
>>109578733le wacky animals is the most creative bankrupt example of creativity.
