2:46:44 of this video:
"Your shit art is good now because our standards are beyond dogshit and we let computers live rent free" is NOT the own you think this is.
To add, AI slop has a worse rap because the vast majority of people are not creative. Just like how a lot of art is mediocre, AI art is even more so because it eliminates the time sink filter. To add to that, a lot of people also are not good at writing and translating their ideas into a succinct way for a model to understand, let alone the learning the technical knowhow for complex workflows. You still need a talented person behind the tool.
I'm catching up and necroposting, but the fact they left out Google shows how hollow the movement is.
Have you seen how buggy the code they've released is? E. g. the obvious bugs in their CLIP tokenizer (there are public bug reports on their repos). Not to mention their "pissfilter" which looks suspiciously like someone forgot about the default avg/norm constants for imagenet and didn't adjust for their own data.
Reminds me of cringe tranny slacktivism and how people draw their favorite cartoon characters beating up ICE.
I linked full articles on nondeterminism. I also provided toy examples of nondeterminism for people who didn't understand how it's possible. The only things I got wrong were assuming people in an AI thread would 1) be knowledgeable about how it works, 2) want to discuss it instead of treating my post as a personal attack.
Okay, but imagine if the llm talks like a valley girl brat.For some reason, that way LLM writes or talks -- especially talks -- is grating to me. Not as bad as "Chris Chan" derping, but still irritating. Whether it's "Grokipedia" or "ChatGPT". And I don't like how LLMs try (and fail) to sound human. I dunno how so many like using LLMs all the time, much less that trying to do that AI "boyfriend" and "AI girlfriend" thing. Like I said, it'd be much better if LLM talked like stereotypical 20th century scifi robots. Like the whole "AFFIRMATIVE", "DOES NOT COMPUTE", etc. thing.
This is why you introduce the randomness at the seed stage, such as by using radioactive decay and then feeding what you get into Mersenne Twister or whatever they're currently using. If you need a lot of "randomness," you can't practically generate it by such means, but with a random enough series of seeds and no way of knowing what they are, it's effectively random from your perspective, but there's a reason they call them PRNGs.
That's the cheeseball way of doing it. For something approaching true randomness, you need something like this:
That's one thing I like about Gemini is they introduced permanent instructions pretty early so I was able to tell it to be straightforward and professional and be extra direct and to the point if it detects I am frustrated (mainly because I start calling it slurs).For some reason, that way LLM writes or talks -- especially talks -- is grating to me. Not as bad as "Chris Chan" derping, but still irritating. Whether it's "Grokipedia" or "ChatGPT". And I don't like how LLMs try (and fail) to sound human. I dunno how so many like using LLMs all the time, much less that trying to do that AI "boyfriend" and "AI girlfriend" thing. Like I said, it'd be much better if LLM talked like stereotypical 20th century scifi robots. Like the whole "AFFIRMATIVE", "DOES NOT COMPUTE", etc. thing.
