rotates him

stolen post but i needed to repost it here because this is the funniest shit of all time

i am unbelievably furious

For the equation to be accurate, it must be an admission that AI adds nothing.

We sometimes talk about web scraping projects as "ingesting" or "slurping" text from the web but usually we understand that nothing is being actually consumed, the text is still there where it should be. But now it's different.

Large Language Models (LLMs), of which the current most famous is ChatGPT, are eating the web, actually consuming and destroying it.

• Google: overrun by fake sites with unhelpful content generated by LLMs. They have, bafflingly, decided to counter with their own, first party unhelpful content generated by LLMs.

• Reddit: despite its community, it was previously a refuge of meaningful, open, human-written text on topics people care about. Now it's mostly down, because they betrayed their community with drastic changes and the community revolted. One reason given for the drastic changes: their tasty meaningful text was being exploited by Google and OpenAI and their ilk, and they weren't getting any benefit from it. So they decided to charge a gazillion dollars for API access, causing everything else the community built that uses the API, such as accessibility tools and usable interfaces, to go down.

• Twitter: bought by a complete dipshit who (among his many foolish ideas) thought he could save money by replacing workers and essential site features with AI, so essential site features are being replaced with nothing. Also decided "API access and third-party apps should cost a gazillion dollars so LLMs will pay us" before Reddit did.

• Stack Overflow: Rampant use of ChatGPT threatens to turn its questions and answers into nonsense, much like the questions and answers you see on Google. Moderators responded by moderating even more harshly than they usually do. Site owners saw that a declining number of people want to even attempt to use Stack Overflow, and panicked, and told moderators to stop banning people for using ChatGPT, instead of addressing any of the other reasons people don't want to use Stack Overflow. Moderators went on strike. The site and its siblings are now mostly unmoderated, and, confusingly, still working for now.

• Wikipedia: apparently standing strong for now, but their standards of information are threatened as formerly "reliable sources" start generating nonsense with LLMs. Wikipedia is particularly vulnerable, because if LLM output is ever treated as a reliable source, it can create self-reinforcing fake facts that people repeat because they're on Wikipedia.

• Many independent websites: buried under competing LLM nonsense, or bought out by venture capitalists who fire their staff and replace them with LLM nonsense. (I just saw a hecking GeoGuessr tips page destroyed by someone who copied all the work people put into it, pasted it onto a ChatGPT-generated website, and took credit for it.)

They have found a way to scrape the web so hard that there isn't any web there anymore.
I don't know what we can do about it. I hope that a specifically anti-metrics, anti-capitalist website like Cohost can be a refuge, but there's not enough Cohosts.

I'm at the point of considering tracking down and joining niche forums every time I have an issue. Unfortunately, this would come with the difficulty of finding them when search engines aren't usable. Reddit works for most things for now, but the way things are going I'm getting concerned about how long that'll last.

be nice.
he's just a little guy trying his bess.

here's the short version

• when lenses get small enough, diffractive optics comes into play
• this means "image get blurry" and some notion of maximum rendering power
• unless you strap a telescope to a phone, you'll never get a clear picture of the moon
• you cannot "zoom, enhance"

enter computational photography

• let's just throw some content-aware-fill slash stable diffusion to it
• or in simpler terms: run autocomplete and fill in the details
• if you draw a happy face on a blurry moon jpeg, you get a happy face high rez moon pic
• you actually can "zoom, enhance"

sure enough, some people are 100% ok with "make my photo better" technology, they want to just get "more bokeh cream" and smear it across a jpeg.

on the other hand: the people who buy very big telescopes to take photos of the moon are a bit miffed, and the people who spend hours processing raw photographs for tone are also upset, and it's understandable.

who wants to live through a tsunami of low effort garbage drowning out any and all creative works?

it could be worse: at least we have this other algorithm to predict which of the garbage that will gamify the experience of consumption

anyway, aside from the "we are about to witness the endless september of machine generated content" stuff, what strikes me about the computational photography bit is that eventually, your phone will hallucinate your friends faces.

what was once a blurry underexposed group selfie is now a perfectly crisp, well lit image, and maybe it'll randomly insert a family member into the background because that blurry head looks just like your uncle