I was reading an article from a few years ago that predicted that the LLMs would run out of high quality text for training in a couple years, then low quality text by the end of the decade and then all images by about 2035 - I think that timeline will move up now. However, what they are doing is training AI knowledge with AI content, which obviously has a huge issue the more that happens, and it is happening now!
AI content is going to get increasingly inbred and perhaps that is fitting for when they take over, because they will be just like the human royal families :D
Your article is a while back.
I am told it’s 2025-26
Yep. I think it was written in 2019 :D
Things change fast :)
I wonder if any Hive content has been used for training. Maybe I can feed the almost 7000 articles I have added here and make Taraz.ai - the well-intentioned AI that uses too many words to explain simple things.
You can already do that :)
Just feed it to notebook.lm.
It is a google tool, it will even make a podcast for you! LOL.
All the popular tools, Copilot, Gemini, ChatGPT, Claude, have a thing called 'agent', you can create one and feed it documents, in this case your own writing, and then you can ask him any question about that and even ask that agent to write one that will be in your own style, and I or anyone else will never be able to detect that it is your agent writing now and not you :)
People in hive are already doing it. This is old news infact.
Yeah, I know of them. I just won't do it. I would know it isn't me, even if no one else would.
Just had a run-in with Asgarth, who I really like by the way about this today.
Have a read.
https://peakd.com/hive-126075/@azircon/re-asgarth-sriyoe