Sheet summary:
🔺️Το AI-generated content is starting to fill the web and this could be bad news for future AI models.
🔺️Τα language models like ChatGPT are trained using content found on the web, and as AI creates more «synthetic» content, it could create an engineering problem known as «model collapse».
🔺️Το filtering synthetic data from training models is becoming an important area of research and will likely grow as AI content begins to fill the Internet.
The sheet:
The uroboros is the famous ancient symbol of a snake devouring its tail. As content generated by AI language models such as ChatGPT begins to fill the web - often to the frustration of the very human editors who work on these sites or in creative writing-related professions - many results of AI's activity are accompanied by many bugs.
And this is a big problem, because the Internet is itself the source material on which these language models are trained, based on algorithms invented by mathematicians in the 70s and 80s (Petri-nets by Soviets originally and neural networks later).
In other words, AI, which only found the opportunity to hatch when a lot of data was illegally over-aggregated by web2.0 companies like Facebook and Google, once hatched, has already started to eat its tail.
This can best be described by continuous photocopies on the same photocopy. In the end, the original content is completely altered., In the same way, artificial intelligence could begin to train on synthetic data full of errors until the very thing it was trying to create became total gibberish. This is what AI researchers call «model collapse».
Experts in this branch of computing warn that artificial intelligence/intelligence could cause the extinction of humans. This is not certain, however, because AI is evolving uncontrollably and without human intervention much faster than the human species and the societies it has constructed. The danger is mainly to be found in certain professions that are in real danger of disappearing. Such as real journalists (the parrots, like most of ours, who only relay propaganda and whatever the APE says, have already disappeared; but they don't know it...) or programmers, notaries or bureaucrats in public administration.
And of course, as with any other human invention, the danger lies in its misuse, such as that which is now being done to violate any notion of intellectual property and personal data, to develop the techno-feudalism of the bigtech internet in place of the mechanisms of neoliberalism. Which has also collapsed along with its famous “invisible hand” (which would have regulated everything, to its horror).
Because of this problem, they are now looking for ways to prevent AI from using internet data generated by AI programs (the problem is equivalent to detecting fake news that is rapidly replicated on the internet due to human stupidity). And they come back to what has been known since the 80s:
✔️Πάντα will need human intelligence for artificial intelligence to evolve. Without the former, AI will collapse!
But human intelligence, which really produces new knowledge and progress, relies on the function of intuition (and then proof), in short, on what we call science.But no machine can create artificial intuition for a simple reason:
Intuition operates in chaotic patterns in the human mind, and is the result of our evolution in knowledge-sharing societies. But no machine can operate as chaotically but only in a linear fashion. In creative chaos (which they are still very much afraid of, as well as trusting their intuition), all machines break down.
What don't you understand about making it a penny in the comments?
After my good morning
