to improve visibility in AI search […], which in the end deteriorates LLM's outputs (Pay to Win) Happened to Google Search too though, right? And it has only gotten worse with time.
We can only hope that AI goes down another road – or that AdBlock works still better on them! (To be fair though, my only real ad problem on the whole internet is YouTube, where AdBlock partially stopped working a year back or so. That's a pain. And sometimes I see them on Google.com, but that's not nearly as bad as on YT. Curiously, this site is the only other site that gets around AdBlock for me! ! Otherwise, I hardly ever see ads. GET ADBLOCK good people!)
It is useful to note that the basic training data of general-purpose LLM's like GPT are more or less "everything that has been written" – it is trained on data equivalent to a couple of billion books, which makes up its underlying knowledge.
But of course there will be biases, mainly because of human-induced censorship/leadership choices &/ training.
I work with "training" AI (American) LLM models such as GPT through what is called RLHF (Reinforced Learning with Human Feedback). We "penalize" it for using profanity, saying offensive things (e.g. racism) etc. (We do not train it not to talk about political leaders though – at least not yet... China's DeepSeek, on the other hand, has a filter against critical questions about Chinese gov't.
Then of course ads will come. But GPT don't have them yet, and even the ad giants Google and Meta mostly still only have ads around their AI models – not in the models themselves. |
