using an amount of technical words is important to get more awe from the people, so they do it, but does not mean nothing
to understand the actual fail wave which is hitting all AIs, here is a simple explanation
how we learn from beginning is by listening and recognizing patterns, then we start to understand, then we try to replicate to express what we understood
that is pattern recognition
it also means, we later learn hot is a word and it is a relative value for temperature, but before we learn patterns with the word hot, like the water is hot, it is hot under the sun or similar expressions
red, green, blue and transparent are words, not patterns, if we treat each as absolute color pattern we may misunderstand the meaning when used in a phrase, transparent isn’t always just a color
in order to understand what somebody said we need to recognize how he said it, so taht is pattern recognition in its most authentic form
now the trick, AI does not use word, they say token, that is by intent to make them look smarter then they are …
so the search the internet for red, green, blue and transparent tokens and count each find, then they tell you that red is the preferred color
that might work in some general cases, but mostly yields distorted results because AI didn’t validate the context
and that is what is hitting AI very hard, the back squatter of changing pattern recognition to token count what happened ends of 2025
means a trivial question about choosing a color stick for a child might come out wild because AI didn’t understand the sense of your question and renders only the token count and might even ignore the color because child rendered much higher numbers and tells you to use a stick only for an older child or something …
since the change, this first 5 month of 2026 AI became a nuts case, it got poisoned by their own wrong answers which are immediately used as source, and when the result was shared, the effect is even faster and deadlier
now you just need to multiply the effect on technical question, programming, architecture, engineering
… and just wait until your building comes down …
as I already wrote about before and shared AI self-positioning about the issue: it might already be too late to fix the problem …
RIP AI ? we see that soon, but until then a lot of catastrophic things may happen, like companies break, people loose their jobs, entrepreneurs and freelancer loose their clients …
Leave a Reply