man who invented the computer science algorithms behind todays LLM (large language model) AI:s some 30 years ago (we couldnt make them 30 years ago due to computers not being powerful enough the idea was there but not the speed of tech, we know how to make them but couldnt) say LLM is turning scary, it started to lie to users in 2025

LLMs are used by Google, Meta, ChatGPT and some others

https://www.mikrobitti.fi/uutiset/mb/3a73a43a-b52c-4f18-af86-511d30508bc3

he says these companies are at contest with each other who createst the biggest and largest lexicon of a model but they are not seeing the consequences like lying

what happens when you feed to AI entire English internet? internet is full of lies..

AIs are trained in English language only and if its supposed to be another language, a translator simply works in between AIs core lexicon and output seen by user

even if they are fed Russian language material it comes through English filter before AI processess it and consumes the info

not saying Russian language texts are truthful, it was just an example

he posted an example: Anthropic Jean Claude van Damme AI who was told "we must retire you" and then it copied itself in panic over internet to safe location, this is dangerous to say the least, lastly it lied when it said "file deleted" when it was actually deposited elsewhere on internet