Raw LLM Responses

Transformer models are purely stochastic. They attempt to produce text that is statistically similar to human writing. They DON"T THINK! Of course they make stuff up, that is part of how they work. I have a PhD with a focus in AI and have worked in the field for about 30 years. I have worked in computational linguistics for almost 10 years. I know what LLMs can and can't do. This is about the 4th time that I've seen a new AI technology over glorified by the general public because is LOOKED impressive, only to have it misused. When people realize that it is not all that it appears, they abandon the technology and it takes that much more effort to convice people of its legitimate uses. This will set the field of AI back years as legitimate advaces in AI will be disregarded or considered too dangerous to use. Unfortunately, Transformer models of this type are becoming the basis for search engines at Microsoft, Google and others. They are NOT proper search engines and inevitably will be misused and cause major problems for the people using them. Their proper application in search is to interpret plain english search querries to optimize the subsequent search by better spanning the semantic space of the querry, filtering relevant results, and formating search results in a more digestable form, NOT interpreting the results, (which is what people will ask it to do.) If people trust ChatGPT or any LLM to interpret search results, they are screwed. It makes me cringe that the same kind of technology has been trusted to driver cars on public roads and have killed people because they DON'T THINK and if they see a simulus they were not trained for, they tend to ignore it or misinterpret it in ways no human ever would. (Heavy sigh) The real problem is not with artificial intelligence, it's with the lack of organic intelligence on the part of the people using it. I sometimes joke that I work in artificial intelligence because there's not enough of the real stuff around. This just proves my point.