Raw LLM Responses

There is one thing that people don't usually understand. What we currently have as "AI" APPEARS very smart, while in fact it is not. It is leveraging the combined knowledge of humans into one powerful tool. For example Wikipedia is made by humans. If you feed wikipedia into an LLM and ask it about stuff, it WILL answer with Wikipedia data. It won't find new things. It is not thinking. How it basically works is, that it tries to take the input you provide ( e.g. "what is DNA?") and "guess" the result character by character via a neural net (which is basically a weighted statistical mapping). So it will run your input through that weighted network and because your input mentions DNA and explain, it will "guess" the characters to output based on the weighted neurons. Or to cut it short, the current iteration is an amazing tool to search through a massive pool of data, but it will NEVER find answers we humans haven't encountered. Everything it produces must be in the circle of it's training data. It may combine several sources and elements of the trained data to produce a new piece of text. But you couldn't ask it to come up with creative things that didn't exist yet, because it wasnt trained with it. However, there are other kinds of AI, that are not LLMs. Those are actually used to produce new protein structures, medicine, molecules etc. But thoe are trained differently and - as per my knowledge right now - not on the same tech as the chatGPT craze. If you have more detailed insight, feel free to share. Maybe I'm not up to date, maybe I got something wrong. Please contribute ❤