Raw LLM Responses

Lambda is a specific AI architecture, called "transformer". That is basically your phone's autocorrect on steroids. If you sit just clicking the next word in autocorrect, your phone will generate a sentence. It is not a very good one, but it will generate a list of vaguely related words. A transformer is roughly the same but much, much bigger. It has a much larger database than your phone (trained on a substantial portion of the whole Internet) and it looks at much more context than the one or two words your phone does (it looks about 2000 words backwards). That context includes everything said, both questions and answers. The AI takes the context with your latest question as the last part and generates it's response, one word at a time, just like when you click the next word of autocorrect over and over. The main point is that it has no other memory. It only has 2000 words of context and a fixed database containing statistical data for probabilities of what word should come next. By just putting suitable stuff into the context, you can get the AI to say anything. It isn't intelligent, any more than autocorrect is.