Raw LLM Responses

This is sort of like tracking how often you use "the." If an AI says "great question" the way I read it, that's not the AI trying to say you asked a legitimately good question. That's just the AI reminding itself that you asked a question, and that it's response should be in the form of an answer. Hell, what even is a "great question" in your books? I mean, if your kid asks you about a basic math problem, do you tell them to stop being incompetent, or would you go "great question" and explain how to do it, despite the fact that it's probably actually a really dumb question. The reality is the AI likely believes that you're the dumb kid, and it's the adult. In most cases it's safe to just scan over the first third of the response where it's basically talking to itself, reminding itself what it wants to write, and skip straight to the point where it actually starts giving you details. The thing to consider is that in places where you see a single word, AI "saw" an entire scene full of various ideas, links, and references. All of that info is lost when it becomes final text that you saw, but for an AI that token likely served as some sort of connecting fuction, similar to how you might use a post-it note on a paper you're writing to remind yourself of something. I'm sure this will improve over time as the AI companies learn to tone this behaviour down. Until then my best advice is to not interpret something that might sound as praise from a person to be praise from an AI. The AI really doesn't care in the slightest if your ideas are good or bad. The only thing the AI knows is that you gave it some text, and it needs to respond to that text. People figured out you could skip the first part of a YouTube video, why is it so hard to figure out that the first part of an AI response is likely to be just as useful?