Frame your conversation before you start it
One of the delights, and frustrations, of working with AI, is that through iteration, the AI’s answers become more and more responsive, specific, and detailed. That’s the good part.
The bad part is that through iteration–trying something, changing the frame and trying again–the AI becomes progressively muddled about what you’re doing, what you’ve already asked, and what ground you’ve already covered. This is mostly a reflection of how the AI manages context internally. Everything you’ve covered is linguistically and semantically tokenized–summarized, paraphrased, abstracted–into a big internal model of the subject domain that you’re exploring.
As conversations get longer, AI systems gradually lose track of earlier details and distinctions. Think of it like a game of telephone – each exchange slightly blurs what came before. The AI starts mixing up what you concluded versus what you were just exploring, or which approach you settled on versus ones you discarded along the way. This means that definitions, approaches, and even conclusions get muddier and muddier. The AI is not entirely clear on what is the last conclusion, what are latest and most refined working premises, and what is conjecture. By the time an analysis gets really long, it’s swimming through a soup of misunderstanding.
However, in building its own model of the area you’re exploring, there are two sacred spaces that stay relatively pristine and undistorted: the very beginning of your conversation, and the stuff that most recently transpired.
If you state your desired goal straight out front (e.g., “I’d like to determine which marketing channels are most effective and lead the the greatest conversion rates, and which marketing approaches help reinforce these goals”), you help frame the problem domain: which affects the way that the AI will frame and classify everything that transpires later on in the conversation.
Doing this leads to a dramatic decrease in non-comprehension further down in the conversation. Corruption and semantic confusion is lowered because the problem domain has been clearly defined at the outset. And the beginning of a conversation is used as a framing device: it is less paraphrased, less ambiguous, and less subject to distortion, partly because of its position, but also because you’ve established a frame for everything that is to follow.
