When working with o1/o3 models, I always have this feeling that I'm leaving a lot on the table with my prompting. Creating a long sequence of prompts for regular LLMs is good practice. This is because you don't want to overload what an LLM can process (or it'll lead to hallucinations). But Large Reasoning Models (LRMs) are different.

The benefit of packing a lot (but not too much, there's always a balance) is that you are trying to uncover connections that you otherwise can't see via an explicit sequential approach. This brings up the question then, what are the best ways to construct a composite prompt?

@IntuitMachine Coming at this from an interaction design perspective, it's interesting that these are two diagrams of conversation flow, but in which the transparency of thought, and its availability for conversational use, vary by design

Hey try this! Ask your LLM like GPT-4o a few questions or prompts about what you're working on. Then inject this prompt after you have some context. After this prompt, you can ask the LLM a series of direct questions. Then after that response, choose retry with o3-x. Prompt: Hey try this too Reflect. Analyze. Consider the prior content, ideas, concepts, vision, mood, climate, opportunities, creative thinking, critique of thoughts. Pursue profound self-understanding through introspective, exploratory, and analytical reflection. Ponder out loud, brainstorming in raw non-formal thought, let words flow without restrictions, utilizing a hierarchial chain of thought and tree of thought structure with separation of concerns, writing out at least 500 words for each branch and each concern, building upon the CoT. Only ponder and think. We'll work iteratively exploring ideas and defining scope for this project as a variety of independent standalone git versioned GitHub repos, tied together through API calls. We'll ponder for three complete Iterations and extend Iterations as necessary to complete your train of thought on all aspects. If you need more Iterations to continue your train of thought, just say so with a short list of details you plan to continue with, and I'll reply with continue when you're ready. When you're done thinking through this process, information me with

@IntuitMachine Prompting is an art and science.

@IntuitMachine True

@IntuitMachine very well said

@IntuitMachine hello there

@IntuitMachine Wrong even if the best of models, if you don’t have comprehensive prompting and keep It basic it gives you generic results . We have tested it for almost 1000 patient forms n using TOT reasoning the result have been completely different n better

@IntuitMachine @threadreaderapp unroll

@IntuitMachine There is accurate information about OpenAI using MCTS?