Description
Describe your feature request
Proposal
I propose a feature that, instead of displaying the raw reasoning content of the model, shows a real-time, summarized, step-by-step reasoning content as the model performs its inference.
Detailed Description
Currently, LLMs undergo a complex internal reasoning process to answer user queries. The content generated during this process can sometimes be unnecessarily long or convoluted, leading to confusion for the user.
This proposed feature aims to address this by not directly exposing the raw reasoning content. Instead, it would generate and display summarized paragraphs of the core content at each step in real-time. This is similar to how models like Gemini or ChatGPT succinctly summarize the essence of their current task in a short paragraph when performing complex operations.
Key Benefits
- Improved User Experience: Users do not need to wade through the model's complex and verbose "thought content." Instead, they can easily and clearly understand what task is currently being performed and which stage it has reached.
- Enhanced Transparency of Reasoning Content: By not completely hiding the model's operation but showing summarized steps, users can trust the logical flow by which the model arrives at its conclusions.
- Increased Readability of Result Derivation: This makes the waiting period for the final answer less tedious and provides a new experience where users can follow the problem-solving process step by step.
Distinction from Existing Functionality:
While OpenReasoningResults currently offers simple, single-line summary sentences, this proposed feature goes further. It focuses on progressively displaying summarized paragraphs that represent meaningful units of progress. This allows users to understand the specific progress of each step in much richer detail.