I built a tool to analyze ChatGPT's "Thinking" and JSON code to uncover the topical map of its brain.

Interesting idea — probing the internal structure of ChatGPT’s outputs and JSON representations can definitely help uncover latent patterns, but in real engineering practice the hard part is not just mapping topics, but validating that those maps actually reflect meaningful internal behavior versus artifact patterns in the generated text.

From an engineering standpoint, one of the key signals I’d watch when analyzing AI outputs is consistency under perturbation — for example, whether slight prompt rewrites or schema variations lead to stable topic clusters versus chaotic shifts.

Curious — in your tool, what validation signal or metric do you treat as the strongest indicator that the topical map is meaningful and not just an artifact of the prompt/decoder patterns (e.g., cluster stability, semantic similarity distance thresholds, or repeatability across runs)? That’s often the difference between a cool analysis and something you can build processes on.

I haven’t used it yet, but I just took a look and it really seems like it’ll work. Looks way better than most of the tools I’ve come across. Nice job!