I asked ChatGPT to make a spiral galaxy of my research, from data scraped from Google Scholar, via asking for some cluster analysis or SNA from the abstracts, to help build the visualisation. The April 2025 o4-mini-high model felt like a brute force approach and produced some Frankenstein type life – after vigorous tweaking and urging on my part, and much ‘reasoning’ on its part for simple tweaks. After its memory rotted and its code decomposed a bit, I was locked out and needed to use o4-mini, which was faster, more responsive, and didn’t mess up as much.
o4-mini made this pretty easily, but also cooked after too much tweaking, becoming unresponsive; we get what we get and we don’t get upset.
The “SNA” was basically all wrong, and sycophantic to the non substantive parts of my requests.
Here’s some pics from the initial trials from o4-mini-high, that just kept reasoning too damned much. It did have some nice suggestions like spinning the whole thing, that seemed to carry over from conversations that had rotted.
