[hodgehog11]

Some very tough computational geometry problems I couldn't solve on my own, nor with the assistance of other AIs or my colleagues. Fable did them all first go. The most impressive built a custom optimizer with a ludicrous number of adaptive switches that absolutely crunched through an error surface with a bunch of nontrivial nullspaces and some wild curvature. That optimizer is of independent interest; it's not totally novel in theory, but the implementation is an impressive piece of engineering.

[mk89]

Was it a better prompt? Have you tried giving the same prompt to other models?

I have found out that the mistakes of other models (which I choose first to save money) help me refine the prompt more and more, until I am fed up and pick Opus 4.8 (for example) which magically seems to get it right, but there is a lot of pre-work there...

[hodgehog11]

Yes I have tried giving these same prompts to other models. The difference has been painfully clear. Switching back to Opus, it is completely unable to do anything that I had asked of Fable without significant conceptual and engineering errors. Functioning code, sure, but not even remotely capable of accomplishing the task to the accuracy I need. Sonnet, GPT 5.5, Gemini, DeepSeek, it's all the same deal. I accepted this in the past because that was just how it was. Now it's tremendously irritating.

I wish Fable really was only a minor upgrade so that it wouldn't be missed, but this feels like the difference between having a post-doctoral colleague and educating a student that I have to constantly guide and correct. It's so profound for me that some of the reactions in this thread feel like they come from another reality entirely. Or maybe they just got instantly diverted to Opus, who knows.