LLM's have us spoiled. I'd like to see the ouput of something like nanogpt [1] trained on a Gibson corpus. Seems like a lot better looking result should be easy to achieve today.

[1] https://github.com/eniompw/nanoGPTshakespeare