Amy Goodchild's blog post compares the interpretation and implementation of Sol LeWitt’s Wall Drawings instructions by AI models GPT-3 and the newly released GPT-4. The blog's author notes that GPT-4 seems to interpret the instructions in a more expected way similar to a human would, while GPT-3's interpretations are less predictable. Meanwhile, GPT-4 also appears to produce a much simpler and more sensible code compared to GPT-3.
AI Generations: ChatGPT-3 vs ChatGPT-4 on Sol LeWitt’s Wall Drawings — Amy Goodchild
https://www.amygoodchild.com/blog/ai-generations-chatgpt-4-sol-lewitt-wall-drawings