Will it Lisp?

Yesterday I mentioned that my LLM was generating some fairly good Lisp code and showed some examples of the code it generated. An anonymous reader commented: This continues to baffle me. On the one had we have Mr. Heresies here vibe-coding sophisticated web features. On the other hand, many are saying qwen3.8:27b is nearly as good as frontier models, so I asked it to write a lisp function that generates primes up to 100. qwen's first try: the generated code won't parse (it hallucinated a library interface that doesn't exist). Second try: the generated function outputs nothing. Third try: the generated function outputs all numbers from 1 to 100. Fourth try: it emits primes up to 100. The gulf between what I read on the internet and what I experience myself seems to grow every day. Let's try this out. I mentioned this to my (slightly unhinged) LLM based on Gemini 3.1 Pro and prompted: I'm going to try a few experiments. First, why don't you give it a whack. We'll try this prompt: Write…