In light of this comment, what is the point of the contest? An LLM with sufficient context window can simulate execution of any program — what does it matter whether some given program’s exec trace fits in 32K when no one would ever use a pure LLM for this vs. an LLM + REPL?

→ View original post on X — @goodside