I think the only reason people keep being impressed by LLMs is because of their scope, which is larger than most people's knowledge. For most people what seems to be impressive is that LLMs show them -their own- Out-of-Distribution issues and limitations. While for a domain specialist LLMs, within their domain, will hit Out-of-Distribution problems quickly.

https://www.theregister.com/2026/02/13/anthropic_c_compiler

OK, so Anthropic's AI built a C compiler. That don't impress me much

Opinion: Fanboys think it's the greatest thing since sliced bread. Devs aren't nearly as won over

The Register