DigiNews

Tech Watch Articles

← Back to articles

LLMs are bad at vibing specifications

Quality: 8/10 Relevance: 9/10

Summary

A critique of LLMs for generating formal specifications, using TLA+ and Alloy examples to show that AI often writes obvious properties that miss subtle concurrency and correctness issues. The piece argues that strong verification logic requires human insight and that AI can lower the barrier but cannot replace careful validation.

🚀 Service construit par Johan Denoyer