Lazy AI - Eval: How we evaluate LLMs (a look at DeepSeek)