AI 모델에 대한 실용적인 글 — 무엇을 할 수 있고, 비용은 얼마이며, 무엇을 골라야 하는지.
6페이지 중 6페이지
One of the most reliable AI tasks — with one mistake that costs people. Where to put your question, and the instruction that prevents invention.
One is an assistant that can search, the other a search engine that writes. The honest split, and where Gemini sits between them.
The split that decides whether this works, what should never be automated, and why deflection rate is the wrong metric.
Why rollouts fail, the three-line policy that beats a document nobody reads, and the shared prompt library that drives adoption.
Material, audience, shape and bans. Plus the widely shared techniques that do nothing, and why examples beat adjectives.
Medium, lighting, composition, concrete nouns and exclusions. Plus everything people add out of habit that does nothing.
Why Claude is the developer default, what all of them still get wrong, and how to read coding benchmarks sceptically.
Reliable because the material is supplied. Attribution and decision-vs-discussion are where it goes wrong — plus consent and retention.