Last week I had to caution a junior engineer on my team to only use an LLM for the first pass, and never rely on the output unmoderated.
They're fine as glorified autocomplete, fuzzy search, or other applications where accuracy isn't required. But to rely on them in any situation where accuracy is important is professional negligence.
They're fine as glorified autocomplete, fuzzy search, or other applications where accuracy isn't required. But to rely on them in any situation where accuracy is important is professional negligence.