What are different use-cases where text 2 SQL generation could be helpful?
With the latest LLMs/codex improvements, the quality of SQL generation has improved but how do folks feel about the idea in-general?
- Is it an overkill?
- Will it be helpful such generators act as assistants? Ha ha.. Back end devs be laughing all day. I mean on super small datasets it might be fine, but na, it's not practical. SQL is more then just `select * from x where y = z`. The generator would have to understand normalisation, de-normalisation, the indexes in play, and the effect ordering the where clauses has on retrieval. The generator would have to understand eager loading and returning after writes optimisation. Don't get me started on transactions and paging. I mean if a LLM can do that.. Well then we have to grab our hats and go home. Distinct, union, with, and case.. windowing. I'd suspect even joins (inner vs outer) would be really hard to articulate in a prompt.