4 ms·
recently saw a demo app built on top of GPT3. Used a rest API for prompts. The backend was hooked to a corpus of PDFs and a SQL Database with financial data. W
by vjust 3y ago
recently saw a demo app built on top of GPT3. Used a rest API for prompts. The backend was hooked to a corpus of PDFs and a SQL Database with financial data.
What changed was the queries were English/natural language. The query language has changed. That's the R in CRUD. I wonder if the C, U, and D will also change.
While this is one of many types of AI, it means commands will be in natural English. This might be a big deal for UI builders because how we ask questions has changed to NL.
- wizofaus 3y agoI'm curious about this sort of use case - how long does it take for a GPT-based system to process a bunch of documents it's never seen before so that you can perform NL searches on them? Assuming something ike a million total pages of text are we talking minutes? Hours? Days? And is it at all feasible to ensure whatever factual information is returned is only sourced from said documents, vs being "hallucinated" by virtue of whatever weights exist based on the core training corpus?