• Docker

• Environment variable OPENAI_API_KEY

https://hub.docker.com/r/aa8y/postgres-dataset

make run-db target runs the database with the dataset.

make

make run-all

make run-db

make run-db-client

make run-chatdb

make run-chatdb-db

make stop-all

• it really depends on the question: "Can you compare different cities" yields just names of cities. What I'd really like is to take 5 cities from the database and compare them using information from a different source. I.e to use data from the database and then combine it with another source.
• It never gives up on an answer
• There is a conversational aspect missing. It could be because of the queries but somehow the results are not being chained in from Openai.
• I don't know what's constructing the query. I see a lot of HF models too

Can you tell me a little bit about JavaScript? yields

SQLQuery: SELECT name, localname, continent, region, population, governmentform
FROM country
WHERE name = JavaScript
LIMIT 1;

Pretty good if javascript were a country.

I wonder if there is merit in doing the reverse - getting results from open AI and really augmenting it with our results and eliminating nonsense. In other words something like stable diffusion.

https://github.com/langchain-ai/langchain/blob/cdfe2c96c530f42fbf2b200c87c617fe5fac6dfd/libs/langchain/langchain/utilities/sql_database.py#L119

• this is a major one, on large databases it fails because of prompt limitations. Example db: postgresql+psycopg2://demo_user:[email protected]:5432/demo
• https://stackoverflow.com/questions/76887233/how-do-i-solve-working-around-models-maximum-content-length-exceeded-errors?noredirect=1#comment135548887_76887233

From a question, how do you figure out where to go if there are multiple datasources?

It would be nice to have a gendb layer to go to