A sanity check for future plugins to access private SQL databases

GPT builders Plugins / Actions builders
43

Hi Guys, I am planning on releasing a database + NL question corpus that gets exactly at the issues we are discussing.

Would it be useful if I built a public corpus around the following schema:

   <table id="stock" writes="full">
        <column id="Symbol" type="text"/>
        <column id="Name" type="text"/>
        <column id="Market_Cap" type="int" lex="Market Cap"/>
        <column id="Country" type="text"/>
        <column id="IPO_Year" type="int" lex="IPO Year"/>
        <column id="Sector" type="text"/>
        <column id="Industry" type="text"/>
        <column id="exchange" type="text"/>
        <primary_key key="Symbol"/>
    </table>

    <table id="quote" writes="full" lex="quote,quotes">
        <column id="time" type="time"/>
        <column id="symbol" type="text" properties="none"/>
        <column id="price" type="real" units="USD"/>
        <column id="volume" type="int"/>
        <unique key="time,symbol"/>
        <foreign_key from="symbol" to="stock"/>
    </table>

The database state will be some subset of the stocks of the NASDAQ and their price at some frequency over the day.

The natural language questions would be like:
“sectors of stocks”
“names of energy stocks”
“names of energy stocks that are down 5% in the last two hours”

Would this be an interesting corpus to try to do a little prompt engineering on? Golly I will eat my hat if
ChatGPT can do this in a robust way.

Let me know if I should proceed.

Michael