With this new database "Query Language" you don't store the compiled final SQL, which has been output according to the requirements and laid out during the lengthy discussion with you.
Instead of Select distinct Salary from Employee e1 where 2=Select count(distinct Salary) from Employee e2 where e1.salary<=e2.salary;
You have:
AIQL.do("Get the 2nd highest salary in the company.")
Two separate equivalent things can be stored:
1. The long discussion, with you telling it first what you want, it asking you questions and together slowly refining the definition ("debugging" it) until you get exactly what you meant in the first place.
AI QL: What do you want from the database? Me: Get the 2nd highest salary in the company. AI QL: You mean a single number, with the salary of one employee, after checking each one of all the employees? Me: Exactly. AI QL: Will you need it for a certain time, or whatever happens to be stored there right now? Me: just what's there now. AI QL: Will you want to know the employee number and the datetime this was changed? Me: Nope. AI QL: OK. Do you want to see the SQL, a description of how it worked, or just the result? Me: Later I will. AI QL: OK this is stored as "Project:Interview. Query: Get the 2nd highest salary in the company. (Only a single number. Negative 1 if something goes wrong)"
2. The summary in natural, clear, concise, and unambiguous English which gives the exact and correct query when compiled into SQL.
Example: Project: Interview. Query: Get 2nd highest salary. Does: Gets nth highest salary by fastest method shown in this StackOverflow link.
Of course the next step will be AI CompuReq for programming computers via requirements and debugging the program by refining the requirements with nitty gritty. But that's for the next idea.
Both these ideas are different from Cody AI now available for Visual Studio VS, because you won't be looking at the code anymore at all, and maybe not even having to understand it!
You'll be looking at a generalized description of what is happening, and if that's not good enough you'll be getting a more and more detailed description. Only in tricky places you'll need an expert to take a look at the actual code (or SQL, in the case of this early implementation of the idea)-- pashute, May 07 2023 https://github.com/features/copilot [pashute, May 08 2023] https://marketplace...cegraph.sourcegraph [pashute, May 08 2023] Up next: drill-down conversations lasting an entire work-day. Do you mean for all employees managed by Conglomerate, from Sub-conglomerate B, or from some other group? Do you mean among those in time zone A, B, C, or D? A tuna fish sandwich or a salami sandwich?-- Voice, May 07 2023 This kinda works with either Sourcegraph's CodyAI or Github Copilot with OpenAI.
But they don't have a standardized repo for getting it done, so you might get some other code next year. We need it to be able to redo what it did last time so that the output is completely working as designed.
No, it's not ad infinitum as Voice suggested. The AI is just using the intelligence it gathered in the past to ask a few "pseudo-intelligent" questions from experience before it fetches the info.-- pashute, May 08 2023 Would the AI be able to say, "The database does not contain an accurate answer to the question you asked, because current working practices do not capture that information, even though the database schema makes it look as if they did"?
Or would it just return the wrong answer without comment?-- pertinax, May 08 2023 "This database is attempting to access HR information to which it has not been granted by a user who is also not registered as having access. Access is denied and this query has been recorded and logged."-- RayfordSteele, May 08 2023 Hello AI QL, I have a query but I am not sure how to ask it. Can you just give me the information I need please? Quickly please. Thanks.-- pocmloc, May 08 2023 Certainly, [pocmloc]: 42 </obligatory>-- pertinax, May 08 2023 I thought I'm writing on Halfbakery, but you guys are seemingly pretending to be as GPTish misunderstandingish as possible. Is that by design?-- pashute, May 11 2023 random, halfbakery