Hello everyone,
I chose the title of my topic was I believe it accurately reflects my current dilemma; how do I adjust my language to best guide Davinci on correcting values in a table (Country, City Description) of IP subnets?
Another way of looking at my problem, “How do I adjust my language to guide Davinci in correcting incorrect data inputted by humans?” There’s tons of software out there that accomplishes this task, but like anything else these mechanics are subject to the second law of thermal dynamics, it breaks down over time.
What I’m trying to correct is that anecdotal 5%. Things like “uNited stats of merica” and “usa” or “cnada” or “CDN”, those typos are easy enough to update. But how do you build your language construct in such a fashion that GPT3 understands that I want her to look for uninformative descriptions like “router alpha bravo 01-mls-gige22” to “Human beings using virtual machines”.
Other mechanics that are intuitive programmatically, like evaluating each row and if a property has a value of $true then apply this update unless there are these other conditions which have their own updates (I’m trying really hard to not use programmatic languages to explain myself!); that kind of logic is tedious to engineer but not impossible. And those are for the known conditions, the tough part about the 5% is it’ll often represent the unknown knowns.
“I know that’s a VPN subnet cause Faye told me from the VPN team, “here are the subnets””. However, in the table the VPN field is $null and the description is useless it says “Vendor Alpha Bravo VpnDeviceModelNumber and some convoluted old junkie electrical terminating numbering series”. I can create a lookup table based on a wishfully accurate CMDB and then use that data to say “Ah, if I see these model names/numbers than there’s a probability low or high that this subnet represents users on a vpn hardware”, then I can make a generic description “Assumed as a human being subnet in CountryA and labelled VPN. I could be wrong by $value”. [The problem with using device interface labels is they are often uninformative as well and need translation. They likely accurately describe something for these grouping of people but all the other humans have no idea what switch-airportcode-moreletters-interface2/2/2/3/4 or whatever… might as well smash a keyboard.]
My intent here isn’t for a generic github/sourceforge copy/paste problem solve + wack-a-mole regex, but rather a language construct that I can then keep updating. Once I have accomplished the above problem solve, then the next step is to make the descriptions more human consumable.
I’m using the ip subnet description use-case as it is small enough for people to consume and understand as an irritant. Also big enough of an irritant in terms of reporting that I can explain it 5-7 word slogan and have it pinned in a power point.
Thank you kindly for your time reading through my post.
Respectfully,
~C