Why Not Use AI?
-
Comments posted to this topic are about the item Why Not Use AI?
-
The conversation about all of that seems like it started on LinkedIn. IIRC, they were announcing that CoPilot was coming out as a part of a new release of SSMS and a whole bunch of people, including myself, asked them to make sure that it could be turned off.
To me, it's not a matter of if I'll use CoPilot or not... Rather, it's a "feature" that is similar to Intellisense or automatic indentation. I can't/won't speak for others but I need the option to turn it off, especially if I'm doing something intense and don't want an even more intense version of "Clippy" popping up and interrupting my train of thought, especially since I would have done the research ahead of time. I've also found the nasty propensity of every offering of it that I've tried to be seriously incorrect even on some of the simple things.
"Clippy" wasn't the end all to be all. Neither is automatic indentation or Intellisense or AI. I'll always want the ability to turn even the best intentions for advance features off because they frequently aren't doing the job correctly for ME.
Regressing a bit, I'm more than just a disappointed in the supposed "helpful features" that MS decided to default to enabled in 2022. I'm also disappointed about the equivalent of TF 1117 not being able to be disabled in TempDB as of 2016 especially since they did NOT fix the problem of SET IDENTITY INSERT ON forcing a full table sort in TempDB!!! Imagine my surprise when that little hank of hair caused a long standing proc to fail because TempDB exploded when 8 files all tried to grow to 100 GB each on a system that almost never grew TempDB to more than 20 GB.
Improvements are not improvements if they break something to make an improvement. The same goes with the breaking code they put in to make conversions from DATETIME to DATETIME2 "more accurate" (also a product of 2016... I won't cover things for 2022.. there's just too much).
The really bad part is that most of the slow stuff in 2022 has been that way since 2019... and they still haven't fixed it. We're still rewritting code because of that "upgrade".
--Jeff Moden
RBAR is pronounced "ree-bar" and is a "Modenism" for Row-By-Agonizing-Row.
First step towards the paradigm shift of writing Set Based code:
________Stop thinking about what you want to do to a ROW... think, instead, of what you want to do to a COLUMN.Change is inevitable... Change for the better is not.
Helpful Links:
How to post code problems
How to Post Performance Problems
Create a Tally Function (fnTally) -
I already use AI and love it, but I'm more concerned about Microsoft's particular implementation.
The demos I've seen have been half-baked at best. They required a lot of copy/pasting data back and forth between different input areas (Copilot, the query window, and the results window), manual work that leads to errors. The AI's output had clear, obvious errors that the demo person would skip right past, which troubles me. If the demo person didn't recognize the hallucinations in the output, that's one problem - and if they're purposely skipping those to avoid discussing them, that's also a problem.
The demos don't have any security awareness. This is a database product, and it encourages people to copy/paste queries and results out to the Internet. While I recognize that MS says they won't log the data or use it for training, they have a really bad track record on that front - the telemetry stuff and memory dumps have been recent examples of "oops, we forgot about that part, we'll fix that at some point."
Microsoft has a track record of shipping half-baked tooling. We can't ignore the history of Data Dude, DACPACs, Azure Data Studio, and the MSSQL extension for VS Code. Microsoft doesn't devote the staff necessary to do a good job on this free tooling, and as a result, the half-baked stuff never gets fully baked - Microsoft's off to The Next Big Thing.
This half-baking isn't just an AI problem. The same thing is present with the source code integration and dark mode, for example. Microsoft's making excuses saying they need more time - but they wouldn't need more time if they focused on less features, and doing a better job on those features. (This isn't even just an SSMS problem - it's spread throughout the data team, with similar problems in the database engine and the cloud products.)
My complaints don't mean I don't like AI in SSMS. I love the idea of it, done right, but this isn't done to my tastes - this isn't even done. It's half-baked at best, raw proof-of-concept at worst.
-
I'm not a DBA but a C# developer who has used SQL Server since V7. In my normal line of work, I've recently written a number of classes and then asked Copilot to create the unit tests ... something like "create unit tests using the XUnit framework for MyNewClass.cs". And lo and behold, there they are. Mostly all I have to do, as the original developer, is to desk check and run them. At the time of writing, I'm loving Copilot - I hate writing unit tests and this has saved me days of irritating typing and wrestling with mocks and all the other gubbins. I'm not sure how it would work for TDD though!
-
I've not used Copilot but do use tools like ChatGPT and Claude. I ask them for help, they don't thrust help upon me. Whatever happens I don't want an AI tool to be like Microsoft Clippy, or something that puts me in mind of the Sirius Cybernetics Corporation with their plastic pal whose fun to be with.
When I use JetBrains PyCharm it subtly indicates when something could be written better but it doesn't beat you around the head with it. Tools like Sourcery.ai are also subtle helpers. It is easier to accept help from a subtle teacher.
-
Largely speaking, I've thus far not used any of the various and sundry "AI" implementations out there (either for SQL or anything else,) mostly because, well, I don't see the benefit. I can get similar or better answers with a decent search on google or the like and applying a bit of "common sense" to mentally filter the results back.
I commented on one of the other recent editorials here about "AI," I had attended a panel put on by my employer about "AI" and it's future in our line of work, and I asked the presenters (both very "rah rah AI" types) if "AI is little more than a very good natural language search engine." That their reply, word-for-word was "yes, it is" I feel is rather telling.
I'm not sure in the real world, that these "AI" "helpers" (As Jeff reminded us all, Clippy, anyone?) will actually "help" or end up being another hindrance that you need to figure out how to turn it off or block it...
-
One of the areas I see looming larger and larger is the question of liability and legal consequences of using AI in the event of an alleged failure and damages to someone, either individuals or corporate entities resulting from its use. The main area so far has appeared to be that of automated transportation, but I suspect that there is going to be an extremely wide range of application which will end up with massive litigation and financial consequences.
I already carry a personal liability umbrella for my home and vehicle which would mostly be due to harming one or a few individuals. This was brought home to me this week as I sat at an auto dealership having a safety recall performed. We are also well aware of the risks associated with industries such as food supply and pharmaceutical companies and the like. I'm already very concerned with the demonstrated abilities of the FDA evidenced in the many reports of recalls.
I anticipate a whole new era of legal battles around the AI phenomenon in the coming decades.
Rick
Disaster Recovery = Backup ( Backup ( Your Backup ) ) -
I do not find generative AI assistance to be helpful. I hate the fact that it is getting thrust upon us, embedding into everything with no way to turn it off; no recourse other than "submitting feedback". We have a product that no-one asked for; that even has to be presented with a caveat "the information presented may be convincingly incorrect".
I am ethically opposed to it: It is intrinsically flawed for most use cases; dishonestly trained, deceptively marketed; thermodynamically unviable. Unpredictable, fragile, expensive, and slow.
It is a dead-end technology that adds nothing, and only takes away from us. It allows "B" people to cosplay as "A" people, who lack the ability to critically evaluate their output.
I'd like to know what 3.2 billion problem it solves?
So, yeah, not a fan.
-
I have used GenAI, mostly in GitHub Copilot. But also ChatGPT, Microsoft Copilot, and Claude AI. But all from home. My employer won't purchase any AI because it costs money and they don't want to spend the money.
Kindest Regards, Rod Connect with me on LinkedIn.
-
My employer (a moderately large software house in the UK, employing ~350-400 people) not only pays for GitHub Copilot, it also encourages us to use it. It's good for grunt work, doing boileplate stuff that's boring and a waste of time. I also use it occasionally when I get weird EF exceptions - yesterday it identified a very niche issue with an incorrect cast to a reference table id, which would have taken me hours of slogging through (someone else's) code to find and fix. "Dead-end technology"? I'm sure they said the same about the mechanical loom.
-
I have used AI for a few things, but I don’t need it all that often. I spend most of my coding time writing T-SQL but I also write C# and PowerShell on occasion. I recently wrote an app to move data between development and QA databases. I hadn’t worked in C# in probably 8 months or more and needed to jog my memory on something. I didn’t have an account, so I tried using CoPilot in Skype. The first pass was rather broken but after a few follow up prompts I got what I needed fairly quickly. It’s also been handy for languages I’m curious about but don’t work with, like python or rust. But even then, I want to read and understand the basics of the language before I ask AI anything so I can have an idea if what AI gives me is accurate. My company just gave us CoPilot Enterprise licenses for Visual Studio and VS Code. I haven’t tried it yet. Of course, the demos look great, but they always do. If it can save me a few hours here and there, I’m fine with it. But, at least at this point, I’m not trusting it to do more than help in small ways and only when I want it to.
- jasona.work wrote:
Largely speaking, I've thus far not used any of the various and sundry "AI" implementations out there (either for SQL or anything else,) mostly because, well, I don't see the benefit. I can get similar or better answers with a decent search on google or the like and applying a bit of "common sense" to mentally filter the results back.
I commented on one of the other recent editorials here about "AI," I had attended a panel put on by my employer about "AI" and it's future in our line of work, and I asked the presenters (both very "rah rah AI" types) if "AI is little more than a very good natural language search engine." That their reply, word-for-word was "yes, it is" I feel is rather telling.
I'm not sure in the real world, that these "AI" "helpers" (As Jeff reminded us all, Clippy, anyone?) will actually "help" or end up being another hindrance that you need to figure out how to turn it off or block it...
This mirrors my sentiments exactly. There are probably some things it could help me do a bit more efficiently, and some things it'd slow me down as I have to review its errors. I'd rather stick with my own intellect, intuitions, and experimentation than have answers shoveled to me by a machine.
Be still, and know that I am God - Psalm 46:10
- jasona.work wrote:
Largely speaking, I've thus far not used any of the various and sundry "AI" implementations out there (either for SQL or anything else,) mostly because, well, I don't see the benefit. I can get similar or better answers with a decent search on google or the like and applying a bit of "common sense" to mentally filter the results back.
I commented on one of the other recent editorials here about "AI," I had attended a panel put on by my employer about "AI" and it's future in our line of work, and I asked the presenters (both very "rah rah AI" types) if "AI is little more than a very good natural language search engine." That their reply, word-for-word was "yes, it is" I feel is rather telling.
I'm not sure in the real world, that these "AI" "helpers" (As Jeff reminded us all, Clippy, anyone?) will actually "help" or end up being another hindrance that you need to figure out how to turn it off or block it...
Worse than all that, I wonder if it'll make the already too intrusive collection of data (which I don't have any IOT devices in my home) even more invasive in the future (is your new TV eaves dropping on you?).
There's a nice little meme for that...
I also can't wait for people to start using AI for all written and visual communication so that I can use the following...
--Jeff Moden
RBAR is pronounced "ree-bar" and is a "Modenism" for Row-By-Agonizing-Row.
First step towards the paradigm shift of writing Set Based code:
________Stop thinking about what you want to do to a ROW... think, instead, of what you want to do to a COLUMN.Change is inevitable... Change for the better is not.
Helpful Links:
How to post code problems
How to Post Performance Problems
Create a Tally Function (fnTally) -
Great comments, and they often mirror what I see. Some people love AI and use it productively, some don't find it helpful.
I'm on the fence. Some things are very handy and small time savers, sometimes it eats up my time. I am torn on how useful it might be in SSMS, and I'm thinking not that useful at this point, as I know most of what SSMS can do. However, like the command palette, it might be a time saver for small tasks. If it can structure some tsqlt tests around procs or queries, I might be thrilled.
-
From a coding perspective the bits I find most beneficial would equally be done with a decent command line utility if I knew where one was.
Viewing 15 posts - 1 through 15 (of 22 total)
You must be logged in to reply to this topic. Login to reply