The way I see it, is that it’s a great tool for people that already know their shit and don’t want to bother writing the code for it. There are some other use cases which are nice, like updating documentation or helping with pull/merge requests.
But it sure isn’t worth all the downsides we currently know of, with the way it’s being implemented.
It is perfect for producing garbage that should never have been produced in the first place. Marketing wank. Nobody cares what it says, so long as it is aesthetically pleasing to the intended audience. It’s perfect for doing shit that never needed to be done in the first place. Perfect for busywork.
It is worse than useless for any technical purpose. Verifying its output requires more effort than manually producing that output.
Yep, used it on my year-end self assessment. Told it what I would have written and it bullshitted it up to the appropriate length. I reviewed it and fixed where it invented some accomplishments out of nowhere that would have been awkward if I ever got called on it. Knowing that my performance assessment would only be glanced at and never read.
Similarly, a design document owed to management for something they have no actual interest in. No one honors the design doc, no one expects it to be informative that actually uses the tech, but non-technical management feels ‘included’. Didn’t even need to be reviewed, because all the ones it concretely matters know not to bother with it.
Stuff I always hated and was pointless, but someone wanted to check some boxes.
Though just had someone clearly just feed my email into a prompt and told the AI to use my points to agree with me, and that was annoying to see someone using AI to try to be a sycophant to me. If you are going to be a suck-up, at least suck-up in your own words. I won’t do it to something someone will read in earnest.
Sadly, at least for as long as I can tell, the horseshit is simply the language of the executives. They feel like they are being smart and insightful while saying nothing. So of course we are going to be stuck with it in any work that has room for bureaucracy to exist.
At my company we have an internal AI bot that uses Anthropic. We use it to investigate bugs by comparing code to expected behavior or finding things in data.
For example, I don’t write code. But when I need to find how data from a client is being mapped. I could either ask the engineer who’s busy with other work with my menial task, I could look though a bunch of code with bitbucket, or I could ask the bot which will give me the code and any ticket related to the code and pull requests.
When finding something in the raw data. I could take a bunch of files and text strings and pull them into excel and spend a bunch of time doing lookup, matches, and index equations in Excel. Something I did for years in the past. Or have the bot do that in 2 minutes. When I’m just trying to find and validate a few things.
It’s peculiar as a thing that acts the same whether it’s right or wrong. That is a pretty big drawback when trying to use it for anything vaguely fact based. The fact that if you don’t know yourself, it sounds credible either way, is super tricky. Particularly when managers that are used to gauging credibility without knowing the material come along. They are complete suckers for AI output since they don’t have good ways to know better.
Exactly my company is heavily pushing for full agentic development and while that does save time a bit, it comes at a cost, for one I’m no longer getting the deeper understanding of the codebase that I would if I did the work myself, and secondly the number of PRs that AI pushes out creates a review bottleneck because people can’t keep up, but management doesn’t care, they’d rather you push something out fast and if it breaks things it’s your job on the line not the AI subscription.
I hate AI with a passion.
For me AI represents a decline in cognitive ability.
In the past when I solved a problem, I actually learned something, and I felt great about that.
When using AI, I don’t get the same satisfaction of having learned something.
The way I see it, is that it’s a great tool for people that already know their shit and don’t want to bother writing the code for it. There are some other use cases which are nice, like updating documentation or helping with pull/merge requests.
But it sure isn’t worth all the downsides we currently know of, with the way it’s being implemented.
It is perfect for producing garbage that should never have been produced in the first place. Marketing wank. Nobody cares what it says, so long as it is aesthetically pleasing to the intended audience. It’s perfect for doing shit that never needed to be done in the first place. Perfect for busywork.
It is worse than useless for any technical purpose. Verifying its output requires more effort than manually producing that output.
Yep, used it on my year-end self assessment. Told it what I would have written and it bullshitted it up to the appropriate length. I reviewed it and fixed where it invented some accomplishments out of nowhere that would have been awkward if I ever got called on it. Knowing that my performance assessment would only be glanced at and never read.
Similarly, a design document owed to management for something they have no actual interest in. No one honors the design doc, no one expects it to be informative that actually uses the tech, but non-technical management feels ‘included’. Didn’t even need to be reviewed, because all the ones it concretely matters know not to bother with it.
Stuff I always hated and was pointless, but someone wanted to check some boxes.
Though just had someone clearly just feed my email into a prompt and told the AI to use my points to agree with me, and that was annoying to see someone using AI to try to be a sycophant to me. If you are going to be a suck-up, at least suck-up in your own words. I won’t do it to something someone will read in earnest.
Exactly the kind of horseshit I’m thinking of. The demand for mindless busywork is truly amazing.
I can only hope that the next big idea in business is a widespread push to “replace” AI production with “nothing”.
Sadly, at least for as long as I can tell, the horseshit is simply the language of the executives. They feel like they are being smart and insightful while saying nothing. So of course we are going to be stuck with it in any work that has room for bureaucracy to exist.
Meh, LLMs and GenAI as tools themselves and as technologies, are not that bad. There are definitely some legitimate and extremely useful use cases.
But it should never have been released to the general public, it will bring extreme harm to the world.
At my company we have an internal AI bot that uses Anthropic. We use it to investigate bugs by comparing code to expected behavior or finding things in data.
For example, I don’t write code. But when I need to find how data from a client is being mapped. I could either ask the engineer who’s busy with other work with my menial task, I could look though a bunch of code with bitbucket, or I could ask the bot which will give me the code and any ticket related to the code and pull requests.
When finding something in the raw data. I could take a bunch of files and text strings and pull them into excel and spend a bunch of time doing lookup, matches, and index equations in Excel. Something I did for years in the past. Or have the bot do that in 2 minutes. When I’m just trying to find and validate a few things.
the problem with ai isn’t the concept, wikipedia is great
the problem is it is not deterministic, often factually wrong and being used by the rich to steal workers pay
It’s peculiar as a thing that acts the same whether it’s right or wrong. That is a pretty big drawback when trying to use it for anything vaguely fact based. The fact that if you don’t know yourself, it sounds credible either way, is super tricky. Particularly when managers that are used to gauging credibility without knowing the material come along. They are complete suckers for AI output since they don’t have good ways to know better.
aka the bias confirmationer
Exactly my company is heavily pushing for full agentic development and while that does save time a bit, it comes at a cost, for one I’m no longer getting the deeper understanding of the codebase that I would if I did the work myself, and secondly the number of PRs that AI pushes out creates a review bottleneck because people can’t keep up, but management doesn’t care, they’d rather you push something out fast and if it breaks things it’s your job on the line not the AI subscription.
I use it to identify useful parts in scrap piles.