birb_cromble
No bio...
User ID: 3236
I deal with this shit all the time. My day job involves maintaining a very old application that has seen continuous upgrades and maintenance over a two decade timeline, but some vestigial bits haven't been touched since the application's inception due to manpower constraints. We all assumed ai would help us deal with the backlog of "not quite important enough to move forward to the top of the stack" items that cause a drag on developer productivity, but always get sidelined by more pressing concerns.
Unfortunately, our codebase takes every AI product we've tried and sends it spiraling down a garden path of schizophrenic recursion.
Before I continue, let me get the usual reflexive responses out of the way:
- We have tried multiple times, with multiple LLMs, including whatever the zeitgeist considered to the SotA at the time. No, I am not going to go back and retest because Opus-4.8-thinking-xhigh-ultra-planner-with-cheese is truly a revolutionary improvement.
- If you claim we are Prompting It Wrong, I will personally figure out how to send you to at least four of the eighteen Chinese hells. Seven different people on my team have tried, all with varying backgrounds and writing styles. Two of us are published authors. One is literally a technical writer who writes concise, complete, and detailed prose for a living.
- No, I'm not going to take a two million line, twenty year old application that runs in regulated environments and rewrite it in Go-lang, even if Claude is really good at it.
The problem is that one single fix or feature can easily involve traveling through over a decade of code-strata, and the LLMs invariably over index on the idioms that they encounter first. The end result is that even a detailed plan on how to replace a deprecated method from a library with a documented replacement tends to make it imagine that it has to do things it doesn't.
I'm going to be vague here, but I'm going to share a personal story from work.
Due to our support contracts, we occasionally have to release security patches for very old releases of our product. We recently had an issue where one of our very old releases was using library $FOO on version N. Version N of $FOO had a very hard to exploit but potentially real security issue that could cause an information disclosure. Normally we'd just tell people to upgrade, but one of our clients operates in a regulated environment and can't do a major upgrade in less than 12 months for anything short of a nuclear bomb going off. Given that, we'd normally upgrade the library and move on with our lives. However $FOO has officially EOLed version N, so no more updates.
We had to do a manual mitigation, and I'm the guy who catches that kind of work, so I got started. A co-worker said "hey, this is high priority, and I think I can get it done in 15 minutes if I use AI". Since it was high priority, the boss told us both to get to work and whatever made it through all our automated tests and reviews first would go in.
I did my fix by integrating with some functionality provided by $FOO that allowed me to catch invalid data before it made it to the offending code. I missed an edge case that got caught by the automated tests, but either way, the PR landed in the codebase with half an hour to spare before the end of the day.
I decided to see where my coworker was, and the answer was "nowhere close to a working solution ". I could only see his draft PRs and not the prompts, but it looked like the LLM tried two different approaches and shit the bed both times. The first time, it looks like he/the LLM tried to catch invalid data at every single entry point into the system - the top of the funnel, rather than the bottom. The problem is, we have thousands of entry points. The code got completely out of control and eventually the LLM started over fitting and trying to apply the mitigation to things that didn't even make sense. Thr second approach involved manipulating JVM bytecode to try and hot-patch the offending class. That didn't even compile, but it looks suspiciously like an old stack overflow post about a different problem with the library.
The other guy is pretty sharp, and in some areas I think he's genuinely more talented than I am. But looking at the work he did that day, you'd assume he was fresh out of a boot camp and in way over his head.
I'm not really sure what to make of it.
birb_cromble
2d ago
Does anyone have any pointers for getting a smooth, even finish with nitrocellulose lacquer?
I'm doing test coats and keep getting tiny bubbles.
birb_cromble
3d ago
Read a book that you've already read before. You already know where the plot leads, so your mind can relax a little and take the body with it.
birb_cromble
4d ago
fallen into the cluster B trap
Been there. Learned my lesson. Never again.
birb_cromble
4d ago
My contractor for last week's home maintenance surprise cashed the deposit, so spending is $1,214.28 higher than the same day last year. If I'm careful and frugal for the next nine days, I should be able to stay cash flow positive for the month and start getting back on track.
birb_cromble
4d ago
Somebody's gotta hang out in East Kentucky and treat the Black Lung.
birb_cromble
5d ago
If we use Bitcoin as a reference, they tended to crap out after about 3 years because of blown capacitors.
Maybe things have improved?
birb_cromble
5d ago
Did he mean it as "increase growth from 2% to 4%", or "increase growth from 2% to 2.04%"?
birb_cromble
5d ago
The comparisons to the dot com and railroad bubbles concern me sometimes.
A railroad line can last centuries if properly maintained. Fiber has a 20 - 50 year lifespan. They were both totally usable by the time everyone finally got over the mania. I'm not sure the same is going to be true about GPUs. The data center physical structures will exist, and maybe the power infrastructure, but even the (IMHO optimistic) projections on GPUs show a 6 year depreciation schedule.
birb_cromble
6d ago
I’m not convinced it’s a bubble
My current layman's opinion is that the current environment is a bubble, but that bubble is entirely independent of the technology itself.
It's clear that at least some people, in some circumstances, are getting value out of the technology. It's not like NFTs, where even the best use cases are better served by simpler, pre-existing tech.
That said, the current economic environment is baffling to me. Every big provider is acting like this is a zero sum game where one company winning will give them a monopoly forever. They're also acting like the progress curve will produce exponentially increasing capabilities forever while operating costs approach zero.
I'm not sure if the market as it stands can achieve profitability that justifies the current AI company valuations if there are 3-4 winners instead of one. They're all priced with the assumption that one of them will utterly own the most transformative technology since the steam engine. If that's not true, people are going to start asking why they're not getting a 10% return on a company that has a 20x P/S ratio. Once people start asking that question, it's going to get uncomfortable for anybody that's not a monopoly already.
They're taking on significant debt, too. Take meta, for example. If just one of their data centers has a twelve month delay, that's a ~3% hit to free cash flow to service debt on an asset that isn't making any money. When was the last time that you saw a construction project more complex than a doghouse finish on time and on budget? Even if they finish construction, there are significant delays getting them powered, and gas turbines aren't a permanent solution. There's pretty enormous systemic risk there. Some companies are better equipped to handle it than others, but none of them are immune. Oracle, in particular, appears to be laundering questionable debt through their investment grade credit rating, which is unlikely to end well for them.
That said, even if Anthropic and OpenAI shit the bed and contagion through the bond market causes a market crash, and Google puts their research back on the shelf, LLMs don't go away. Local models exist. China is still plugging along with much more reasonable objectives.
I don't know exactly what the future holds, but either way, it'll have LLMs in it.
birb_cromble
6d ago
Except crypto was almost always purely in the realm of theory-applications.
If nothing else, Bitcoin can always buy you a pizza, although the only topping will be regret
birb_cromble
6d ago
I'm in my 40s. I haven't had a drink since I was 22, when I absolutely obliterated myself in a blackout bender after a bad break-up.
A lot of my peers have stopped or curtailed their intake significantly. Alcoholism was rampant in my social circles, and everybody seems to have cut back or died.
birb_cromble
6d ago
Thanks!
birb_cromble
6d ago
Because we apparently picked ourselves up by our bootstraps and created from scratch all the text which is used to create LLMs.
This is clearly proof that the Saurian Overlords of Agatha in the Hollow Earth taught us language.
More seriously, isn't there a lot of research going into using synthetic data safely? I thought that the current consensus was that you can avoid model collapse with synthetic data if it's properly labeled as such.
birb_cromble
6d ago
This sounds like it has a lot of promise. Can you provide a link to what you mean when you say downspout adapter? I have a huge gap in my handyman skills when it comes to this kind of thing.
birb_cromble
7d ago
My one gutter downspout is clogged. My ladder isn't tall enough to reach that corner, and my usual gutter guy retired on short notice this year. The downspout connects directly to an old pipe and has concrete poured around it. Even with an extended pole I can't go in through the top because it has gutter guards. All the other services in the area either aren't calling back or are booked til mid summer.
I'd like to try and clean it out, but my only ideas right now are to buy a large and expensive ladder, or to try and find a point where I can take the downspout apart and come in from the bottom.
Do any of you have ideas?
birb_cromble
7d ago
At least for me personally, I just hope this leads to less retarded mandates from my higher-ups about using AI X times a month etc. (we're literally tracked on usage and it can affect our raises/bonuses).
I work at a dinosaur of a company, so I can't speak to this directly, but a friend of mine that I recently mentioned gave me an update the other day. They've gone from "you must burn as many tokens as possible to maximize your performance review" to "we must use our token budget wisely."
The timing is interesting. It happened right around the same time Anthropic started putting the screws on its customer base with increased token usage and tighter rate limits.
I really feel like the company that sits on a "good enough" model and aggressively cost cuts is going to win this particular war.
birb_cromble
7d ago
I've definitely seen a decline in chicken quality over the last five years. I'm not exactly sure what the cause is, but it's real. Woody breast, in particular, is almost a guarantee unless you get chicken from a farmer's market.
birb_cromble
9d ago
I would put a cornucopia on the fruit of the loom logo and never speak of it again.
birb_cromble
9d ago
This is one of the reasons I'm taking a break from BJJ. Over half the guys were openly on something, and I have a feeling half of the remaining half just kept their mouths shut about it. It got to the point where people didn't believe me when I said I wasn't getting outside help.
birb_cromble
10d ago
The best explanation I've heard is that nobody is an objective 10. A 10 is a 9 that does it for you, specifically.
birb_cromble
10d ago
Goddamn. That's a right kick in the balls. I hope it works out for you.
birb_cromble
10d ago
Every time somebody makes an "AI related" move like this, I go look at their financials. Once again, I am unsurprised. From Variety:
For a full-year 2025, Snap reported a revenue of $5.931 million (up 11%) and a net loss of $460 million (compared with $698 million in the prior year).
Looking at filings, I don't think they've ever been profitable over the course of a full year unless you're talking positive EBITDA, which is something of a nonsense measurement.
I think my bigger question is: how the fuck is a company that was founded in 2011, and IPOed in 2017, employing over five thousand people while losing hundreds of millions of dollars per year, still in business?
It really seems like that's the deeper question here. Ever since ~2021, the economics of software companies have increasingly decoupled from the fundamentals that are supposed to describe a healthy business. I look at that and think "no wonder people are having trouble finding entry level jobs, some of the biggest sectors in our economy are very sick".
Am I missing something here?
birb_cromble
10d ago
Implants are expensive, particularly when there are complications.
I think we recently had to have a stern-yet-loving talk with a customer who was still resisting an upgrade after eleven years.
More options
Context Copy link