"We did it, Patrick! We made a technological breakthrough!"
A place for all those who loathe AI to discuss things, post articles, and ridicule the AI hype. Proud supporter of working people. And proud booer of SXSW 2024.
AI, in this case, refers to LLMs, GPT technology, and anything listed as "AI" meant to increase market valuations.
I work in finance IT at an AI pilled company, I not only have to deal with the increased prices, I have to deal with management who generates slop all day long while proclaiming how fantastic it is, as an AI skeptic it hurts everytime I go to work.
I am currently looking for a new job, my stress and depression is through the roof.
We were forced to switch from agile to some kind of AI native bullshit with smaller teams, because that's what the large companies do.
It's a train wreck. The SDLC model in itself sounds pretty nice in theory, but requires everyone to self-manage, and most importantly for the teams to be self sufficient and not blocked by others. I was curious how would it work, so I looked up as much as I could about it.
I think I'm the only one who did that.
Of courss they ignored the core ideas, disregarded and broke the main principles it's build upon to work, and just made a mess in our team structures while saying "AI will help with this". Everyone is confused, has like twice as much meetings, and our development has grinded to a halt while everyone is dealing with byrocracy. It's also a model for 100+ developers sized companies, and for a different industry than ours. We have like 10.
It would be funny if it weren't so sad.
I was willing to give it a try, if it was a serious attempt. But what infuriates me even more is that they don't even know the basics of what they are doing.
I would love it if companies/organizations actually evaluated AI properly.
I mean like this:
Tale three teams of developers, all equal size and experience, and give them all the same task, but set requirements like this:
Team 1: You guys go all in on AI, just go ham!
Team 2: You guys may use AI as a manual, you may not copypaste code from AI, you can't ask it to make what you want, but you can ask it for examples of how to do specific tasks.
Team 3: You can't use any AI, you can use Google, Bing, Duck Duck Go to find help, but you may not use any AI response.
Once the teams tasks are completed, and working, all of the versions of the code is pushed to test and is run for 3-4 months each to evaluate stability and ease of maintenance.
Once that is done each teams code is scored on an index of development time, stability, and ease of maintenance.
It would be interesting to see an accurate test between the different methods compared.
Management loves to shout how much time AI saves, but I have yet to see any evidence
You see, the ones running all this don't care about "AI" or how useful any technology is for that matter. They just want to get their hands on your money and they care little about what they have to say to achieve that goal.
I think this comes down to the usual capex vs opex situation. Anyone who has bought a car or a house is familiar with this problem. Should you buy something crazy expensive that costs much less to maintain? Should you buy something cheap that will drive you crazy and cost you a fortune in the long run?
I would expect that, companies prefer to spend 5 hours on making a whole bunch of code that will require 1000 hours of maintenance within the next 2 years. At least thatβs the exact philosophy I followed when buying my first cars.
It's still funny lol
It's a generalist tool that cash do most things at a subpar level, some things decently and apparently has been excelling at a select few things.
It's not what we want to replace everything with
Most importantly it's a non-deterministic tool that people are showing into things that can be easily made deterministic, but that's not cool enough or something.
I love making automation scripts, but I'm never given time to do them properly, so I had to shove it between compiling and whatnot. So, most were unfinished and needed just a bit of polish.
Then, a three week long AI-tooling sprint comes, because suddenly we have time for that, and they replaced all of it with Agent based solution, where one run costs tens of dollars in tokens, and it also sometimes just straight up halucinates results, becasue it ran into an error and you asked for data, so data it will give. It has already ran a wrong build when it was confused, halucinated results, and in general it's faulty.
If I was given three weeks to finish my scripts, we'd have the same results, but actually reliable.
Hell, we've replaced our regular daily chat messages (like, post your standup here thread) from a normal scheduled message to, for some inexplicable reason, messages posted by AI. The jobs fail to fire like 10% of the time, it posts the reminder in the evening, or only when you ask it about it. Like, why?? We had a perfectly deterministic solution for all of that.
I would kind of understand if people used AI to build the deterministic scripts, but noo, we can't have performance test running script, we have to add AI to the loop so it can be more flexible. It's fuck all flexible, just sometimes spews bullshit, runs the wrong thing, and we get data that no one checks and people just trust. We already burned so much mandays on trying to investigate something that was just a bullshit halucination.
I hear you. At my work we are expected to experiment with it for coding, see where it can improve our workflows. I think that's fine.
What you're saying sounds like a nightmare. Heck give me three weeks of time to do whatever I feel like and I'll improve things a lot more than my hallucinating buddy ever will.
Iβve literally listened to middle management demoing βan example of stuff you can doβ and βwell it doesnβt work right now but you can see the potentialβ and then announcing his leaving the company a few weeks later.
Gee thanks for nothing you useless crusty skid mark.
While you wait for RAM to get cheaper, may I suggest you revisit older games? A lot of them were really good you know, and they run like a champ on even the lowest-spec machines today.
No its not about gaming, and many peoples dont buy gaming pc for actual gaming, it is just the term we have given to some powerful consumer PCs that's it
Are you trying to build AI models?
Prices expected to increase through 2028 and beyond
AI, like with corporate downsizing, is a convenient excuse. The real cause is the RAM and HDD cartel. Gamer's Nexus already has multiple videos on this subject.
they can stick their ram sticks up their arses
Pretty sure my laptop has faulty ram and I'm just gonna hope it doesn't get significantly worse because 32gb of laptop ddr5 ram is gonna do more than just break the bank lol
Fucking AI
market being messed-up means sacrifices and lower expectations. if you're in the u.s. here's a starting point for you.
$890 5500/2x8gb/500nvme/9060xt(8gb): https://pcpartpicker.com/list/ZgGxfp
if you have a microcenter available to shop in-person at, you can save $50 on mb/cpu/ram ($250 bundle), and $10-20 on a case, so $820-840.
or hopping on fb or cl and buying used.
Seconding that last part. Economic times are tough and some people have other priorities rather than processing power. I have seen more than a few decent gaming rigs in my area selling for less than $500 on FB marketplace. They go fast and obviously beware of scams, but they are out there still.
AI from big (US) corps are a biased scam by definition, never ever use an AI app inbuild in your OS, it will spy you and wasting a hugh amount of RAM and Bandwith. If you need an AI for some task, use a specific one for a function you need, a lot of these you'll find for free, withot the need of an account, FOSS and made by little startups and devs, which don't have the need to spy on you. You cand find a ton of these eg. in HuggingFace, ready to use.
Go outside, get some perspective. Yes they suck super hard and should be reined in. No, a modern gaming computer is not a need, just a want. You will survive.
reined in
Thank you