Ain't that interesting ?
Borkzilla is having to to police its own workforce on the use of its shitty Clippy.
Maybe other companies could take a hint ?
Concerned about cost, Microsoft is reportedly warning employees against wasteful AI use. In an email seen by 404 Media, Microsoft Executive Vice President Jay Parikh warned that individual divisions would be given targets and could face restrictions. "Tokenmaxxing is not what we are optimizing for. I want all of us focused on …
Or maybe...even though it seems a bit old fashioned today...they could work on making AI less resource intensive? Focus efforts and resources there.
The whole AI infrastructure build out feels like the developers have taken over...it's the typical "throw RAM and CPU at it" on an absolutely egregious scale.
If a Russian woman on a shit laptop can shave 60% off a pirated game using compression in clever ways, why can't a multibillion dollar company shave 10% off an AI model?
Perhaps they should start looking at trimming the fat from training data...does an AI really need to know how to potty train a toddler for example?
If AI is going to go anywhere, it needs to be far less generalised and it needs to stop with the scatter gun one (fucking massive) size fits all approach.
Last year Microsoft were telling their engineers that it was more important to be using AI than doing a good job, and presumably the ones who didn't decide it was time for early retirement have taken that to heart.
It's still kind of astounding to me, though, the way they're determined to shovel technical debt into their products like coal into a steam engine. It will necessarily and obviously have repercussions down the line and the effort to undo that harm will be vastly more expensive and time-consuming than doing things well the first time would have been.
If the answer to anything is repeatedly throwing your source code tree through the mincer one more time until it works or just throwing it all out and getting an AI to build something again from scratch then there is no technical debt... if all you do is build little noddy apps.
Most apps are noddy apps...there's very few "original" apps out there that aren't 99% the same as everything else, they may have a different coat of paint and a different dataset behind them, but they're largely the same...I'd be amazed if the majority of so called developers out there could raise their hand and honestly say what they've built doesn't share 99% of it's code in common with everything else that uses the same frameworks etc.
There are only so many ways to connect to databases, parse JSON, loop through an array to build a table etc.
The actual use case of a given app aside, how many of them use a homegrown database connection library, email library, json library etc...development has been flat pack IKEA bullshit for a long time now. Which is why AI can do it easily.
Pre-AI did you ever watch any developers do their thing? It was mostly autocomplete, intellisense and tab complete...using plugins like Emmet for VSCode etc...average developers haven't actually been writing code for a long time.
Every now and then you might come across a developer that will write code raw using a text editor in a terminal because they don't need the VSCode wheelchair to push them along...but mostly, IDEs have been doing a lot of heavy lifting for a long time.
I dumped IDEs a long time ago because I started getting sick of all the suggestions and the wanky incorrect autocomplete etc...as time went on more and more of that crap became default and I had to spend more and more time turning it off...until I realised all I wanted was syntax highlighting, a decent legible font, smooth scrolling and a tree view of the project, which you can achieve with gedit and a file browser.
I'm sure it's a matter of perspective, if the software is meant to be a 10 year solution and as the developer, you're only working on it for 1-2 years, then they probably would say there is no technical debt...because when the "LTS" framework they've used goes out of support after the long 2 years it's supported for, they won't be around and if they are its *sucks teeth* full rebuild mate.
The reality of course is that a development team might take 1-2 years to build a solution, usually on technology that goes out of support within at least 1 year of going live, at which point, they've fucked off.
I haven't met a developer for a long time that develops with long term use in mind.
It used to be the case that long term development would be done using lower level languages so in the future you can just recompile the code to work on newer architecture...this would at least buy you a decade until problems arise like libraries becoming defunct etc.
It's all "Javascript" these days...and by that I mean React and Vue etc...so not really even proper Javascript...90% framework, 10% code actually written by the dev.
I've recently had an uptick in requests to migrate a lot of backend tech to make it a lot more future proof because some places are fed up with "upgrade gymnastics" required to maintain a solution built in something like NodeJS...I've managed to port dozens of APIs to Rust with minimal third party libraries...any libraries I do use, or I am forced to use, I stash with the codebase so the project can compile without needing to "fetch" the libraries.
The first time I did this was about 8 years ago, I ported a crappy (and I mean crappy in the strongest terms, specifically because of the tech) Django backend project. I rebuilt it in Rust...it has seen numerous OS updates and database updates and it still compiles and runs...the maintenance on it is incredibly low, it's the occasional bug fix when weird edge cases are found essentially...the deployment pipeline for it is insanely simple as well...you drop the binary on a server, drop in a .env, whack it in a service, enable the service and you start the service and it's deployed then it will run until the box dies or is benched for going out of date.
There is still technical debt of course, but it's a manageable amount of technical debt...there isn't a massive scramble every 2 years to "fuck me, upgrade now, jesus we haven't done it for 2 years"...and therefore, no massive unexpected rewrites, compatibility issues etc.
Yes, paying a Rust dev is more expensive up front (they have dignity and know what they're worth and there aren't warehouses full of them in Bangalore) but they aren't expensive 2 years down the line, 5 years down the line, 10 years down the line because you won't have to rewrite everything from scratch. You just have maintenance...you might have to swap a library or two, or adjust the code to work with a new database authentication method or something...you won't be swapping out KungFuNinjaTosspotAwesomeSauce.js and doing a whole new dev cycle because it's no longer trendy for TaiChiJeetKunDoWanker1337.js because it is trendy and "everyone uses it now, it's the standard".
My fairly minimal Go back-end is an order of magnitude easier to maintain than my Javascript UI that relies on libraries that make breaking-change version jumps every six months. By this point moving the UI to the latest version of the core platform it's built on is almost a full rewrite. I don't think the people excited to add features to their tools really consider that not everybody using them can afford to be spending resources on constantly upgrading and migrating to stay current.
Quite right, a lot of the crap that's put out there is built using bloated frameworks because it makes it cheap up front and usually these cheap solutions are actually the proof of concept that just gets a bit of spit and polish to make it "prod" worthy.
Then the company operating the platform, taking payments etc neglects to base their pricing around what the development would actually have cost...only the hosting costs...so when a rebuild comes around, usually within 2 years, they have no budget for a rewrite...so they do the same shit over again...same cheap tools, same cheap solution, slightly higher versions of everything...over the course of a decade it eats into your profit margins and makes your company finances look worse than they otherwise would and your customers have to constantly put up with platform changes every two years...each time you do it, you're guaranteed to lose customers that got "used to" your platform and you have to gradually decrease your prices or leave them the same in order to try and entice new customers in, who get harder to find and convert when your platform is 10 years old and still shit...moreover, if you somehow manage to reach the 10 year mark, and you have 10 years of meagre growth behind you, investors will think your product is dead with no potential left in it...so you're fucked for an exit, investment capital etc.
You save nothing by spending less up front, they made considerably less money over time, they have company books that look shit to investors and customers that aren't loyal...it's dumb. Hands up how many of you are in exactly this situation...most people I'd imagine.
Cheap dev, results in a cheap product which results in a cheap business run by a cheap bastard who will eventually be taken advantage of by a nefarious techie/dev somewhere who has them over a barrel.
Sounds like the usual managment method to bullshit a budget cut.
Cut training budgets = "employees are now now 'digital natives' and no longer require training"
Cut patching budget = "technical debt is a myth we don't need to patch or update anymore"
None of it true but it suits the management to believe it.
At the huge mega corp I work for, one of those near invisible ones hardly anyone has heard of, we’re currently in experimental mode, but a fast tracked one.
A lot of devs here are super aware this free lunch is going to lurch abruptly to a halt.
My take is this is when the redundancies will occur.
Rather than blame silly token over usage on deterministic tasks, there’ll be a headcount reduction.
Can’t have the coolaid drinking C-suite having egg on their face over rapid adoption, can we?
Anyone not onboard due to “using it wrong” will be ejected.
A few years later? They’ll be crying out for tech talent to dig them out of a shit storm.
Computer resources left unused are wasted, so the first step is to encourage users to use more by providing access nearly for free. Once there is enough usage that resources are in short supply, this makes the second step seem fair: Raise usage rates to allow those who place more value on the resources an ability to get them.
In other words, Microsoft sent an internal memo about being more efficient with tokens simply to validate the price increases for everyone else.
That happened to us at work with OneDrive.
A few years ago we were encouraged not to keep much on our local machines or fileservers, but to stick everything on OneDrive.
Last year the memo came round. Essentially it said 'Everyone take your OneDrive use down to below 200GB, or Microsoft are going to charge us shitloads to use it'.
Doesn't surprise me that they'd do that with AI tokens too.
At this point everyone knows he's naked, but it's no longer enough to simply claim the non-existent tailoring work is highly fashionable and enviable, now you also have to point out which of the imaginary fabrics are so stunning that it makes your own work measurably more productive and fervent. But don't compliment too much, that's wasteful!
Sure glad we're progressively burning the planet for all of this
I have tried using Copilot within Microsoft's own products to do things like fix formatting throughout a document, and my experience is that it will claim that it's done something that it clearly has not, or it will execute poorly. By contrast, Claude will quickly do the right thing (again, in my experience--put down the torches and pitchforks, please). I can't even imagine using Copilot for coding: best case, you get nothing; worst case, you get something worse. I guess if you're a Microsoft engineer trying to look busy, an agentic coding loop in Copilot could be very helpful; otherwise, one can only imagine what kind of crap you'd wind up with. 0_0
I mean, co-pilot did the correct operation for once, as one valid solution to the problem of proprietary software, is to delete it (besides, practically that's just short circuiting the lifecycle anyway).
Copilot will eventually proceed to delete the whole program and any backups too I reckon, solving that problem completely.
This post has been deleted by its author
The original comment was deleted, but magically it's not copyright infringement if a LLM is used to do the copying?
What's the difference between such process and re-encoding a hollywood movie for H.264 to AV1?
After all, the bytes on disk of the resulting video file are completely different and the output pixels are somewhat different too - must be fair use right?
Shouldn't the "super thinking machine" be able to do simple tasks like packing a dishwasher, provided a robot body (with a camera and arms), if it had any intelligence?
Currently, it seem such programs can only degenerate bad copies of pre-existing text?
It sounds smart. But think about it and you should realise it's just another version of the tech singularity, except with money. And I don't think rich people would ever allow us access to a machine where anyone could go, "Claude, make me a billion dollar business."
So you’re telling me that M$ and in extension GitHub are openly proclaiming their “AI-first” approach, yet this publication is still seemingly standing behind them and proving material support to their cause? And people are still buying their products, using their services? Seems like a shame to throw ethics and morality into the wind just for a paycheck, working against your fellow man, supporting the interests of “The Man”. We’ve come a long way, baby, in the worst sense of the phrase.
Threatened with the sack by management if they didn't use a lot of unreliable AI but still expected to produce solid work, they did the logical thing.
They did the work themselves as they always had done and burned as many tokens as they could getting the AI to translate the great works of literature into Japanese haikus.
Yet still management complain! No pleasing some folk.
I do honestly long for the day when companies start to think about measuring employee productivity in terms of their actual output - and factoring AI usage into that. As in, if you have two employees, both producing a similar output and yet one of them is burning through tokens and the other isn't. Clearly the second one - the one who is capable of doing the vast majority of his/her job effectively with minimal AI usage is actually providing better value for money.
As I've said all along, if AI is that beneficial, those that use it more should produce more - and it's this combination of output balanced with AI token consumption that should determine an employee's true value to the company. The current fascination that "those who use the most tokens are the best employees" is a deeply flawed methodology that needs to die...
When it is mentioned in the warm and fuzzy chat about your shit pay rise that your AI use is too low to give you a better rise, then Tokenmaxxing starts to become more of a focus to you than being good at your job.
It's amazing how long AI takes to do relatively simple tasks that used to take me moments to do, but at least I'm shooting up the AI use league table.
Oh, and sorry world for all the electricity and water use and all the waste heat I'm now responsible for. I'd love to not be a dickhead, but a bigger dickhead than me has decided this is what is best. The company is BCorp, but the effects off this diktat about AI use seems to have been lost on the management. Or maybe they just see BCorp as a marketing thing...
I've reached a point where my pension pretty much matches my working salary. Up to now I enjoy my job and have avoided any vibe coding, so haven't left, but am already planning my retirement to be before I'm forced into using all this bollocks.
Anon, obviously, my boss doesn't know yet!