The Tokenpocalypse Is Here: Companies Are Scrambling To Stop Spending So Much on AI | Leaked audio from Accenture says a big source of AI token ‘chewing’ is people just converting PDFs to presentatio…
https://www.404media.co/the-tokenpocalypse-is-here-companies-are-scrambling-to-stop-spending-so-much-on-ai/Open linkView original on lemmy.today441
Comments120
When you tell the people to mash the easy button and then they do exactly what you wanted. They created the monster.
Eat shit you greedy corporate assholes, i hope all of your companies are damaged beyond recovery, you useless fucking tools.
This is 100% the thesis I've been shopping personally... Many Boomers and Xers in executive positions had a "magical millennial" that they quietly kept as a secret "AI" to split/edit PDFs, set up an Airtable base, add columns to a google doc, etc. There was a tacit, silent agreement in this symbiotic relationship for the bulk of the last 20 years - you'll make sure I don't look completely incompetent in tech matters and I'll backchannel on your behalf to senior leaders and people who "matter" to help you advance.
Gen AI essentially allows the laziest input, gives a half competent output that "feels" fine and has the bonus of telling the boomer/Xer that they are actually amazingly capable, and could have done this themselves all along even, but they rightly delegated the task to their magical millennial, and now to the AI of choice.
So they fired all the magical millennials, because they knew too much about the before times. Now that they are fucked without a life raft, costs soar and they will cling for dear life because they will be exposed otherwise.
Edit: through a twist of fate, the iPad kids grew up technically incapable and relied on the magical millenials as well. They could only offer praise and loyalty really, or a boomer, Xer recruited them in and talked the MM up as a "wiz" to seek out. Anyway, now that the MM are gone, the Zoomers and gen Alpha kids only have one strength remaining, the old people have no idea what they are doing or how to quantify their success, outside of "use more AI". So the fragile balance remains for now, with a vulnerable, hollow center where the magical millennials used to live.
This post was a fuckin rollercoaster, well said
Using an LLM to parse stuff is like using a rocket launcher to kill an ant.
You can accomplish the same thing using a million times fewer resources with a purpose-built program.
Well, if your KPI is how much you use an LLM like in some reports - this is an easy way to get those good indicators. Also, LLMs are super easy to use to parse things, whereas many special programs like IDK grep isn't exactly user friendly. Not to mention not finding patterns really. Though here I'm thinking things like looking at various logs on computers.
Maybe we could at least nudge the LLMs in the direction of suggesting appropriate tools and giving hints how to actually do that when applicable instead of blindly brute forcing every task. Of course that does not solve the issue of stupid corporate incentives, but I feel like by now most companies have realized that burning as much money as possible is not a good goal.
Yes but you're in an open office and all you have available in your open office hell are: rocket launcher, your sanity, Debra's snappy attitude.
These stupid companies are getting everything they deserve and I'm loving it.
Nah. The bailouts will only be two or three tomes what they squander.
Gonna bump my gaming rig to 64GB as soon as these asshats get what we've all known was always coming to them
What a fuckin’ clown circus…
You know, if we were taught to use something like LaTeX, this wouldn't be an issue because of BEAMER
That wouldn't fix the issue though. The problem seems to be that most people only put out a PDF file when sharing slides, and never end up sharing the source file (the
.pptxor.texfile).This is likely because PDF became the "file that everyone can open", just in their web browser. It's the next best thing to a web page for non-techie consumption. Yes, there's no reason people can't open pptx in most cases, but I bet various endpoint protection and just not understanding how to even pick the right program to open the file steps in.
I'll take typst any day over latex lol
Me reading the intro in their docs:
"Ok so this is just like Typora"
"But it uses weird symbols for everything"
"Oh wait, now they're referencing a figure…"
"Holy shit it can make a bibliography"
This is a pretty awesome tool at first glance. I'll have to check it out further.
And Accenture itself reportedly started requiring senior staff to start using AI or risk losing out on promotions.
Every time companies urge employees to use AI and then regret the cost. The fuck is wrong with people? Why are they pushing it so hard? Does Sam give them hand jobs if they use the most?
I don't understand this need to pressure staff into using something and threatening punishment if not. Are they worried that their employees are not efficient enough? Pay them the token prices on top of their salary and see how stuff changes.
This is actually very common across businesses. My company actually has our bonuses tied to AI adoption, so we have dashboards showing people's AI usage. Other major companies have done the same, which lead to the practice of "token maxxing" where people were using AI to make more AI calls to boost their numbers up.
Amazon did this and regretted it, canning their leaderboards It's crazy to me that this is considered normal. Please use this fair dust product that will eventually replace you, kill the planet, and make some douchebag rich.
Tales as old as KPI.
That's the corporate hive mind, all afraid of missing out of a great productivity tool. And they think that because media these days just copies what the richest people say and hype it up because the rich these days only speak to yes-men.
Reality doesn't need to obey yes-men.
Fuck Accenture, they just abuse low level analysts to make profit.
We got the hell we deseve :D
This dumb bubble can't pop soon enough and wipe away most of the US economy while doing so.
True except the 80% of people who will suffer further because of it, and the 20% that will gleefully be laughing to the bank and buy up everything at a discounted price. Few of us should be cheering for the failure or success of AI.
Look I'm getting stabbed in the gut regardless let's just get it over with.
If it's going to fail I'd rather just get it over with.
Sounds the people they hired to do the shit work don't actually want to do that type of work. I, for one, am shocked. Shocked I tell you!
The problem seems to be that it takes competent employees to get anything useful out of an LLM in the first place. However, it is these very employees whom the greedy CEOs want to replace. So the result is that an incredible amount of money is being spent on absolutely nothing.
The logical conclusion, then, should be that it would make more sense to replace these useless CEOs with AI. Since they’re just making idiotic decisions for a lot of money anyway, there could be lots of savings.
Unfortunately, however, that will never happen, because contrary to all that talk of KPIs and such, what really matters in the upper echelons of management is never efficiency, but rather ruthlessness and brown-nosing.
I recently learned of a term "reverse centaur" and found the ideas in that article were very entertaining as a way to explain why CEOs are obsessed with LLMs. The author of the book being interviewed has a few hot takes that I think are pretty relevant here too. https://arstechnica.com/gadgets/2026/06/how-to-burst-the-ai-bubble-strike-at-its-roots/
Indeed!
I think that pretty much sums it up.
Doctorow is the Cassandra of our time.
AI is about to join the list of "stupid technologies that people should really wait and see before investing on", which includes
Folding Phones Sales Continue To Increase since 2019, showing people seem to like them
Two time buyer, can confirm. They're legitimately useful and durable enough for me.
I like the one that folds like a GBA SP.
Not that I'm gonna buy one based on that.
But if I'm choosing between two identical models but one folds (and I guess it isn't that much more expensive), I'd go for the folding one.
Except, of course, for the fact that the next phone I buy will be Commodore's Callback.
Two time buyer as in you liked it so much and bought another, or two time buyer as in the first is already inoperative and got another?
Not the same guy. But I've been using foldables since 2019 as my primary phone. Samsung fold 1 -> 3 -> 5. No case, no protectors, keys with phone in same pocket. Never had an issue.
My fold4 lasted 3 years, and still works. I'll concede it was only able to unfold to about 85% open at the end when I decided to upgrade. I was happy to upgrade to the fold 7 which has a redesigned hinge and better dust protection. I acknowledged the risk of mechanical failure as a possibility compared to a slab phone and after using it for years I decided I was still very impressed with the flexibility of having a tablet in my pocket. The fact that my 7th generation is significantly thinner and has a 200MP camera compared to the 4th is what sealed the deal.
It's not perfect for everyone, but I wouldn't go back to using a slab, there's too much functionality I'd be pissed giving up.
Granted, I tend to upgrade my phone every 2 years or so anyway.
I find only 3 years of durability to be absurd 0_0
I don't think they are for me but I honestly would not include them in that list above. First of all, there is no investment bubble around them and secondly some people seem to like them and are ready to pay for them. They also do have legitimate benefits (but also downsides)
Yeah I didn’t dislike the 3d monitors/tvs they just had too many caveats at the time and VR kind of ate its lunch.
3D stuff has the fundamental issue that VR and 3D views are just incredibly straining on the human user. Folding phones have no such issue and their durability is also good enough to be competitive (yet clearly worse than regular smart phones). They are really not comparable to 3D monitors and VR.
I think the strain is really dependent on the person as it never really bothered me, it definitely is a problem for some people though. Either way I don’t think it belongs in the same category as blockchain bullshit
It is straining for everyone but yes, some can handle the strain better than others. It just takes so much more mental energy to handle. Most can't handle it well though, which is why those technologies never manage to break out of their niche.
I do agree however that it is quite different from NFT and other scams. It is a really fascinating technology with real use cases but just some foundational issues that prevent it from leaving their niche.
If it lasts 5 years, maybe its durable enough.
I just replaced my S21 which I bought new, and it still works fine for everything but being a phone because it had a short or blown cap or something that kept the SIM reader from working.
This could also be explained by the entire rest of the market being different flavors of the exact same thing. A little over a decade ago, we actually had choices in what type of phone we wanted. Now, if you want anything other than an identical slab, foldables are your only choice.
I wish someone would make another widescreen slider already. Droid 2 was the shit at the time.
Disco Stu says disco record sales have trended upwards for the entire decade of 1970's. If this trend continues... eyyyyyyyyy!
Don't forget NFTs... though I guess that can fall under blockchain.
yeah but NFTs were so stupid that they deserve their own dishonorable mention here.
NFTs are so fucking stupid the satirical yearly award for this shit should probably be the NonFungible Trophy cus it's a literal fucking physical trophy.
It's just the same as crypto
Some of these turned out to be useless, some just haven't been adequately delivered yet.
In 2040 folding phones and autonomous cars might be great, but blockchain will still be a solution in search of a problem.
Problem: You can't walk from America to Europe.
Cold fusion. Green hydrogen. Hyper loop. Just off the top of my head.
Hyper Loop was an insane idea (in a bad way).
Motherfucker wanted to invent subway, but for cars, making cities even more dependent of cars.
hyperloop was very low air pressure subway tubes for high speed trains. Works for research tesring tracks impractical to infeasible for real world applications. But it apparently what killed high speed rail in california (because wait this will be better)
The cars in tubes thing was a seperate stupid idea.
I don't know that he really wanted to fully execute on Hyperloop so much as build hype (and Tesla stock price) around the idea while sabotaging funding for California's high speed rail project. But yes, end goal to keep people buying his cars.
Average massively shitty car CEO idea
Cold fusion is a scam, not a bad idea. There is no scientific basis for that. Is as bad as “infinite energy engine”.
All of the things I mentioned are scams lol
Green hydrogen is water electrolysis with solar power, not scam, it works, it is just a question of making it economically viable. Hyperloop is just stupid, but within the realm of possible. Cold fusion is the only scam here.
The scam isn't always "fake tech" so much as "distracting from better alternatives for monetary reasons". Even if hyperloop worked perfectly, they're not going to build it, because not building high speed rail was always the point.
Your definition of “scam” is not what other human being use. Causes unnecessary back and forth.
Oh, put a sock on it.
This one I slightly disagree with. I got my headset on a black friday and it was super cheap, but VR documentaries are friggin' amazing and I hope museums will invest heavily in it in the coming years.
Definitely. Makes me feel good for people who make their living driving trucks.
The metaverse isn't VR in general, it was meant to be a virtual space in VR where users could be advertised to and buy/rent things and space like in a physical city.
It failed because those were the intended starting points, and it didn't solve any problem other than a shitty attempt at a "I want to live in a ready player one world" and didn't have any compelling reasons to actually use it, let alone use it and pay ridiculous amounts to do interesting things there. They always just wanted to be the middlemen, offering space for others to pay for and do something interesting in. The most interesting thing they came up with is having a meeting with avatars instead of faces on a screen (and most people don't even want to turn on their video and just do a voice conversation instead).
Plus Second Life already tried the same idea and failed, and did so without requiring several hundred dollars of specialized equipment per user like the Metaverse did.
I showed multiple morons how to doctor up their safety photos using copilot in hours that it cost the company i contract for insane money. I dunno if it she's but I'm doing my part?
Further evidence that most CEO's and SLT's are just confident, lucky, and born wealthy; not competent, smarter, or much better than average.
Also they are addicted to sycophancy. Now that they have a convincing yes man in their pockets they are uniquely vulnerable to AI psychosis.
I don't get this, We've had OCR for a while. All around San Francisco Ive been seeing ads for "llamaparse" with the tagline "we parse pdfs", like is that all you do? How do you afford this marketing budget?
PDFs are so shitty to work with, it's like translating them, it's impossible without using a tool like Google translate.
I fucking hate PDFs as much as I hate Adobe.
I HATE PDFS TOO. I hate them! 99.999% of the time I'm given a PDF file it would be more useful as an HTML file.
All this money that is being spent, where does it end up?
Various designing firms.
After the bubble bursts, law firms, because everyone starts suing everyone to get back money they didn't necessarily have, but still spent, because somebody will soon sue them.
Well at least there's a happy ending
Nvidias pocket maybe. CEOs? Billionares?
To build new data centers
Leadership are mostly idiots. They don't know how work gets done. They think like "wow it produced so many lines of code!" and don't know that's not a useful metric.
My job had Microsoft do a four hour copilot demo for the entire team last week. (Surely an immense expense. They won't pay for most people to be full time with benefits). The guy used copilot to make a regex to parse html. A little surprised zalgo didn't show up.
I'll one up you.
Google's "principal evangelist" for Gemini did a 2hr presentation for almost 1,000 people at my company on Google Meet a few weeks ago.
40 mins late due to nobody knowing how to get more than 500 people in a meet.
He then spent 90 minutes showing everyone how to do shit that literally everyone knows how to do with AI nowadays; touch up old photos of your grandparents, make shitty marketing mascots and logos, etc...
Here's an image I took during the presentation. For a laugh, look at the background text:
A C-Suite exec got impatient and asked when we were going to cover how to use agentic tools and he said "oh I've got a little section at the end for that." Ffs
Literally nothing useful happened during this meeting and nearly 1,000 employees were idle for no reason. A complete and utter waste of fucking time. Gemini dude didn't even bother to consider what use cases we might be looking into, he just did a boilerplate ELI5 of AI that would have been outdated in like 2022.
Absolutely insane.
ROFL. Thanks for the screencap, that’s priceless.
Lastly, poor Buc-ee the beaver deserves better.
CEOs them to make power points using the already pre built company presentation template then talk about how they couldn't have made it without AI like that's supposed to show how useful AI is and not how incompetent they are
Yeh but the people trained ‘number’ widget went up on their numbers dashboard. Good reason for another self awarded bonus!
Did they think everyone was gonna be creating return value based products for the company to make a fortune on?
Instead they making posters and having crazy convos.
It’s like someone has just discovered salt and every restaurant has it. Each thinks they are unique and gonna be the GOAT.
Can't read the whole article bc paywall... But if they are really worried about token cost for converting PDF to a Powerpoint, they ain't seen nothing yet. Agentic coding the way AI companies push it (multiple agents in parallel with Claude, looping etc) uses way way wayyyyy more tokens than this.
Archive link: https://archive.ph/c6DSo
Have you used Opus 4.8 at API costs? Without using agents, I can burn $20 an hour no problem. I use Kimi K2.6 and GLM 5.2 these days.
How do you use it without agents? Just prompt the chat interface?
When I used claude code, I'd go through the smallest tier subscriptions 5h limit in half an hour sometimes if I used Opus lol, reckon at API pricing it would've been 100 bucks an hour
Chat prompts and then accept the changes when they are worth using. LLMs are entropy apparatuses, injecting that shit straight into your code base is crazy. It took one module/subsystem done with agents for me to realize that it was completely unmaintainable. Not a single eng can make a change to it and the only option is to rewrite. So I banned agents. I know some people still try to get away with that style of LLM coding but I encourage people to instant deny a PR if it looks like garbage.
If you accept changes then you're probably using an agent though?
Agent mode is disabled, only chat mode works. This is with GitHub Copilot harness in JetBrains Rider or Visual Studio Pro. Agentic coding means it can make decisions without a human.
I used to understand what computers were for and how they worked.
If you just throw every task at it, I can completely imagine that it'll cost you more than any gain in efficiency. To some degree I think this is true of some coding tasks (not all, most are pretty efficient pattern matches, and that's what it does well).
But using LLMs to build tools and pipelines that stand alone (no AU built into it) and enable human productivity seems like a far higher leverage use.
The programming version is building the libraries and abstractions that are robust and well tested, so that regular developers can quickly build and refine the features.
Or building the reporting dashboard. Or whatever.
The cost is only going to go up, and the companies that lock in some non-AI process improvements before the hike will likely be smiling
Especially because a statistical model of language has very limited valid usecases. Many tasks that people use LLM for do not make any sense and cannot be accomplished by a statistical model of language. It can only output a statistical approximate of what a solution to the task might look like, not an actual solution.
It constructed an entire feature-full library for a UV sensor that just came on the market, by me just throwing the PDF datasheet at it and saying "make library pls".
It's the actual solution. Maybe the statistical approximate just coincidentally lined up with the actual solution? Either way, works for me. Second time it did that, too. Worked for a LiDAR sensor earlier.
Yes, right, and this can happen. I didn't say they are a bad approximation. LLMs may be the most advanced and sophisticated statistical models ever created (if there are other examples of statistical models that are more sophisticated, I would love to learn about them). But given what an LLM actually objectively is, a statistical model of the next token to follow from a sample of language, what other explanation could there be?
We need to keep in mind what the tools that we are working with actually are.
As a code generator, they can produce great results, especially simple stuff like generating a script or some function implementation. Once you get to software engineering tasks like designing system architecture and designing maintainable code, it starts to fall apart really fast. You end up doing all of the work for it in natural language and just using the LLM for a usecase that it is actually great for: translation, from detailed spec to code.
https://archive.ph/20260625141336/https://www.404media.co/the-tokenpocalypse-is-here-companies-are-scrambling-to-stop-spending-so-much-on-ai/
LLMs as a technology are incredibly powerful tools that have so many useful applications. The issue is trying to turn them into this all powerful omnitool that could do everything, and then shoving that product up everyone's ass at every turn.
There was always bound to be a correction to this bubble, because the current model is unsustainable. You can only keep making unreasonable moves for so long before everything starts to crumble and your forced to scale back.
its not a presentation
its a "deck"
i fucking hate these people
Ever since LLMs where provided free to the public, I've been pen testing them. Not intentionally, I just like tech and have wanted to explore them. After getting all the info they're designed not to provide(which is so fucking easy, even on the most recent models that have corporate PC baked in to the core can't do shit) I turned to token burning. I could easily get 1M/24hr, but they've brought that cost down to like $0.10 so its not doing much anymore. I think I saw a model specific script that's doing 52k/min.
Point is it takes a day of coding and a shit PC to just sit there burning money directly out of their budget. It wouldnt take that many people committed to doing this to make AI less feasible.
Sorry bro, but ai budget should come from the salary
relevant Pic though, as Ai has really boobiefied padme. like GOTDayum
No, from yours
What a stupid take. Why should employee then even consider using it, when the cost for it would directly reduce their salary?
Exactly
But the employers want their employees to use it, (because of some unfounded promises about performance boosts). So I don't really get your point.
Employees will be motivated by doing less actual labour
Personally I do it to clock more hours than I've actually done, why else?
Well that's just working-hour fraud, you can do that even without using AI.
Is it? I'm spending my own money on 4 currently bottom tier AI subscriptions (fucking hell I'm not gonna pay 200€ a month for one sub) that I'm trying out and working on 3-4 projects in parallel. Lots of experimentation been done in my own time to get things working nicely too. I won't say I deliver 3-4x as much work in the same time because obviously I don't, I have to double check everything and with small things the AI is slower than I am... But even just 50% more work is 50% more money.
The alternative is going and saying "hey I want 50% more money per hour because I deliver 50% more work in the same timeframe now" and my employer telling me to pound sand because that puts me over what they charge for my work when they rent me out. Instead I bill them more and they bill their customers more and nobody needs to raise their rates.
This probably is a best case scenario for AI usage being a productivity boost, I'm not forced to use it (and often don't), my employer/customer is okay with me billing extra hours IFF I'm getting things done in less time than reasonable estimates. Their customers still get the things done in the costs (hours) that they agreed to.
Well that is such a unique case. I've never heard any employer being okay with clocking hours that were not actually worked. And in your case it's essentially just a raise with extra steps.