Workplaces Look For Cheaper AI As 'Tokenmaxxing' Fades As a Corporate Fad (apnews.com) 87
An anonymous reader quotes a report from the Associated Press: A corporate fad of "tokenmaxxing" on artificial intelligence technology is hitting its limits as workplaces throwing AI at everything are seeing the costs rise without a similar spike in productivity. What started as tech industry-fueled springtime hype over squeezing as much AI-generated work as possible out of products like OpenAI's ChatGPT and Anthropic's Claude has shifted to a summertime backlash. [...] Just a few months ago, Silicon Valley executives were promoting high token consumption as a signal of high-performing employees. The stereotypical tokenmaxxer was staying up late -- perhaps ignoring their significant other -- while orchestrating an army of 24-hour AI agents performing work on their behalf. [...] The trend boosted revenue for leading AI large language model developers like Anthropic and OpenAI, but it fizzled as it became apparent it wasn't necessarily the best strategy for everyone else.
[...] Bain & Company management consultant Jue Wang said many of the big businesses her firm advises have been taking a closer look at returns on their AI investments. "The token cost for them has been doubling, almost every other month," she said. "Let's say $200 per developer per month. Multiply that by 20,000 developers, which is often what we're dealing with at these companies, and that quickly gets you to a number that is not a line item that any general manager has planned for." Sometimes that just means not using the AI equivalent of a sledgehammer to crack a nut. "Not everything needs a Claude Opus 4.6," she said of one of Anthropic's more capable models suited to software engineering or deep research. "And yet you see so many companies, so many users, default to using Opus for everything, including generating emails." That's led to a search for tools that do AI "model routing" -- in which easier queries get automatically sent to cheaper and more efficient AI systems and more complex tasks go to more powerful models.
[...] At the same time, those who favor racking up as many tokens as possible are having a field day with new open-source models from Chinese startups like Moonshot's Kimi or Zhipu's GLM, which nearly match the capabilities of top U.S. models at a fraction of the price. "There is some validity to the theory that this could push tokenmaxxing a little bit further," said Raffi Krikorian, the chief technology officer at Mozilla. "But if we look at the industry overall, I think it's realizing that tokenmaxxing is a dumb thing." It's similar, Krikorian said, to how software companies once considered how many lines of code a programmer wrote to be a good metric of productivity. That later fell out of favor. "I think tokenmaxxing is moving through the exact same pattern," he said. "I think this is going to be an interesting blip that we're all going to look back to laugh at in a year."
[...] Bain & Company management consultant Jue Wang said many of the big businesses her firm advises have been taking a closer look at returns on their AI investments. "The token cost for them has been doubling, almost every other month," she said. "Let's say $200 per developer per month. Multiply that by 20,000 developers, which is often what we're dealing with at these companies, and that quickly gets you to a number that is not a line item that any general manager has planned for." Sometimes that just means not using the AI equivalent of a sledgehammer to crack a nut. "Not everything needs a Claude Opus 4.6," she said of one of Anthropic's more capable models suited to software engineering or deep research. "And yet you see so many companies, so many users, default to using Opus for everything, including generating emails." That's led to a search for tools that do AI "model routing" -- in which easier queries get automatically sent to cheaper and more efficient AI systems and more complex tasks go to more powerful models.
[...] At the same time, those who favor racking up as many tokens as possible are having a field day with new open-source models from Chinese startups like Moonshot's Kimi or Zhipu's GLM, which nearly match the capabilities of top U.S. models at a fraction of the price. "There is some validity to the theory that this could push tokenmaxxing a little bit further," said Raffi Krikorian, the chief technology officer at Mozilla. "But if we look at the industry overall, I think it's realizing that tokenmaxxing is a dumb thing." It's similar, Krikorian said, to how software companies once considered how many lines of code a programmer wrote to be a good metric of productivity. That later fell out of favor. "I think tokenmaxxing is moving through the exact same pattern," he said. "I think this is going to be an interesting blip that we're all going to look back to laugh at in a year."
The only thing I find it genuinely useful for (Score:4, Interesting)
From what I understand if you put a bunch of time into learning how to prompt it it can generate mostly workable code. I find when I ask it to generate code though the code inevitably has a few mistakes and bugs. And with some of the Python code I've worked on porting from version 2 to 3 it gives you some of the most bizarre code I've ever seen. Like a mishmash between the two versions. The default python conversion tool handles most stuff anyway so I was only really hoping the AI would handle the handful of things it doesn't but no dice.
On the other hand if you don't know how to program I suppose it could be useful. And I've seen some people do impressive things in the retro game programming hobby. Supposably the guy working on the mortal Kombat 2 3DO Port used AI to Port some of the arcade code over to the 3DO. Ironically it was the code for controlling the computer AI
Of course the last thing you would want for a port of mortal Kombat 2 is arcade accurate ai. Mortal Kombat 2s AI was the cheapest and shittiest imaginable
Re:The only thing I find it genuinely useful for (Score:5, Interesting)
I'm mostly working on a project with a million lines of C++ and Python code, much of which is 20 years old. It's very useful to be able to ask the AI to explain how some feature in the code works because it can usually figure that out in halt the time I'd take... though sometimes it will come up with some completely wacko explanation and I have to ask it 'are you really sure about that?' so it will go and take another look and figure it out.
And for code where I know how the changes should look I can typically tell it what needs to be done at a junior-dev level and it will go off and do it much faster than a junior dev would.
Today I was estimating the time required for a bunch of bugfixes and was able to get it to do the grunt-work of analysing the code to verify that I hadn't missed anything important that would blow the budget. It also found some danger areas where we really shouldn't change the code if we didn't have to (because the system has to stay up as close to 24/7 as possible and the change would require an outage) so I gave it an alternate plan instead.
It's definitely useful, but we're only seeing maybe a 10% increase in productivity on average because most of our time isn't spent touching the code.
It's also nice at home to be able to tell my slow local AI setup to make some changes to my personal code and come back an hour or two later when it's done.
Re: The only thing I find it genuinely useful for (Score:3)
I spend most of my time on strategy and concepts, I estimate that my coding throughput with Chat is roughly double now but mostly bacause I've already laid out a design for it to follow, I'm not guessing or reworking much.
How we are subsidizing your habits. Let's stop. (Score:1)
Today I was estimating the time required for a bunch of bugfixes and was able to get it to do the grunt-work of analysing the code to verify that I hadn't missed anything important that would blow the budget.
Ah, so you are one of the people that we-who-are-not-using-AI are being forced to subsidize. You're talking about your own budget, but how about you pay me for the delta you caused on memory prices? There is no such thing as a free lunch and I want my money.
I'll reach out to you offline, I'm looking to get a new xbox.
Re: (Score:2)
Re: How we are subsidizing your habits. Let's sto (Score:2)
Ah, so you are one of the people that we-who-are-not-using-AI are being forced to subsidize. You're talking about your own budget, but how about you pay me for the delta you caused on memory prices? There is no such thing as a free lunch and I want my money.
You aren't forced and you aren't subsidizing anything, you're being priced out by people who want something more than you do and they're paying for it. It's supply and demand. It's like living in a gentrifying neighborhood, I don't know what else to tell you but nobody owes you a damned thing here. When people with more money than you do buy up all the inventory that's what happens.
Jesus dude, if we were actually talking about housing prices or closer to the issue, electricity rates, I'd have a little empat
Re: (Score:1)
Re: (Score:2)
Re: (Score:3)
Yeah, I have the same as well. I was given access to the company hosted AI models (because we don't want to let source code leak) and while some people attempted to vibe code their bug fixes, I've been using it to ask far more pointed questions.
And it was impressive - when a code merge resulted in a JSON file not building, I asked the AI to help me figure it out and it managed to find the issue in about 5 minutes, while for me it would've taken far longer. I later looked at the reasoning chain and realized
Re: (Score:2)
Heh, a system designed to parse huge data sets and compare/contrast search strings is really good at compare/contrast searches.
Re: (Score:2)
Writing dos batch files to do specific tasks. I've been using Dos since the 80's, and have a workable knowledge of batch files, but I have to look up more complex commands and syntax.
Pasting in something I've written for a grammar/typo/consistency check (knowing full well it's probably adding that text to its training, so nothing important.). It may not be perfect but it's light-years ahead of grammar tools I've briefly looked at in the past. Mostly it can catch s
Re: The only thing I find it genuinely useful for (Score:2)
Actually it really is terrible at checking facts. It's a probability engine. Try asking it anything about music. This is the kind of result you'll get. Pure hallucination. It can't even figure out what key the works is in. And recommends a recording that doesn't exist.
https://chatgpt.com/share/6a6873a5-b874-83e8-bc1d-5e21a302769a?ogimg=plain
Re: (Score:2)
Re:The only thing I find it genuinely useful for (Score:5, Interesting)
I hate creating Powerpoint slides, so I find it genuinely useful that I can type a few sentences or paragraphs into Copilot and it will do 100% of the slide creation and revision. It'll even pull content from documents.
I recently told it something along the lines of: pull a list of the APIs from this specification and create a slide summarizing the purpose of each API. APIs X and Y aren't mature yet, indicate that on the slide. API Z is the current area of focus.
It created a very professional looking slide with all the APIs summarized and color coded, with a legend for the color codes. A few more paragraphs of text resulted in more slides providing a solid overview of what's been done, what's in progress and what our targets are.
Only a few people need to actually read the API documents from start to finish. A much larger group of people need to be kept informed as quickly and briefly as possible of what the status and progress is.
I can verbally describe something to people, but they'll forget it immediately. I can type up a lengthy text description, but they won't read it. If AI can turn my text into a picture, they'll likely remember it.
Re: (Score:3)
From what I understand if you put a bunch of time into learning how to prompt it it can generate mostly workable code.
So can putting a bunch of time into actually coding something, with the bonus that you actually understand it 'cause you actually wrote it.
On the other hand if you don't know how to program I suppose it could be useful.
Maybe as a learning tool, but if you don't know how to program, how will you know if the generated code is any good?
I'm shocked! (Score:3)
Shocked, I tell you! I keep thinking this whole AI thing must be modeled on a pump 'n' dump scheme.
Re: (Score:2)
All AI hypes are. This one is no different. It is just larger and even dumber than the ones before.
This is a good sign (Score:2)
Maybe they hype bubble is starting to burst.
Yes, AI is useful for many things. I genuinely *like* having AI write most of my code. But the pipe dream of orchestrating a bunch of AI bots doing the work of a whole team, is just a pipe dream. Humans can't manage that much work all at once, and neither can AI.
The point though is this: If companies are finally starting to scale back on AI expenditures, it will force a reckoning in the AI world. The AI companies won't any longer be able to just charge whatever th
Re: (Score:2)
"a bunch of AI bots doing the work of a whole team, is just a pipe dream."
You're not doing things right, then. Or very well. It's pretty low intensity to author/implement 5 substantial features a day, test, fix the bugs, run QA, and do 10-20 odd commits. Give or take, given the complexity of the plans/features and how much they let the LLM fill in the blanks and how thorough they were in scoping the implementation. Maybe 100M-200M tokens, depending on your model/skill/skills/etc. More, if it's a language li
Re: (Score:2)
It's pretty low intensity to author/implement 5 substantial features a day, test, fix the bugs, run QA, and do 10-20 odd commits.
Sure, you can do that. But here's what you can't do in that day: read through all the code it generated, understand how it works, and confirm that it's cleanly written and doing the correct things. You just trust that it's ok and check it in. Keep adding to the pile of buggy, incoherent code that no one really understands.
Hopefully no one will ever be expected to maintain that code.
Re: (Score:3)
Ah, but they'll just get a future AI to maintain the code, so it doesn't matter that it's a pile of spaghetti that no human has ever understood.
Some of the people I work with have been getting the AI to write the code and then periodically getting it to refactor everything into something that isn't a pile of spaghetti, but if you do that then suddenly you lose much of the supposed productivity gain of letting the AI write the code.
Re: (Score:1)
Sure, you can do that. You can do that with buggy code that compiles, or python, too.
For AI, you can also have another agent and/or model review and fix it. You can ask it adversarial questions until you get it right. You can ask it to explain the architecture and design (or: you can plan it out ahead of time so that you know the design and then verify it).
It's no different than starting on an existing code base... just faster.
Re: (Score:2)
So it's turtles all the way down [wikipedia.org]?
Just get another AI to review the code generated by the first one. Because if you don't trust the first model to understand the requirements and implement them correctly, surely you can still trust it to understand the requirements and determine whether another model implemented them correctly! If you want to understand how the code works, just ask the AI to explain it to you. No need to study the code to determine how well the explanation matches what's actually implemen
Re: (Score:1)
Re: (Score:2)
The Mythical Man-Month is a classic. I've read it twice, and given talks on it.
How does it fit into the AI discussion?
Re: (Score:2)
f: (Managing teams of people) -> (Managing teams of AI agents) is an isomorphism.
Re: (Score:2)
If you think "managing a team of AI coding agents" is an isomorphism for Brooks' team structure, you should re-read the book.
His structure was very explicit, and had only exactly two programmers on a team: the "Chief programmer (surgeon)" and a "Copilot".
All the other roles were auxiliary
- Administrator (scheduling),
- Editor (documentation),
- Toolsmith (creating tools to enable the programming process),
- Tester (QA),
- "Language Lawyer" (kind of an architect).
You can suggest that a programmer with a "team" o
Re: (Score:2)
The underlying isomorphism point is solid imho. And I'm not actually suggesting that a "team" of AI coding agents is a thing that works, just that agentic AI software development has the same flaws as human software development.
The pattern of use by software engineers, who are by and large unqualified to lead teams of humans, is to use teams of agents any way they can, currently (observations in an industry with unlimited money resources).
There's no rhyme or reason to either their own management of the A
Re: (Score:2)
Thanks for clarifying, I missed the sarcasm.
Re: (Score:3)
If your code doesn't *have* to be right, sure. But AI makes incorrect assumptions very frequently, regardless of which model you are using.
A quick example. I asked Claude to add a feature in my genealogy software, to "repeat the last source provided." So you add a source for one individual, and you want to add the same source to the next, just click a "Repeat Last Source" button instead of having to look it up again.
Claude assumed I wanted to persist the "last source used" in the database. It went to the tr
Re: (Score:2)
That's a model specific problem, to a large degree.
The "AI assumed something stupid and destructive" problem went away when I started using better planning skills and models which were not anthropic's.
I've not had any of those "wild destructive hallucination reasoning" problems with GLM or Qwen models, even small versions, which I fought with regularly with opus.
Re: (Score:2)
That's a model specific problem, to a large degree.
Not in my experience. I've tried Claude Sonnet and Opus, Gemini 3.1, 3.5, 3.6, GPT Codex, GPT 5.5. I've not tried Grok, though I have doubts about even going there. They all pull the same stunts.
CEO brain (Score:4, Informative)
2 weeks ago my large company you've heard of had a townhall meeting where the CEO, who had hitherto been quite levelheaded and sensible about the AI stuff, finally threw all-in with the AI boosters and encouraged everyone to use as much AI as we could, for every possibly use we could think of.
3 days later we were out of tokens and even the useful stuff like support ticket summaries were unavailable.
Re:CEO brain (Score:5, Informative)
He was probably told by the board to get on board with AI.
Wait (Score:3)
Wait... you mean there were actually people trying to use as many tokens as possible?
It's one thing to try to exhaust a usage quota (which has never been difficult) because it's a subsidized commodity. It's another thing entirely to just burn tokens.
How can a CEO say stupid shit like this? Oh right. CEO brain.
Re:Wait (Score:5, Informative)
Yes. [slashdot.org]
Re: (Score:3)
Yes. In our tokenmaxxing org (not AMZN), many employees were fixated on how many dollars they burnt during the day and how their AI spending compared against their peers. Employees came up with strategies to burn more money to catch up and overtake their peers in dollar spending. It was not uncommon to hear people talking near the coffee machine saying, I just overtook xyz in AI dollar spending. I too burnt money making AI write some throwaway code.
Now after burning millions, the talk has changed. Now peopl
Re: (Score:3)
Some bonehead made "tokens used" a performance metric.
Re: (Score:3)
Some bonehead made "tokens used" a performance metric.
As with many metrics, it's easy to collect data, is reasonably accurate and most importantly, repeatable to identify trends; and of course totally useless except to see where you are burning money.
Re: (Score:2)
Yeah I do this. Sure I could whip up a dashboard in Excel in a few seconds from pages and pages of data, or I could just ask Copilot to generate dashboards over and over again until it guesses the one I like, all the while taking longer to do the job and costing a lot of money. Eventually the business is sure to stop telling us to "use AI for everything".
Although corporations are probably just dumb enough that rather than admit AI shouldn't be used for everything they'll probably pay some consulting a shitt
Re: (Score:2)
Re: (Score:2)
It's always the same game with "supreme simple" metrics.
When it's "lines of code" the developer will be wearing out the enter key to juke the stats.
When it's PR reviews, the developer will rubberstamp every PR they come across to juke the stats.
When it's token utilization, the developer will throw absolutely everything they can into an AI to juke the stats.
Don't use stupidly simple metrics in an attempt to see how efficient complex workstreams are. If you do, the players in the workstream are going to maxi
Re: (Score:2)
Dogbert's Playbook (Score:5, Funny)
"We're not a fast-growing company."
"And we never will be if we don't lose more money!"
Sovereign Systems (Score:2)
I've mentioned this several times. Standalone open source models have several advantages - fixed operating cost, high security due to lack of 3rd parties, control.of behavior, etc.
https://www.scry.llc/2026/04/0... [scry.llc]
https://www.scry.llc/2026/07/1... [scry.llc]
Not like lines of code (Score:2)
It's similar, Krikorian said, to how software companies once considered how many lines of code a programmer wrote to be a good metric of productivity.
Here's the difference. The fad of judging people by lines of code lasted for years. Judging them by tokens is already going out after only months. AI is accelerating everything, even corporate fads.
Local AI is the way to go (Score:1)
Getting the AI out of the hands of the "Mag 7" and into the hands of the companies that need it is a much better strategy than being tied to them. The open source models, wrapped with the right security and guards, is a way better solution. http://www.fortalezaai.com/ [fortalezaai.com]
Re: (Score:2)
While I do not disagree, who is going to do that? Enterprises are all in "The Cloud" and do not know how to run their own computers anymore.
Re: Local AI is the way to go (Score:2)
I'm not sure what the value proposition is of fortelezaai, seems like they just strung a bunch of containers together, probably made a nice admin UI, but starting at $3k/mo ?! You must be kidding.
data sovereignty is a big issue where I work and I just set up something similar with vllm, but honestly the hardest part was compute for inference which these guys totally gloss over.
So much horseshit (Score:2)
a.k.a. throwing money at rubbish ideas, and seeing what sticks.
The greatest business strategy of all time /s
Re: (Score:2)
When your customers are idiots with delusion, it seems to be working quite well. For a time that is.
Data centers? (Score:3)
Gee, it is almost like it is silly building giant data centers when companies can run their AI models in-house.
Re: (Score:2)
Gee, it is almost like it is silly building giant data centers when companies can run their AI models in-house.
They can run inferior models in-house. If they want models which do all the things they're used to, though that is nowhere near the promises made by AI providers, they will either need to spend a shitload on hardware or pay someone else for the service. There's no cheap way into having high quality LLMs, for what that's even worth.
The models you can run on reasonable hardware are more impressive than many people might think, but they aren't anywhere near the capabilities of the larger models.
Fable (Score:2)
I've been tokenmaxxing Fable 5 and I have to say I'm pretty happy .. except for the fact that I'm running out of tokens too fast. I did try with Opus 5 and it's ok, but it does get confused more than Fable 5. For what I'm doing Fable gets it right sooner. But the tokens are killing me.
What i do now is make it pause until my quota resets again. But this is clunky.
Strange article (Score:2)
If tokenmaxxing *fades* this is no argument for (but also not against) cheaper AI. To have a logical conclusion you would either think they'd say
- "Workplaces look for cheaper AI as employees are tokenmaxxing" (logical conclusion: more tokens mean cheaper AI to be able to afford them)
- "Tokenmaxxing fades as trend" (no implication for or against cheaper AI either way)
Re: (Score:2)
Reducing tokens would be a path to cheaper AI.
You seem to assume cheaper AI *must* only be about cost per tokens, but practically speaking they don't care about that detail so much as they care about the overall cost.
This has been playing out in my organization, where the consensus has been that folks burning through tokens like mad are, at their best, doing only as well as people using it in a more curated way. So the guidance has come down that people are only to be using it in a curated way, to reduce e
Re: (Score:2)
Reducing tokens like in "token efficiency" in reasoning (and fluff around the answer) is a metric for making the AI at the same cost cheaper to use, but the other option is to use a cheaper (possibly worse) model. And the third option is to decide where to use AI and how much to save tokens (best combined with token efficient models). Stopping "tokenmaxxing" is a reasonable thing anyway and the decision to additionally use cheaper models is kinda independent from it.
Re: (Score:2)
AI at the same cost cheaper to use
If it is cheaper, then it isn't the same cost. Yes, there are various ways to reduce cost and reducing the volume of work is only one of them. It doesn't work to accuse the article of being 'strange' for saying that using less of the services is cheaper. Because that is obviously one of the ways to be cheaper.
Re: (Score:2)
You are confusing cost as in money with cost as in computational effort. Token efficiency means solving the same problem with fewer tokens. If this means less money depends both on token efficiency and on per-token cost.
money = tokens * token_price, and token_efficiency = problem_size / solution_tokens.
So solution_tokens is a money independent cost metric.
Re: (Score:2)
I'm not confusing anything. I'm saying you are applying a rather picky definition of 'cost' to disagree with the assertion that 'using fewer tokens is a strategy being used to reduce AI cost'. The assertion makes sense so long as you don't get very weird about limiting the definition of 'cost'
Re: (Score:2)
Okay, so now for you focused on monetary cost: Your current AI is costly. What can you do?
1) Reduce the number of token needed by using a model that "talks" less for the same task with the same result. You need to pay for less tokens. This is known as token efficiency.
2) Reduce the size of the model so that the per-token cost is lower.
These options are not mutually exclusive.
One device models (Score:2)
just run your local AI (Score:2)
Re: (Score:2)
Here is news for you. Most "established companies" are not even capable of running their own computers. Instead they use something expensive, insecure and unreliable called "The Cloud" and think they are doing it right.
Re: (Score:2)
I have a friend who worked for a decent sized startup. Their product was basically a web interface to a container around a Python interpreter. I asked him why anybody would pay for that and he told me that lots of companies can't install Python.
Somebody is going to sell a bunch of GPUs in a box with a nice fat service contract of course. Oh wait, they already do:
https://www.nvidia.com/en-us/d... [nvidia.com]
Well done, marketing dept. (Score:2)
AI companies and their marketing department somehow managed to convince the world that the number of used tokens represents productivity.
What I would argue it actually represents unproductivity. After all, if you've got to use thousands of tokens to do the simplest jobs right, it means that productivity per token (and unit of cost) is low.
So yes - well done marketing departments. You've managed to turn the reality around and convince your customers to consume as much as possible and pay ridiculous amounts o
Re: (Score:2)
Division is hard. People have trouble focusing on more than one thing at a time, and divison has both a numerator and a denominator. Usually the numerator is sexy and the denominator is kind of boring so people forget about the latter. In this case the denominator is sexy so they just treated it like it was actually the numerator.
Such a surprise (Score:2)
LLMs not all they are cracked up to be. News at 11.
Seriously, have these people still noticed nothing? The great AI revolution is not happening in this 9th or 10th attempt either. Some smaller improvements on some things are to be had, but that is it. No getting rid of experts. No reduction in the competence that engineers need. No good code from the incompetent. No "AI does it better" for almost all things.
Re: (Score:2)
Mostly, I'm with you. I think of AI as a tool. Most of my experiences with AI have been pretty poor or some close but not good enough.
For very specific tasks what we used to call machine learning really works.
I used Claude Code recently for a little personal python project, the free web version. I don't use python much. I had cobbled together some working code by hand that was working but I wanted to polish it. Claude was able to rework my code really quickly. Its only a few hundred lines and easy to revie
Re: (Score:2)
I am using "AI" as shorthand for LLMs. I am well aware that other AI fields deliver. After their respective hypes had died down that is, for many of them.
I am also not arguing that when learning some tech while already having good skills in a related area, LLMs can be helpful. But to not expect them to ever be able to deliver production code.
Joke's on them (Score:2)
Re: (Score:2)
Yeah, bullshit. From naive Bayes filtering your email to image generators, and LLMs quantifying the massive amount of natural language data we collect, lots of AI is very much cheaper. Even the smart chatbots probably make some things cheaper. The good applications fade into the background of civilization and people only recognize the silly mistakes.
Re: (Score:2)
The biggest problem OpenAI and Anthropic have is they don't have AI, they have a crappy version of the Library Computer Access Retrieval System (LCARS) or a super Wiki. Now that could be useful, if it always knew that 1 + 1 = 2 but due to the fact it has terrible data inputs and "hallucinates", the accurate statement would be pulls incorrect data or calculated incorrectly, constantly they've done somethi
Re: (Score:2)
There was a time when actual computer scientitsts frequented this site. Naive Bayes is a classic AI algorithm, and natural language context filtering is a classic AI problem.
No comment on AI (Score:2)
Silicon Valley executives were promoting high token consumption as a signal of high-performing employees.
All opinions on AI / LLMs / Agents etc aside resource consumption is a pretty poor proxy for productivity/efficiency/activity.
Sure a macro level it works, oil usage by nation states does track with economic activity. At the house hold level not so much. If the Simths buy 20% more gasoline this month it might just mean Mrs. Simth picking up and dropping off Mr. Smith to and from work because his little efficent coupe is in the shop and she needs the Scoccer Mom Edition Expedition to cart the kids around all
The accountants are tired of paying for it all (Score:2)
This was expected. Apologies to Joe Walsh.
Sounds like it might be time... (Score:2)
...to re-read The Midas Plague by Frederick Pohl.
https://archive.org/details/ga... [archive.org]
Lack of accountability (Score:2)
My biggest issue with all these fads is upper-management "running with it" and when the fad dies, they pretend as though they never championed it, fall totally silent, or if called out simply state "we're re-evaluating our direction", of which nothing ever comes out. They make grand statements, our jobs are on the line to make them look good, and when we don't... well I guess we're screwed and they get to continue a fruitful career in "leadership".
It's the beginning ... (Score:2)
Go look back at the dot-com collapse and all the people who bought a catchy domain name without a viable business plan. Those who don't remember the past ... yadda, yadda.
Bubbles (Score:2)
Every sane person has been laughing at it from the first second.