Any software
This is a great article and the points made apply to any significant deployment of a proprietary product (ERP, HCM, et al.).
"It is context, workflows, and institutional memory" oh yes!
OPINION The days when you could jump from one frontier AI model to another at the drop of a hat are going away as vendor lock-in starts to kick in, and prices increase. Once upon a time, say last month, people thought nothing of jumping from one AI frontier model to another. One week, the hottest AI model was Gemini 3.1 Pro, …
Employees in companies that aggressively deployed AI over the last 2 years already have already ceded many of their former capabilities over to AI. Lost knowledge and forgotten experience will be expensive to re-learn and re-train.
The drug dealers giving away drugs for (nearly) free now have at least some victims firmly on the hook.
It would be pretty ironic, if exactly those companies, that were told for years now to be "too late" to adopt AI, now emerge as the ones that avoid becoming a victim of predatory AI pricing strategies and will enjoy major cost and quality advantages in the future...
American corporations have this very odd business philosophy of discouraging repeat customers, and putting its current customers out of business.
I don't get it, but then I'm not some nepo-fail-baby-trustafarian with the right connections. I'm just another plebe/NPC, so what do I know?
"and putting its current customers out of business."
Mmmmmm, if you have a business that's heavily reliant on a sole-source outside supplier that's way bigger than you are, do you really have a business?
The county where I live was using free API's and satellite imagery from Google until Goober decided to charge the moon with very little notice. Now the county has to completely rethink how they can supply GIS services as they can't afford to pay. It was great while it lasted. I could put in a parcel number and get a satellite view with the lot lines overlaid and be able to go out and get images of the land with the drone for clients. It's much more office work now to do the same thing with a bit less accuracy. Fortunately, I don't need cm precision. For me, having the data easy to get was a convenience and let me charge customers less. Now I've raised my rates to provide a similar service. I've seen other companies use a free or low cost service and find themselves with no viable business when things change.
Better to be as vertically integrated as possible and anything that needs to be outsourced available from multiple vendors.
The same mistake as
- Systems Integrators
- Web Hosters
- Outsourcing Partners
- Offshoring Providers
- Cloud Providers
- SaaS Partners
You are losing control of your domain knowledge, workflows etc.
They never learn. Perhaps starting AI in the Boardroom will make for better decision making, better risk taking … and best off they don’t need inflated salaries, pension contributions and fucking stock option grants and bonuses. ….
"most senior executives are completely clueless about what AI is and how to deploy it."
Indeed. A modest problem with both Capitalism and Socialism that Marx failed to mention. When your leaders are lost and confused one has a bit of a problem. Especially when, as often seems to be the case, they are not aware that they are lost and confused.
"Switching model vendors is no longer just an API migration. It is context, workflows, and institutional memory."
And he wrote his reply using a chatbot. "Blah is not just blah. It's blah." is as tell-tale a sign of a chatbot reply as an em-dash. Has he no shame?
Maybe, but that's also a relatively common thing for humans to say, which is one reason LLMs picked it up, and most of the times I've seen it from an LLM, it was used as an intensifier rather than a contradiction like this is. Perhaps we should stop ourselves from taking something as simple as "not just" and jumping to conclusions like bad models do.
Relatively common in certain countries perhaps.
I'm Kenyan. I Don't Write Like ChatGPT. ChatGPT Writes Like Me.
There’s a growing community (cult?) of self-proclaimed AI detectives, who have designed and detailed what they consider tells, and armed their followers with a checklist of robotic tells. Does a piece of text use words like ‘furthermore’, ‘moreover’, ‘consequently’, ‘otherwise’ or ‘thusly’? Does it build its arguments using perfectly parallel structures, such as the classic “It is not only X, but also Y”? Does it arrange its key points into neat, logical triplets for maximum rhetorical impact?
To these detectives of digital inauthenticity, I say: Friend, welcome to a typical Tuesday in a Kenyan classroom, boardroom, or intra-office Teams chat. The very things you identify as the fingerprints of the machine are, in fact, the fossil records of our education.
True if you believe Ed Zitron https://www.wheresyoured.at/ Who should I believe, Zitron or Altman, etc. al? On the whole, to me at least, Zitron seems more credible.
Thanks for that link !
Generative AI subscriptions are like if Uber charged users $20 a month for 100 rides of any distance under 100 miles, and if gas was $150 a gallon, and Uber paid for the gas because somebody insisted that oil would one day be too cheap to meter.
"The current price hikes still do not cover the costs of both training and inference. "
Many new technologies have an expensive early life as people work out how to do it better/faster/cheaper over time. I never estimate the price of a product based on the first prototype or even production unit I make. The question is whether there is any path towards that better/faster/cheaper that doesn't keep requiring mo machines, mo power and mo cooling. I haven't seen any papers/reporting on that. If it's brute force and MW's all the way, something will snap. All the investors that have been looking to make a load of money fast will get out and put their money on the next thing they think will make them money fast with very little effort. Many bit barns will be left owing the power companies that finally had to turn off the flow and also take back the leads/transformers to use elsewhere as they'll be hurting too.
I need to go pick up more solar panels this week.
"This is yet more proof that your C-level executives don't have a clue about what they're doing..."
To add a tiny bit of fairness, the can run a business that already has some record in an industry that's been around for some time. There just isn't any historical information on AI, robots, etc to guide decisions so it's down to the promises of the people looking to get funding for their haphazardly researched new enterprise.
True - but what "token" represents in value is impossible to pin down. One reply can be succinct and to the point, the next can go off on a wild turkey chase. On the backend they are twiddling the knobs constantly to try to minimize losses, while staying on par with competitors performance.
The kludge that is “Prompt Engineering” (or, as some have later rebranded it, “Context Engineering”), means that there will always be extensive work to be done when you change model.
Assuming you have even assembled a test suite (aka “evals”), in order to then keep tweaking the prompt until it sort of works again (for some value of “works”).
It’s as if all of these things are built on a pile of sand.
So, all this AI shit is finally going to start costing real money. And the companies pushing it will find out the hard way that their "wonderful" product is going to be dropped like a hot potato. And all those ginormous data centers are going to run at a loss and maybe we'll see less of them being built.
So, cheers all around, eh ?
Except for the CEOs who thought they had invented Web 4.0.
Not a great loss.
Unfortunately, even if the CEO gets the AX, it is a golden AX worth millions/billions. Certainly large enough to live comfortably for the rest of their life, or start another co and do it all over again. The tens/hundreds of thousands of minions have gotten the don't let the door hit you on the butt on the way out to pay for the AI/golden parachute.
"Unfortunately, even if the CEO gets the AX, it is a golden AX worth millions/billions."
When my mother is looking at stocks, she does all the investing for the family trust, one of the big things she reviews is executive compensation, if that can be dug out. She won't buy anything if the C-suite is overly compensated or will be able to retire due to a golden parachute. They have to work for a living and have some skin in the game.
Three things:
First:The largest corporate users also own stock in those AI companies. They will keep goosing that stock it until the pump and dump pays off or they, both the users and vendors, go bankrupt
Second: Because of the above, don't expect an exodus any time soon if at all: Amazon and Uber are two good example or companies that were losing BILLIONS EVERY YEAR, yet somehow survived. When they by all rights, SHOULD NOT HAVE.
Third: the farther this goes, the worse it will become because the world has never seen GIGO AND data theft on this scale before. And even after the implosion, short of an ELE, we're stuck with some form of this madness. Look at the current state of the Internet. The dot com bomb didn't filter the good from the bad, we just got more slop from those with enough money to survive. And the few remaining good things have been co-opted and turned into slop.
Amazon and Uber had massive early costs, but the business model was clearly sound and they deliver services which actually address a real-world demand.
How can you claim the same about AI companies? They are basically a gimmick, selling their inference at a loss, and they will at some point need to raise charges to cover the costs and make a profit. This in turn would likely make the charges so high that they will lose a large proportion of their business, which in turn may well make them unviable without further cost increases, etc, until they go bust.
What am I missing here?
"Copyright for thee but not for me. Japan and Singapore tell artists to eat shite."
Creative works have a copyright, but factual data does not. In some cases there is a compilation copyright. One of those old fashioned magazines could use a stack of licensed materials and claim a copyright on how they've put it together. A printed phone book can't claim a copyright on the pages of listings. That was until they made stuff up and convinced a judge that it was enough creativity that somebody copying it could be sued.
I've never had to have an attorney appear in court for a copyright matter. The fines for infringement are so high and the case law so clear that attorneys on the other side tell their clients to work a deal and pay up right away. The arguments the LLM companies are using are fanciful and they are trying to argue "fair use" which doesn't apply and that there's no way to say that something they've used in training is expressed in a similar enough way to be an infringement. If I study a bunch of photography books, I can be influenced by those images and that's the argument that's being put forward for the machines. I don't have to buy those books. I can go to the library or book store and just look at them. It's all rather new and I hope that good case law comes out of it that forces the companies to license the materials they are using for training. To get them by downloading torrents is a huge infringement.
This post has been deleted by its author
As long as governments don't bail out the idiots like they did in 2007, the "AI singularity" will be simply "a black hole into which AI fell, never to be seen again".
Sadly, most governments are still patting themselves on the back over how smart they were back then, albeit with other folks' money.
Went on an Azure/AWS training course many years ago where the instructor was hyping cloud computing as 'generic as electricity'. It was about as generic as a wall socket but with a huge amount of transformers, rectifiers and power conditioners lurking behind. Sure you can change cloud/electricity providers but you'd then have to configure a whole new set of transformers, rectifiers and power conditioners.
This post has been deleted by its author
This post has been deleted by its author
AI Migration Expert, as the skills are unknown as are the requirements , and you will pay me a lot as your system burns , as you don't know how to do it, I don't know, so its down to how much insurance you have , which for the first failure will be amazingly cheap, but that's all you need, 1 big job, the exit. (Checks IR35)