Man who uses AI to write his own claimed output states that AI has no use.
At its core, the message is that AI euphoria is so mad that it is now not just infecting US stock exchanges, whilst massively distorting the allocation of capital resources within the US economy towards a purpose which appears to have no useful human goal
And, well, if AI’s only use was to increase Spud’s output then, well. But isn’t his lack of self-knowledge here wondrous?
So he knows what the optimal capital allocation should be in the US economy? I’ve never come across anyone more deluded.
To be a little fair to him, the East Coast tech bro’s approach to capital allocation is to throw money at everything whether or not is has any chance of success .
Which is not unreasonable considering how much money they have and can afford to spunk.
This is the game, and it’s been the game since at least the dot-com era:-
1) find a technology that is cool
2) spend some money on a company doing it
3) give the product away, while losing money on it, which generates users, buzz and hype.
4) watch as the media write about it, adding to the hype
5) go public, selling it to the idiots that can’t tell the difference between user volumes, revenue and profit.
Some of these companies, like Meta and Amazon, eventually figure out a business model that works, years later. But the venture capitalists behind them aren’t trying to make profitable companies. They just want to dump them on mugs.
There’s quite a lot of real-world, profitable uses of AI, but mostly in commercial contexts. And that’s being done by companies already doing particular software, that have high quality training data and for clients that don’t mind spending good money. AI is a technology, not a company, especially since DeepSeek and Qwen came along, which are open weight – you can train them on your specific data. And that’s why I believe that OpenAI and Anthropic are basically fucked. They’re trained on things like Wikipedia, Reddit, any old books they can get hold of. So, generic crap. And the people using them don’t want to pay.
My wife works in corporate finance and is currently taking some professional development course – quite complicated stuff. Part of it includes answering some complex questions based on large data sets. She’s very bright and is averaging 85% (the pass mark is 55%) but for a laugh I showed Claude the questions on one of her papers. Took it about three minutes and it got 100%. Claude costs me £18 a month.
AI works and adds value. Sorting out how we deal with the fact that most people have no useful skills is going to be a challenge for society.
Are the answers already on the Internet somewhere? There was a brag from one of the AI vendors that their LLM had aced the Maths Olympiad, but they shut up when it was pointed out the answers were by then online.
Came across a factoid the other week which disturbed me. The cash rich “AI companies” are expensively buying the whole stocks of second hand bookshops, scanning those they don’t already have, then shredding the lot. This is encouraging the bookshops to restock from charity shops, etc, for further reward.
This removal of the past printed word is not a good thing in the digital age.
I suspect the shredding is intentional.
It’s not. The book scanning machines need the covers removed. So, you “break” the book and at the end of it, have a load of pages.
Personally, I can’t get that worried. This is going to be Edwardian trash fiction about cads and bounders, forgotten cold war thrillers, old ZX Spectrum manuals. Not the last copy of The Odyssey.
Digital has done more to give people access to old books because of the lower production and storage costs. A publisher doesn’t have to put up the cost to get 500 copies of an old Miles Burton novel produced and store them somewhere. They can just get the text onto the Kindle store at whatever that costs in their time and if it sells 20, make a few quid on it.
It’s the same with music. There’s a lot of music that stopped being sold on vinyl, never made it to CD that is on digital. Because you don’t have the upfront investment. Records used to get “deleted” which was after they’d sold the stock, if it was barely selling, they didn’t bother making any more.
Thanks WB.
You’ve made this bookworm happy!!!
That’s just the core model though. You add in a RAG pipeline to get more specific output.
So he knows what the optimal capital allocation should be in the US economy?
No problem, he’ll have that sorted by breakfast and the rest of the world done by coffee break.
“towards a purpose which appears to have no useful human goal”
This is how he uses AI
Gamecock believes there is irrational exuberance in the growth of AI. The industry is gorging on capital. It is not making money. “Some say” the bubble will burst in 2027.
I don’t know what is going to happen, but there has to be a change. What ever wondrous claims are made about its potential, it still has to make money.
Even though today is cold tea and cheese sandwiches, tomorrow is champagne and caviar dontchakno
AI saved me a day’s work yesterday by pulling data from a crappy PDF file that was all I was given to work with.
And that was just yesterday. If Spud ever did any actual productive work (especially with incomplete records) he would know how truly useful the LLMs are in real-world situations.
Out of interest. Have you tested its accuracy or just accepted it.?
As per the above my wife works in corporate finance and is currently taking some professional development course – quite complicated stuff. Part of it includes answering some complex questions based on large data sets. She’s very bright and is averaging 85% (the pass mark is 55%) but for a laugh I showed Claude the questions on one of her papers. Took it about three minutes and it got 100%. Mind you all it is doing is looking at the rules and regulations and applying them to the facts.
The problem is that it really can foul up spectacularly and you need to know when.
Humans might misjudge distance on a road, and bump into the back of someone. AI will mistake a shadow for the edge of the road and drive off it.
https://www.youtube.com/watch?v=frGoalySCns
Because it’s just running a load of stats based on prior information and not reasoning. As a human your brain understands it’s a shadow, not just because it looks like one, but there’s the tree and there’s the sun.
You’re absolutely correct. The advice I was given was to treat these LLMs like the office junior, to never trust their outputs and never give them source documents.
It’s a tool like any other and a bloody good one if you use it properly.
What I don’t think it’s ever going to do is what a bright person can do,Reach a correct decision based on information that is both incomplete & possibly incorrect.
That’s right. It’s just doing probability based on prior data. It doesn’t understand anything. It has no understanding of sunlight, trees and shadows like a person does.
But cheap information sorting using probability to find potentially good answers to be investigated can still be useful.
The jobs apocalypse is nonsense, because it can’t replace real intelligence.
But how many jobs today require real intelligence?
Never trust it! Check totals and cross references every time.
But that’s always been the case – people make mistakes so check totals etc have always been there.
In my experience, AI is way faster and actually less error-prone.
But assuming perfection is just dumb.
It writes a lot of my code now. Sometimes it gets things wrong, but I review it and fix things manually or tell it to change it. A lot of code is boring to type, to look up syntax I have forgotten. It does that for me. But I still do the bits that are tricky, like figuring out how to get the price from a wine page, when there might be lots of different £ signs. Original price, sale price, price for a case, price in bond.
How much it’s going to cost once everyone stops subsiding it, I don’t know.
This I why I don’t understand why you don’t think it will take your job. If it can write code it can write code without you.
Did your AI provider make any money?
We pay a licence…
Not an answer to the question;)
Being paid =/= making money.
I neither know nor care whether they make money or not. They charge a price – we like it – we pay it – after that it’s their problem.
I love his reference to money being invested by the private sector is allocation, or rather misallocation of resource. I.e. money is resource and a lack of money means resources not being allocated. For the public sector though, oh that’s different
A government doesn’t need money; it makes it by allocating resources. And the world can’t turn without it.
Its so good that he has such a clear view of technology and the future of humanity.
With this sort of economic insight its clear why he is so wealthy and has such a prominent position in industry and government.
One of my frequent projects requires submission to HMRC of lists of investors. They require this in a certain format. Alphabetical order, surname first being one of a number of requirements. Clients never have the information in the format I need and usually lots of detail I don’t need. Used to involve a couple of dull boring hours reformatting. After establishing a set of rules with ChatGPT I now upload whatever crap I get sent by the client and a couple of minutes later it pops out a perfectly formatted excel workbook for me.
I’m using it to help with complex planning too. Lots of back and forth and constant cross checking of references but it’s turning two or three day planning sessions into a day’s work. It’s really good at HMRC clearance letters and drawing up checklists for client actions on projects.
You need to understand the subject, know what to ask it and be able to spot when it goes off track. So I wouldn’t trust it to do the planning, but it’s a f’ing competent assistant.
Used to involve a couple of dull boring hours reformatting.
Dull, but billable hours 🙂
I suspect the dull but billable hours are being billed. If it was a gradate dissertation it would be plagiary. In many other areas straight fraud.
And therein lies the problem. For someone with a decade or three of experience, spotting the error is easy. But if the LLMs are taking away all the junior positions, nobody is building the years of repeatedly being hit over the head with experience that’s needed to effectively manage them.
I fear that the LLMs will be put in charge of overseeing each other, which will lead to an almighty mess caused by an hallucination. But given we’ve already got that several of those with the PPE grads running everything, so maybe it wouldn’t make that much of a difference.
Clogging the entry to the pipeline of experience is a chronically recurring problem across society.
I was thinking this regarding a friend’s daughter now in her mid twenties. After uni she went into property management & apparently is now a high flyer. Now I haven’t a clue about what she actually does but I presume property management must include repairs, maintenance & renovation. That means dealing with contractors. And having been one, I’m fully aware of the multitudinous strokes contractors can pull. It’s a learning experience keeping up with them.
How does someone straight out of uni do the job? The only way you’re going to learn to deal with contractors is experience. Why property managers were usually of maturer years. Many had come over contracting side . Poachers turned gamekeepers. And it’s the same mindset that enables dealing with dodgier sort of tenants. The money, legal side etc isn’t really hard to learn.
Dunno. Maybe she’s in the marketing side. In which case the tits & legs will be her main assets.
How the hell does an LLM deal with out of the ordinary names?. Spanish & Ports with two surnames, the first taking precedent?. Some English have two non-hyphenated?. Orientals where the first name is the surname?
I can remember when my skill in negotiating shareholder registers was an asset.. The same people turning up on different company registers with their names in different forms.