The AI emperor has no clothes
jeffgeerling.com
jeffgeerling.com
But the revenue going into AI will not go down. The job loss will not go down. What you're seeing isn't AI being an empty promise, but rather many different people trying to win the market.
It's like "tech"... 20 years ago, there was a category of companies we called tech and it was getting outsized valuations. Now every company is a tech company, whatever that means – there's as many programmers at Goldman Sachs or Walmart as there are at Airbnb or Dropbox.
IT has been the biggest winner of the AI wave, as it's allowing us to suddenly start delivering on asks from the business that previously would have been deemed impossible/impractical with our resources.
This is just a case of the user being unable to see far enough into the future. Yes, there's huge future profit to be had.
AIs don't understand reality. This type of data generation would need a specific sort of validator function to work: we call this reality. That's what "experimentation" requires: reality.
We already have this right now, with the AI training ingesting AI crapgen, with StackOverflow posts no longer happening. That would seem to point to a degrading AI training set, not an improving one.
Here's a discussion with Isomorphic Labs (Google DeepMind spinoff) on this line of thinking: https://www.youtube.com/watch?v=XpIMuCeEtSk
I have no idea what fraction of our economic productivity is wasted doing these sort of TPS reports but it's surely so massive that any software that lets us essentially develop more software on the fly to cut that back even slightly is highly valuable.
Previously only the most moneyed interests and valuable endeavors could justify such software, like for example banks flagging sus transactions. Current AI is precariously close to being able to provide this sort of "dumb first pass set of eyes" look at bulk data cheaply to lesser use cases for which "normal" software is not economically viable.
Boring ass code reviews come in super handy because of the better familiarity, getting exposure to the code slowly, exposures to the 'whys' as they are implemented not trying to figure out later. The same with buyers overlooking boring paperwork, team leads, productions planners. Automating all that is going to create worse outcomes.
In a sane world if we could take the fluff away we would have those people only working 5% of the time for the same pay, but we live in a capitalist system where that can't be allowed, we need 100% utilization.
>Boring ass code reviews come in super handy because of the better familiarity, getting exposure to the code slowly, exposures to the 'whys' as they are implemented not trying to figure out later.
But to what extent is this truly necessary vs a post-hoc justification? Workers are pushed to work right to the limit of "how little can you know about the thing without causing bad results" all the time anyway.
>In a sane world if we could take the fluff away we would have those people only working 5% of the time for the same pay, but we live in a capitalist system where that can't be allowed, we need 100% utilization.
<laughs in Soviet bureaucracy>.
The catholic church was making fake work for itself for about 500yr before it caused big problems for them. It's not the capitalism that's the problem. It's the concentration of power/influence/wealth/resources that seems top breed these systems.
so ive been in the industry ~8 years now, and this is the exact same discussion that was held during the NFT/blockchain hype. this cycle seems inevitable for any new disruptive tech.
AI may be overhyped but the actual use cases and potential ones seem very clear.
It's still pretty unclear to me why you'd actually want a blockchain for managing it though, rather than a traditional database hosted by the central bank that is responsible for the currency that the stablecoin is following. You'd get vastly more throughput at much lower costs, and it's not like you really need decentralization for such a system anyway. The stablecoin is backed by something the central bank already has authority over.
But to be honest where I am (northwest Europe) we already have subsecond person-to-person transactions via the normal banking system, no matter which bank the sender and receiver use. So stablecoin-ifying the Euro wouldn't make a huge difference. There might be more to gain if the region doesn't have that kind of payment infrastructure yet.
Maybe compare with something the development of motion pictures, with dozens of small studios trying different approaches to storytelling, different formats, etc, with live performances being the legacy thing getting disrupted. And then the market rapidly maturing and being consolidated down to a few winners that lasted for decades.
Sadly for those in unstable countries there are lots more risks in life, including financial. I'm not sure crypto is the best long term solution here
The use cases where it gets applied are far more than the valid ones
As for NFTs ... there never was (and never will be) a valid use case - it does not matter if you "own" a digital asset (like an image): a screenshot of it is good enough for 99.99999...% of people, so why pay for the "real" thing?
Yeah, and im seeing the same pattern with AI i think
And afaik, that was the only use of NFTs that actually worked. The whole “omg you can buy skin in this one game as an nft, and then have it across multiple games” (as well as other bs of a similar nature) was always a ridiculous thing, as it still requires gamedevs to collab and actually do all the required integration work (which is rather infeasible). And even if they decided to ever do that, they would almost certainly find it much easier to accomplish without relying on NFT tech.
AI also has a ton of hype and pipedreams that are vaporware, of course (just look at all the batches of chatgpt wrapper startups in thr past couple of years). However, what the AI hype has (that NFT never did) is a number of actual functional use-cases, and people utilize it heavily for actual work/productivity/daily help with great success.
Claud Code/Gemini CLI alone can be such a boon in the hands of a good software engineer, it is undeniable.
I feel the idea that there's only "one thing it's supposed to be good at (writing code)" is largely down to availability bias, in that we're in programming circles where LLM code completion gets talked about a lot. ChatGPT reportedly has 700 million weekly users - I'd assume many using it for tasks that I'm not even aware of (automating some tedious/repetitive part of their job).
It sucks at replacing software engineers. So yes, if you expect it to design a complex scalable system and implement it for you from start to finish, then sure, you can call it bad at "writing code".
I don't care about that use-case. I think I am decently good at writing code, and I rather enjoy doing it. I find Claude Code/Gemini CLI extremely helpful at both saving me lots of time by freeing me from dealing with annoying boilerplate, so I can focus more on actual system design (which LLMs fail at terribly, if we are talking about real production apps that need to scale) and more difficult/fun parts of code (which LLMs cannot handle either).
That's the real power of it, multiplying the productive output of good SWEs + making the work feel more enjoyable for them, by letting those SWEs focus on actual tricky/difficult parts, instead of forcing them to spend a good half of the time just dealing with boilerplate.
And I am not even gonna bother getting into tons of other non-coding tasks it is already very useful of. I find it amazing that I can now not only get a transcript of my work meetings (which is already massively helpful for me to review later, as opposed to listening to a ~25min video recording of it), but also ask an LLM to summarize it for me or parse/extract info from it.
Watch the engagement plunge, and with that, the dreams of profit.
And MS et al will grease the palms the the politicians running against them. Lobbying is easy when you're a huge corporation.
A tiny number of them are actually wearing something, like a fig leaf. They are the ones won't be caught in embarrassment when an innocent child points most emperors are walking around in the buff.
---
[a] Cloud providers, who do earn profits, are service providers to emperors, not emperors on their own.
I'm not suggesting GPT has no value. But in hindsight everything I described above was pure bull. I suppose it could be suggested that Sam Altman and his engineers didn't understand their own algorithms. But I don't buy that fairytale.
Marketing, pure and simple
People relate better to something that sounds humanesque (even though it is not) vs calling it what it is (in this case, a massively-backed (ie LLM-based) Markov Chain generator)
Same reason they hyped up how worried they were about "safety" of the "we have to make sure these things are 'aligned' or they'll become Skynet!" variety: Altman was bullshitting to hype up the company. Even the "safety" stuff was just hype. "It's so capable it's literally scary, or soon will be... better invest in / use our product! Imagine how bad it'll be for you if you don't!"
I was on the fence about how serious they were until I finally got around to skimming the "Attention Is All You Need" paper. LOL. LMFAO. No.
A long way away
And here is why - driverless cars are a thing ... essentially making the car the "robot"
General-purpose robots are an amusing scifi trope, but have no practical benefit in reality
Purpose-built robots (even ones that can flex within that prupose to different applications) make far more sense (and have already been around for decades)
And it’s true. Industries have shown time and again that they’d rather send the work to paupers’ hands in countries without rather than automate to metals hands within.
I disagree, to the point I'd say the statement is nearly false.
You have to look at what the 'expensive' point of the work is.
In some cases it's electricity, those things generally don't get sent overseas, but instead are highly automated.
In some cases it's waste by product. Those get sent to countries where they don't care if you dump it and poison it.
In some cases it's proximity to other manufacturing of parts you need.
In some cases it's cheap labor without the need for paying healthcare.
Surely it was for the cheap labor regardless of how it was achieved.
Does that mean the AI emperor has no clothes? I don't think so.
I've already seen enough significant, real-world value being created by AI technology that I am a believer that it will have a big impact. It will also destroy a lot of investment money along the way, but that's just normal bubble stuff.
The actual emperors? They are dressed (or, at least, in the fabric section of the hobby store and have a sewing machine at home)
Eventually, the music will stop and people will start looking at the cash flow.
It may take a bubble pop, but eventually business will figure out some good use cases that make sense. This happened with the DotCom.