LLMs remind me of blockchains
I used to be quite fond of the idea of cryptocurrencies – in some ways, I still am. I remember when I could send my friend some beer money (mETH was a practical unit back in those days, and it was called funny) which arrived in seconds and cost virtually nothing to transfer. I remember when I could send a friend a bunch of money, me being in Europe, him being in the US – it was a Sunday, and yet the money arrived within minutes and it cost virtually nothing. It seems like an appealing alternative to traditional banking, slow and expensive, stuck in its old ways.
Of course that didn’t last. Blockchains were somewhat cool, and an effective proof of concept that did work for a while, but started showing their weaknesses the moment they became popular. Things got slow, things got expensive, and it became apparent that this thing will not deliver on its promises. It was – and still is – good enough to do crime with, but absolutely useless for an ordinary person, and pretty much a straight downgrade from its alternatives.
I still things cryptocurrencies are a neat idea, and I wish they’d take off. It would be nice to have something decentralized, trustless and low-friction. Perhaps if it existed, internet would be less ad-centric and a better place. Maybe GNU Taler could be that thing? I’m looking forward to it, or something akin to it.
Blockchains, however, aren’t it. A decade in, they’re still shit for this usecase. And given their track record so far, it feels to me that even if we did somehow solve all of its problems – make them reliable, efficient and sustainable – they still wouldn’t make the world a better place. The drawbacks are too many, and while the underlying idea may be appealing, the implementation is doomed to be a net negative for humankind.
The primary goal may be worthy and sound, but the implementation didn’t deliver, and it’s time to give up on it.
Reminds me of the Next Big Thing, the next gold rush.
AI does seem like a cool thing to have. A very smart personal assistant. A piece of software that excels at dealing with the unexpected. Summarizer? Analyzer? Reasoning engine? All at once? Sign me up. I don’t think we’re anywhere near it though, and I doubt that we’re even moving towards it.
LLMs are a very cool concept. The idea that if we double down and down again on a Markov chain so hard that it develops sentience still sounds absurd, but it’s quite amazing how close we’ve gotten to it working out. We’ve had conversations where we could genuinely argue that maybe all sentience is just a probability engine based on the sum of all of our experiences? Maybe we’re all just predicticing the next token/behaviour, over and over? Maybe that’s what a mind is?
And so we’ve put a few years, an obscene number of graphics cards and an eye-watering amount of Watts into this idea. Powered through with the promise that with just one more recomissioned power plant and one hundred more GPUs finally the sentience will pop into existence. Any second now. Just one more datacenter.
And almost 3 years later… it’s still garbage. Every written text has a distinct LLM smell to it, and the imagery is even worse. The information is unreliable, the logic is flawed, the sentience is not, the king is naked. The once astonishing tech demo engine is not improving much any more.
The “AI” we’ve gotten is great for posting spam, generating outrage on social media, writing bad code and dodging homework.
It’s terrible at its job, and only good enough to replace a human if you’re expecting next to nothing from the quality of work.
It occasionally makes a business sense to have it replace an employee;
provided that you have little to no expectations of the employees’ output, but you sure want it cheap.
And cheap it is, but we’re not paying the real cost of it.
It’s still being funded by gullible suckersvisionaries looking for the next big thing.
It’s good enough for how cheap it is, but will it still be cheap enough when it’s forced to not lose gargantuan amounts of money? Once the prices rise (or the quality degrades, like with the smaller models popping up these days), will any usecase still be worth it?
It’s likely I’m wrong on this, as I’ve been on many other things in the past. Maybe I’m just a negative nancy, an ignorant, the AI is actually great, better than ever, and replacing me as we speak. Maybe GPT 4.20 would write a better blog post than whatever this is, and cheaper!
And that takes me back to blockchain. Maybe all these problems can – and will – be solved. But will that make the world better?
Will it all be worth it if ChatGeppetto does actually outpeform a search engine, an artist and a secretary?
Will we all benefit once a machine will write this blog post for me, and your RSS reader will read it and summarize it for you?
And of course, maybe those are just stepping stones. Those are just the PoCs, the tech demos, and the AGI is coming soon. But to what end? What’s it going to do, aside of generating more convincing blog spam than ever before, manipulating people better than ever and replacing us all at our jobs (which isn’t very attractive in a non-post-scarcity economy)?
Blockchains continue to facilitate fraud and abuse, no matter how much we improve upon them. Why would better “AI” stop degrading the society that built it? It’s proved valuable now, it will surely continue to be valuable once the technology is better and cheaper.
If the optimistic scenario doesn’t sound like a good thing, maybe it’s time to stop.
I can rant all I want, of course, and this shit is here to stay.
So far it’s giving me work opportunities at deploying AI-scraping mitigations, so that they’re not DDOSing valuable websites. Without a doubt, as time goes on, I’ll be spending increasingly more time reading LLM-generated articles, reviewing LLM-generated code and fact-checking GenAI produced fake images and videos.
Progress doesn’t care that I consider it to be a regress.
But while I still can rant all I want, I will. And if you got this far, thank you for reading. If this resonates with you, do let me know. We may at least find comfort in being stuck with this together.