Famously Steve Jobs said that the (personal) computer is "like a bicycle for the mind". It's a great metaphor because- besides the idea of lightness and freedom it communicates- it also described the computer as multiplier of the human strength- the bicycle allows one to travel faster and with much less effort, it's true, but ultimately the source of its power is still entirely in the muscles of the cyclist- you don't get out of it anything that you didn't put yourself.
Bu the feeling I'm having with LLMs is that we've entered the age of fossil-fuel engines: something that moves on its own power and produces somewhat more than the user needs to put into it. Ok, in the current version it might not go very far and needs to be pushed now and then, but the total energy output is greater than what users need to put in. We could call it a horse, except that this is artificial: it's a tractor. And in the last months I've been feeling like someone who spent years pushing a plough in the fields, and has suddenly received a tractor. A primitive model, still imperfect, but already working.
A tractor does exactly what you tell it to do though - you turn it on, steer it in a direction, and it goes. I like the horse metaphor for AI better: still useful, but sometimes unpredictable, and needs constant supervision.
[delayed]
Having recently watched Train Dreams it feels like the transition of logging by hand to logging with industrial machinery.
I prefer Doctorow's observation that they make us into reverse-centaurs [0]. We're not leading the LLM around like some faithful companion that doesn't always do what we want it to. We're the last-mile delivery driver of an algorithm running in a data-center that can't take responsibility for and ship the code to production on its own. We're the horse.
[0] https://locusmag.com/feature/commentary-cory-doctorow-revers...
Clever Hans is how I describe LLM agents to non-techies
"Computers aren't the thing. They're the thing that gets you to the thing."
My favorite quote from the excellent show halt and catch fire. Maybe applicable to AI too?
Something like that used to be Apple’s driving force under Steve Jobs (definitely no longer under Tim Cook).
https://youtube.com/watch?v=oeqPrUmVz-o&t=1m54s
> You’ve go to start with the customer experience and work backwards to the technology. You can’t start with the technology and try to figure out where you’re going to try to sell it.
> You can’t start with the technology and try to figure out where you’re going to try to sell it.
If those LLM addicts could read, they'd be very upset!
ChatGPT, tell me how I should feel about this!
That works when you are starting a new company from scratch to solve a problem. When you're established and your boffins discover a new thing, of course you find places to use it. It's the expression problem with business: when you add a new customer experience you intersect it with all existing technology, and when you add a new technology you intersect it with all existing customer experience.
Apple was a well established company when they came out with the iPhone - I don't think anyone but Jobs would've been able to pull off something like that.
That sort of comprehensive innovation (hardware, software, UX - Apple invented everything), while entering an unfamilar and established market, I'd argue would've been impossible to do in a startup.
Isn't that why the big tech companies switched to acquiring up-and-coming scaleups?
> You can’t start with the technology and try to figure out where you’re going to try to sell it.
The Internet begs to differ. AI is more akin to the Internet than to any Mac product. We're now in the stage of having a bunch of solutions looking for problems to solve. And this stage of AI is also very very close to the consumer. What took dedicated teams of specialised ML engineers to trial ~5-10 years ago, can be achieved by domain experts / plain users, today.
> We're now in the stage of having a bunch of solutions looking for problems to solve.
We've always had that.
In olden times the companies who peddled such solutions were called "a business without a market", or simply "a failing business." These days they're "pre-revenue."
Maybe it will be different this time, maybe it will be exactly the same but a lot more expensive. Time will tell.
I think you’re missing the point. Of course you can make such a product. As Steve says right after, he himself made that mistake a lot. The point is that to make something great (at several levels of great, not just “makes money”) you have to start with the need and build a solution, not have a solution and shoehorn it to a need.
The internet is an entirely different beast and does not at all support your point. What we have on the web is hacks on top of hacks. It was not built to do all the things we push it to do, and if you understand where to look, it shows.
> excellent show "halt and catch fire".
I found it very caricature, too saturated with romance - which is untypical for tech environment, much like "big bang theory".
IMO it really came into its own after the first season. S1 felt like mad men but with computers, whereas in the latter seasons it focused more on the characters - quite beautiful and sad at times.
I vaguely remember that they tried to reboot it several times. So the same crew invented personal computers, BBSes and the Internet (or something like that), but every time they started from being underfunded unknowns. They really tried to make the series work.
It's still very good I'd say. It shows the relation between big oil and tech: it began in Texas (with companies like Texas Instruments) then shifted to SV (btw first 3D demo I saw on a SGI, running in real time, was a 3D model of... An oil rig). As it spans many years, it shows the Commodore 64, the BBSes, time-sharing, the PC clone wars, the discovery of the Internet, the nascent VC industry etc.
Everything is period correct and then the clothes and cars too: it's all very well done.
Is there a bit too much romance? Maybe. But it's still worth a watch.
I never really could get into the Cameron/Joe romance, it felt like it was initially inserted to get sexy people doing sexy things onto the show and then had to be a star crossed lovers thing after character tweaks in season 2.
But when they changed the characters to be passionate stubborn people eventually started to cling to each other as they together rode the whirlwind of change the show really found its footing for me. And they did so without throwing away the events of season 1, instead having the 'takers' go on redemption arcs.
My only real complaint after re-watching really was it needed maybe another half season. I think the show should have ended with the .com bust and I didn't like that Joe sort of ran away when it was clear he'd attached himself to the group as his family by the end of the show.
Nice! I added this to my AI metaphor collection.
Another one I like is "Hungry ghosts in jars."
https://bsky.app/profile/hikikomorphism.bsky.social/post/3lw...
AI is not a horse (2023) https://essays.georgestrakhov.com/ai-is-not-a-horse/
We don't know it, up to the point we observe it.
AI is a quantum mechanic
But since the act of observation influences the object observed, who knows what then becomes of it?
Maybe AI is a centaur??
After Deep Blue Garry Kapsparav proposed "Centaur Chess"[1] where teams of humans and computers would complete with each other. For about a decade a team like that was superior to either an unaided computer or an unaided AI. These days pure AI teams tend to be much stronger.
How would pure ai ever be "much stronger" in this scenario?
That doesn't make any sense to me whatsoever, it can only be "equally strong", making the approach non-viable because they're not providing any value... But the only way for the human in the loop to add an actual demerit, you'd have to include time taken for each move into the final score, which isn't normal in chess.
But I'm not knowledgeable on the topic, I'm just expressing my surprise and inability to contextualize this claim with my minor experience of the game
If you had a setup where the computer just did its thing and never waited for the human to provide input but the human still had an unused button they could press to get a chance to say something that might technically count as "centaur", but that isn't really what people mean by the term. It's the delay in waiting for human input that's the big disadvantage centaur setups have when the human isn't really providing any value these days.
Baxtr, JAMES BAXTR? That's the exact comment I'd expect of someone named that.
Or a reverse-centaur ? https://locusmag.com/feature/commentary-cory-doctorow-revers...
Or a reverse centaur [1].
[1] https://locusmag.com/feature/commentary-cory-doctorow-revers...
Maybe from the client's point of view, although it's more likely a Tamagotchi. But from the server side, it’s more like a whole hippodrome where you need to support horse racing 24/7
It's also a big bloatey gas bag that needs constant de-farting to function
So essentially a cow?
Oh horses fart a lot too.
Horses poop a lot. A lot.
I had to search about and it's indeed a lot:
"it is quite normal for a horse to poo (defecate) 8-12 times a day and produce anywhere from 13 to 23 kilograms of poo a day."
That's what you get when your primary source of nutrition is very calorie-poor and largely indigestible.
Yup. I’ve noticed that with my dog going to meat from kibble. Poop sizes reduced by 80%.
More than pooping a lot, they literally cannot hold it. Humans don't poop that much, but imagine if everyone just did it on the floor at a moment's notice regardless of where they are
"2024 AI was a horse". People really like to imagine that the last 6 months constitute their true observation of the new eternal state of the future.
When did a horse ever give anyone psychosis?
So it’s a car.
"No, I am not a horse."
Horse rumours denied.
*sweats profusely* https://imgur.com/a/PszeiAu
That's something a horse pretending to be AI would say.
So... are we having AI races?
That's not from the last week, so obviously doesn't matter.
All true apart you can only lead it to water - it drinks ALL the water regardless of anything else.
Except when you want it to improve something in a particular way you already know about. Then god forbid it understands what you have asked and makes only that change :/
Some times I end up giving up trying to get the AI to build something following a particular architecture or fixing a particular problem in it's provious implementations.
Some day, I imagine one will be a senator
We only have enough budgeted for one joke in 2026 and this is the one.
A horse that can do your homework.
Yeah, well... not really.
I used to tell my Into-to-Programming-in-C course students, 20 years ago, that they could in principle skip one or two of the homework assignments; and that some students even manage to outsmart us and submit copied work as homework, but - they would just not become able to program if they don't do their homework themselves. "If you want to be able to write software code you have to exercise writing code. It's just that simple and there's no getting around it."
Of course not every discipline is the same. But I can also tell you that if you want to know, say, history - you have to memorize accounts and aspects and highlights of historical periods and processes, and recount them yourself, and check that you got things right. If "the AI" does this for you, then maybe it knows history but you don't.
And that is the point of homework (if it's voluntary of course).
I was expecting a spin about the faster horses
Ai is a horse, i get it! I have a horse, and I put money in the front of the horse, and get "ponyium" out the back.
Through many attempts to make ingesting the ponyium more bearable, I’ve found that taking it with more intense flavors (wintergreen mint, hoppy hops, crushed soul, dark roast coffee, etc) improves its comestabilty. Can’t let it pile up. We’ve always eaten ponyium right, and we all like it, right, guys, folks?
And the salesman always says it’s great while it’s in fact lame.
I've always said that driving a car with modern driver assist features (lane centering / adaptive cruise / 'autopilot' style self-ish driving-ish) is like riding a horse. The early ones were like riding a short sighted, narcoleptic horse. Newer ones are improving but it's still like riding a horse, in that you give it high level instructions about where to go, rather than directly energising its muscles.
you rather don't want it in your bed
this post is aging like milk
This micro blog meta is fascinating. I've seen small micro blog content like this popping up on the HN home page almost daily now.
I have to start doing this for "top level"ish commentary. I've frequently wanted to nucleate discussions without being too orthogonal to thread topics.
"I've been through the desert
On AI with no name
It felt good to be out of the rAIn
In the desert, you can remember your name
'Cause there ain't no one for to give you no pain"
you forgot to write pAIn and it reminded me of this: https://youtube.com/watch?v=nt9mRDa0nrc
>2 views
I'm not saying that's your video but it sure looks like that's your video ;)
If an AI aims at the thing we call it hallucinations, when humans do it we call the delusion goal setting.
Either way it is an imagined end point that has no bearing in known reality.
Or your typical American teenager.