Oh and I was forgetting that in the process of this failed pivot they also made multibillionaire out of snotty kid who now is into building drones that dont work.
I do work in software, the entire situation in big tech and downstream in software dev works all over is complete lalalaland, driven by opaque, dark forces.
If you've been working in this sector and you've not been preparing for the music to stop anytime now, you're dumber than a rock.
Twitter laid off like 90pct of the workforce and the site is still there. No LLMs involved.
META not 5 years ago burned hundred of billions on a completely failed pivot that would have killed pretty much any other company/executive.
Every analysis of AI should start with "why did the Internet fail in procuding anything of value beyond burrito taxis and a couple people getting richer than God, and also ushered in the Nazi?"
And it is going down _exactly_ as a huge "big nice centralized massively diffused cognitive multiplier buckles under stress" event.
Not talking about AI but the Internet.
Again, the proof if the pudding is in the eating.
The rise of the most powerful cognitive multiplier since at least the printing press is coincident and is causing, by and large, the rise of _the literal fucking Nazi_ in the US and half of Europe.
- I might add, a multiplier that can be centralized by less companies and people that can be counted on one hand is worse of one that can be reproduced of for which an open market exists.
- A multiplier that fails unpredictably is better than one that fails predictably.
- A multiplier that communicates its limits is better than one that does not.
- A multiplier that beckons is worse than one that is passive.
Etc etc
Not saying that this is not directionally correct, but when you make a general case and then all your specifics are flimsy, maybe you should take another look at the general.
There is a big set of properties of "intellectual multipliers" that make them better vs worse in their function.
I agree in principle, but this is very fraught and I think it shines through in the fragility of the examples:
- People who watch TV all the time are predictably idiots
- Using Google Maps for navigating your neighborhood *will* come biting you in the ass
- Restaurant reviews absolutely suck
I broadly agree but 2 fine points:
- 75k here is for a couple. If you take tax rates for France (!!!) convert to EUR PPP, you'd be taxed 5% average on that income.
- "just tax the rich" *was* pure slopulism but is becoming quantitatively correct because income distribution go brrrrr.
Of course LLMs pick up personas above all so if the "consensus smart" turns out to be dumb they will also fail along those lines.
My logic is that when I interact with someone in professional capacity and I see them doing or saying something dumb, my expectations of their political views being nazi shit also increase but this is fully mediated by dumbness.
Throwing this out there: what if the vector is not good/bad but smart/dumb (or wise/ignorant) and the good/bad effect is carried by the socratic theory of evil?
LLMs seem to prove that you can emulate this decently with statistics.
But the most mysterious part is why this ends up potentially being a powerful heuristic to search for *real* stuff, ie the language-object itself is imbued with "unreasonable effectiveness".
I'm on the fence about the whole thing but I think it's a common experience in all kinds of language production (speaking/writing/coding/maths) to "just let the words flow".
I think the big problem with LLMs is that they (successfully) manipulate the machinery of language, so you're basically building on top of an already mysterious construct.
Ultimately I suspect most of their real world utility will boil down to the "unreasonable effectiveness" of mathematics.
Its a cool metaphor but the stuff that comes after is bad and betrays the low value of the metaphor.
A cookie cutter CNN is not "grown" and actually structure is *strongly imposed*, also viz techniques for thesr are established and its easy to retrieve consistent shape/texture matchers.
Basically almost every single cent of saved "efficiency" from large scale farming goes straight into making red meat. And the kicker is we know industrial scale red meat is bad almost on any conceivable level.
If you compare calorie for $ between annoying everything-by-hand crops thst have a bit of calories (e.g. bananas, specialty potatoes) and basic ass chicken, the vegs still come out on top by a mile. I've visited banana farms in the canaries they're minuscule. They sell that shit in the supermarket.
Then, farms are like 1% of gdp. That's it. Even if you drop efficiency by like 5x, it would make 5%, thus refucing gdp by 4%, not "make economy go boom"
Do consumption and prices probably look different in non-mega-agri world? Obviously.
But the claim that you just can't make it work are absurd no matter how you slice it.
For starters, a lot of the technical advances of past 2 centuries also enormously increased productivity of small farms.
A little farm (1ha) produces 10x/20x in vegs what an individual consumes.
You talk about "efficiency of the farms" but 80pct of the farmed biomass in the US is used as livestock feed.
The livestock converts calorically at <5pct efficiency.
more freebasing than freelancing tbh ngl
Yeah I think the disconnect for me happens because (possibly as a reaction to megaregulationism bringing about abject market failure in real estate in blue states? Abundance agenda?) many want to frame this as "technology automatically gud" vs "luddites".
Also our dude from 1726 got hands.
Let's not make too much noise lest we wake up the dudes from 1526 and 1626.
I'm sure all the "technological innovation" happened in those ~15 years since 1850 where the working hours drop like a stone. Absolutely random that all the big drops track precisely with the history of trade unions.
