Really makes you wonder who is going through the reasoning traces and how many of them they are parsing through per-hour for their eyes to completely gloss over at seeing this rather than recognise that this, in fact, completely legible.
What's also interesting is the use of emojis to represent dead ends. I guess they really have become morphemes now.
Err, yes, until the surplus kills off all other competition and allows the supplier to jack prices up sky high, or otherwise bend consumers to their will. There's a reason most countries will stop foreign firms from doing this to them.
Except there's lots of competition for creating that surplus, including from open source locally-hosted LLMs, and while it's behind the frontier it's not that far behind the frontier.
The dumping -> non-competitive price increases playbook is historically very, very rare, and relies on a monopoly (or in a few cases oligopoly) with large externally-enforced barriers to entry. The oligopoly case is highly unstable and doesn't last, and besides we don't have notable barriers to entry; we have both market competitors and locally-hosted imperfect substitute goods.
There's essentially no reason to believe the dynamic you're predicting could succeed here, because we lack all the conditions that make it more likely to succeed, and it's very rare anyway.
In this case, its not competition in the form of foreign company.
Its competition of engineers, scientists, and intellectual labor being atrophied due to overuse of LLM.
Pushing costs at 1/100 for 'thinking' gets intellectual labor people hooked and dependent. Then when costs go 300x, leaves people dumber and less capable of doing things on their own.
LLM companies, by not accurately charging for services, are directly dumping on world-level society and devaluing and addicting people to outsource thinking. Thats the problem.
In reality all subsidized and 'free' services do exactly this. LLM token vendors are making a play against human thought.
> So it is absolutely strange and contrasting to see you believe that LLMs are so weak as to create negative value while the CEO is asking about regulations because AI is too powerful.
You wouldn't ask a chemistry professor to write code. So just because LLMs create negative value for software development doesn't mean that they can't be helpful for bioweapons synthesis, especially considering the range of chemistry and biology sources Anthropic would have fed to its LLM that wouldn't be publicly accessible. The LLM doesn't even need to be particularly accurate so long as the amateur bioweapons researcher takes adequate precautions before following its instructions and does some background research beforehand.
This is a ridiculous stance to take. That LLMs are simultaneously negative value but can also help synthesise bioweapons. It’s the sort of stance you take when you already feel ideologically against AI. I don’t think it’s coherent.
It's more about information availability rather than intelligence. An LLM has had access to more information during its training period than you'd ever even come across over a hundred lifetimes. It has been trained on billions of books and articles across every single subject that exists on the planet. Can you imagine what real intelligence could do with all that information?
From your example, perhaps you mean "competence does not imply knowledge" or more accurately in fact "lack of competence implies lack of knowledge" i.e. !competence -> !knowledge, in that competence && !knowledge is common but !competence && knowledge is rare.
> If we're all broke/starving/being exterminated, who will the rich sell to?
Themselves. The economy is a big cycle where money changes hands to drive production i.e. things getting made. AI will simultaneously greatly increase production (especially once humanoid robots are as dexterous as humans) and make the humans whose jobs it will do economically irrelevant.
So the rich will buy and sell very nice things to each other while the rest of us get left out in the cold because we simply cannot compete with the robots. And because they will capture and control all resources (either by law or by force) we won't be able to create a functioning parallel economy either.
> So the rich will buy and sell very nice things to each other while the rest of us get left out in the cold because we simply cannot compete with the robots. And because they will capture and control all resources (either by law or by force) we won't be able to create a functioning parallel economy either.
Here's another framing for you: at this point _there are no longer rich and poor people_. There are fewer people, but we knew that was going to happen as a consequence of declining birthrates. The elderly are taken care of despite an otherwise unsustainable dependency ratio, because robots can manage the actual business of survival. In that world everyone is a member of the nobility by the virtue of being _human_. There are a few holdouts - mostly religious nuts and other cults - but by and large everyone who is willing to accept the machine's gifts has their every material need catered to.
There is no world where legions of filthy rich AI barons lord it over the technologically illiterate peasants, though. How could there be, when literally anyone can plop down $20 and get access to a frontier model? When open weight models trail _at most_ a year behind the closed ones and compute continues to proliferate?
One of the few things we have figured out about AI is that productivity gains are mostly captured by the people using the tools, not the person paying for the model. In other words, using an LLM is a skill and there is still no substitute for the human driving it.
> How could there be, when literally anyone can plop down $20 and get access to a frontier model?
For now. And that too at a massively discounted rate to drive adoption.
> When open weight models trail _at most_ a year behind the closed ones and compute continues to proliferate?
Open-weight models require computing power to run. Consumer hardware prices are rising because of AI build-out, so much so that companies that used to serve ordinary consumer markets are switching to serve only datacenters. Megacompute does indeed continue to proliferate.
> One of the few things we have figured out about AI is that productivity gains are mostly captured by the people using the tools, not the person paying for the model. In other words, using an LLM is a skill and there is still no substitute for the human driving it.
Will this be the case in 20 years? Agentic workflows have come as far as they have in about two years of existence. Do you really need the problem between chair and keyboard will be needed after another 10? And do you really think that in 20 years time that we will all be paid to prompt increasingly advanced and independent LLMs?
> everyone who is willing to accept the machine's gifts has their every material need catered to
The way automation is going, knowledge work will be automated first before any physical production processes are. A lot of people will lose their livelihoods before goods in particular become "the machine's gifts". What do you think happens then? Will the capital owners who have captured this reduction in costs reduce prices proportionally? Or will they keep the gains for themselves? Do you think governments around the world will tax the upper class to the point of being able to give everyone their current livelihoods through government benefits?
You are pretty much just describing some sort of fantasy automated communism. Not to mention, in your world, gatekeeping the machines would instantly become the most profitable venture possible.
> Not to mention, in your world, gatekeeping the machines would instantly become the most profitable venture possible.
Yes, it would! That's why frontier labs don't open-source their models :)
The point is that the technology is already too democratized for anyone to hold onto it. Google had chatbot LLMs in 2019 and tried to keep them under wraps, how many years did that buy them?
> Do you really need the problem between chair and keyboard will be needed after another 10? And do you really think that in 20 years time that we will all be paid to prompt increasingly advanced and independent LLMs?
I think that things are going to get so much cheaper that we'll still be paid more than enough.
> The way automation is going, knowledge work will be automated first before any physical production processes are.
So far, LLMs are great and all, but they only really "fill in the blanks." That's a fundamental limitation of the entire concept of modelling in general; you cannot generalize to out-of-distribution inputs. The bottleneck is going to end up being human beings no matter which way you slice it. Because the bottleneck will be people, more and more of them will be hired, even though each individual is incredibly productive. This is also called Jevon's paradox, when making a resource less expensive leads to overall market growing.
> You are pretty much just describing some sort of fantasy automated communism.
If you went back a thousand years ago and told someone carrying a bucket full of water that one day pipes would run across the civilized world and water would literally be free basically everywhere, they might react the same way. If VLA-driven robots start reducing manufacturing prices, is it so unreasonable to slowly expect more and more things to go that direction?
> The point is that the technology is already too democratized for anyone to hold onto it. Google had chatbot LLMs in 2019 and tried to keep them under wraps, how many years did that buy them?
They were hardly the only ones in the space. OpenAI has been around since 2015. GPT-3 was released in 2020 and ChatGPT in 2022. Not to mention, I wouldn't call something produced by a handful of megacorporations worldwide particularly democratized. In fact, Google's transparency is what allowed it to be democratized, because it published its findings about transformers publicly.
> So far, LLMs are great and all, but they only really "fill in the blanks." That's a fundamental limitation of the entire concept of modelling in general; you cannot generalize to out-of-distribution inputs. The bottleneck is going to end up being human beings no matter which way you slice it.
This is a laughably naïve take especially when LLMs have a) been trained on quite literally all the data the world can provide and b) are being trained more and more using reinforcement learning techniques - which don't rely on data at all and instead on producing emergent behaviour from a set of ground rules. With every new release their agentic capabilities improve and they become more independent, requiring only the impetus to get going.
> This is also called the Jevons paradox, when making a resource less expensive leads to overall market growing.
Oh yes, there will definitely be more software. That is guaranteed. What is not guaranteed is how many humans will be involved in making it. Just as more coal is being mined than ever but fewer people are involved in it. Efficiencies in coal mining aren't what made the average coal miner's working conditions or income better, regulations are.
> If you went back a thousand years ago and told someone carrying a bucket full of water that one day pipes would run across the civilized world and water would literally be free basically everywhere
If you told a Roman this, they would not be as surprised as you would think as aqueducts already existed back then. They would be more surprised that the common man had the ability to vote in most countries. I doubt it will stay that way with improvements in AI, at least not without a great reduction in population.
How is it inaccurate? If I only care about buying apples, and apples get 10% more expensive, and my salary only increases by 5%, then I can't buy as many apples as I could have before. How many apples I do actually buy in the end is irrelevant to the calculation.
The person you're replying to erroneously interpreted "stay even" as "avoid going into debt," instead of your income's purchasing power remaining constant.
Consumer price index is about consumer goods. This is why tarrifs and such are considered regressive - they hit people harder the less money they have because a larger percentage of their spending is consumer goods.
If I invest half my income and spend half my income, and the prices of goods goes up 4.2% and my income goes up 4.2%, then I've made progress; I'm now investing more than half my income, because the half of my income I was spending has stayed even and the half I was investing has increased.
> If I invest half my income and spend half my income, and the prices of goods goes up 4.2% and my income goes up 4.2%, then I've made progress; I'm now investing more than half my income, because the half of my income I was spending has stayed even and the half I was investing has increased.
Let's say your income is $100. You spend $50 and invest $50. The prices of goods goes up 4.2%, so to keep your current living standards, you must now spend $51.20. Your income increases by 4.2% to $104.20. After expenditures, you now have $51.20, or exactly half your income, to invest. So you haven't made any progress. And investing $50 now is equivalent to investing $47.98 before in terms of what you could have bought instead of investing.
It's hardly speculative when it is effectively what happened just after the Industrial Revolution, but with more power ceded to capital. In many ways, it's already happening.
No, that was not "effectively what happened" in the Industrial Revolution. That was an enormous change, but it didn't "completely eliminate the need for a human workforce." That's just hype.
Fine, it is not effectively what happened then. It is worse. At least workers are required to run factories (even though working conditions were ridiculously horrible back then). With AI, in maybe 20 years, 95% of all white-collar workers will be economically irrelevant. You won't need accountants, or programmers, or designers. And we can't all become lawyers and surgeons, or tradesmen.
The Industrial Revolution indeed did not completely eliminate the need for a human workforce. The AI Revolution will.
That is the status quo now. If LLMs get better and better, managers will simply direct them, well, directly. That's what the parent comment means by the machine learning to do your job. It completely, not mostly, replaces you.
I have some doubt here, because I do know that managers of at least some seniority have the political antennae to be wary of taking responsibility for something they don’t understand blowing up in their face. If they use LLMs directly, then the number of production outages will increase with the use of these powertools in unskilled inexperienced hands. The causes will probably just tilt towards design mistakes and unforeseen exceptions rather than coding language defects. I can see them wanting to keep some techies around simply to have someone to blame. You don’t get far up the ladder without learning how to cover your ass.
To all the managers reading this and thinking "phew. so long suckers!", I'd be willing do what you do for 50% of the pay. Surely it is _I_ that will be irreplaceable!
I don't think they'll let the chain of managers above you handle the llm directly. That is just too much risk of incompetence. Instead, there will be micro teams (1 dev, 1 sre, 1 product owner) that are meta manageed by a LLM. And their llm reports directly to a higher up's llm. And software will diversify to prevent all these supply chain attacks we've seen lately.
Managers can barely direct me without shitting their pants. What saves them most of the time is my ability to say "No". Until LLMs can do that, which seems quite hard to do so far, good luck replacing me.
This really underscores how so much LLM "intelligence" is based off of people's experiences that they wrote about. It saddens me somewhat to see that it has basically all been captured by corporations now, and perhaps in a few years there will be little point in knowing things yourself simply because the LLMs will have gotten so good that there is no point to wasting the extra effort. Of course, this might lead to the atrophy of people's thinking muscles and a dumber and more subservient populace, but does anyone really care? Judging by the meteoric rise in the capabilities of LLMs over the past five years, is it really naïve to expect most knowledge work to be obsolete in the next ten?
The other stuff in this thread about decompiling firmware is a lot more interesting to me, though, seeing as it used to be a fairly demanding but rewarding task that has now been "solved" by Claude. It's a magic trick that is a lot harder to pull off than the other things in this thread.
What's also interesting is the use of emojis to represent dead ends. I guess they really have become morphemes now.
reply