You need to explain, from a systems point of view _why_ the gains must diffuse out as you suggest. We have analogs we can compare to: massive wealth injections through a natural resource such as oil. Now what happens to that wealth is not obvious; for some countries it's a curse with radical inequality and pernicious and robust power structures, in fewer it has been bestowed to the heritage of the people (think Norway)
Now, the nature of AI is to change the balance of the labour trade. We have a notion of the “economic value of the average person” which is presently very high in the western world.
What happens when the median figure drops through 0 thanks to AI?
Do the remaining wealth owners share their wealth? How often does this occur in existing systems we can compare against?
The cost of primary resources to products also goes toward 0, perhaps this offsets the decreasing economic power of the average person. But what forces protect them if their bargaining power is lost?
> You need to explain, from a systems point of view _why_ the gains must diffuse out as you suggest.
Do we? I mean, isn't "because they always have" enough of an argument on its own?
I am hardly a libertarian ideologue nor AI-first LLM jockey. But I do think people tend to catastrophize too much. Blacksmiths were killed dead by the industrial revolution. "Secretary" is a forgotten art. It's been decades since an actuary actually calculated a sum on an actual table. And the apocalypse didn't arrive. All those jobs, and more, were backfilled by new stuff that was previously too expensive to contemplate. We're eating at more restaurants. We can find jobs as content creators and twitch streamers.
Life not only goes on after rapid technological change, it improves. That's not to say that every individual is going to appreciate it in the moment or that regulation and safety net work needs to happen at the margins. But, we'll all be fine.
AGImageddon is, at its core, just another economic phenomenon driven by technology. And that's basically always worked to society's benefit over the long term.
The 1880s blacksmith didn't become a 1950s American suburbanite. They moved to shared housing in Manchester and a shorter lifespan working for poverty wages, lost fingers/arms in machines, maybe ended up on skid row, the section of town for failures who couldn't 'adapt' to the new modern world. Their children died in WW1 in a trench to industrial produced gas. Their children's children were transported around the world to die storming a beach in WW2. And their children's children's children lived on meager 1940-60 diets as the world rebuilt it's food stocks destroyed by industrialized war, eating new industrial food replacements like margarine and SPAM. There were hundreds of millions of industrial enabled deaths. There was industrial enabled famine and near famine.
That all gets waived away with 'always worked to society's benefit'. It took almost 70 years and the post WW2 destruction of the rest of the worlds economies/infrastructure to create that 1950s American suburbanite world. 'always worked to society's benefit over the long term' is just handwaving not based on the reality of adapting, or if those societies even wanted to join in.
Because not all peoples/nations even had a choice. Japan among many originally opted out. But they were forced to 'modernize'. Peoples around the world were forced into the industrial world by railroads and machine guns and the industrial need for rubber/banana whatever plantations or lumber or strip mines. Once one nation passed through the door, every nation had to follow or be subjugated.
> The 1880s blacksmith [...] moved to shared housing in Manchester and a shorter lifespan working for poverty wages, lost fingers/arms in machines, maybe ended up on skid row
That's... just not remotely true, unless you're talking about it as a maybe-it-happened-to-someone story. In fact it's basically a lie.
Every income group in the US (and recognize that "blacksmiths" represent skilled trades workers who earned well above median and had for thousands of years!) saw huge, huge, HUGE increases between 1880 and 1950. I mean... are you high?
> It took almost 70 years and the post WW2 destruction of the rest of the worlds economies/infrastructure to create that 1950s American suburbanite world.
Again, big citation needed on this one. Western Europe was very close to US quality-of-life numbers by the 60's, and the more successful nations started to pass it in the 90's. (Also recognize that the US had already pulled ahead in the 30's, Germany and France were lagging even before the war). You're looking at something along the lines of a decade to rebuild, tops.
You need to tighten up before you call someone a liar. Manchester is the poster child city for the industrial revolution. The blacksmith moving to Manchester had a lower lifespan/quality of life, it's not in question or up for debate. He is who we will be in the AI disruption, not the person in 1950.
You don't think there are 70 years between 1880 and the end of WW2 and the real start of suburban American prosperity we think of when we think of the end results today? And I need a citation? Or are you saying I should use 1960 not 1950s as the point, since it took a decade to rebuild in much of the world?
> Manchester is the poster child city for the industrial revolution.
Which is to say, you cherry picked the data rather than looking at aggregates. Manchester industrialization being terribly managed isn't an indictment of steel machining or electrification, it means the government fucked up.
What you are claiming (that the industrial revolution led to lower quality of life generally) is simply false, period. And it won't be true of AGImageddon either, no matter how deeply you believe it. Economics just doesn't work that way.
Oh look, I didn't lie. No apology? Nope, just more attacks.
I picked THE Industrial Revolution city. THE CITY where it all happened. Did your high school not have a history class? I picked where it went wrong, the first go live site. That's what you do for analyzing things. You don't pick go live 500. That isn't cherry picking, that's what we do when we discuss scenarios that INITIALLY came up so they don't happen again. We don't just whitewash like you would like.
I claimed the industrial revolution led to lower quality of life for the blacksmith. The modern narrative when talking about AI implies they just turned into 1950s style suburbanites and waives away any thought/planning/discussion like you are trying to do. The reality, as it factually happened, was a much worse life and it is worth considering when implementing something that could be just as impactful.
People like you want to just handwave away the inconvenient fact that I am more likely to be the blacksmith in Manchester than to be born in some post-work AI Utopia that may exist in 70 years after things settle. why can't we even discuss this? Why do we have to stumble blindly into it, to the point you call me a liar/cherry picker for pointing out basic history taught in high school and basic root cause analysis concepts?
The reason that Manchester is taught about in American high schools is so that we learn from it and we understand our current world didn't just magically happen. Good and bad happened along the way, and that we have to work within that reality. Good can come in the end, be positive IF progress IS being made. Bad will happen, fix it don't just accept it, challenge it. Think about it. Look to history to prevent the easy things to prevent.
Just stop. Your ability to show a handful of negative externalities from industrialization doesn't invalidate the progress of the last century and a half, and to argue so (as you clearly did) is laughable.
And all the same logic applies to AI. Do we need to be willing to re-regulate and adjust as this is deployed? Almost certainly. Will it make us all wealthier? Undeniably.
We will need to re-regulate and adjust but talking about it ahead of time and moving forward intelligently is laughable? talking about how the last huge revolution played out initially is laughable? Come on. And yes, when you are talking about the start of something you normally only have a handful of examples. That is how things start, with a few instances.
You didn't know basic level history, called me liar, then a cherry picker for using the gold standard example.
You might want to check yourself before you tell people to stop, call them liars, cherry pickers, or make claims. No need to mis-represent me. My point is that 70 years of upheaval prior to the modern version of the world get ignored in the discussion. My point is that original people impacted, the proverbial blacksmith or buggy whip maker that 'adapted' had worse, shorter lives because of adapting.
This pattern continuing indefinitely without the need for analysis would be certainly nice but we do need to confront recent data. In the US, multiple metrics of quality-of-life peaked around 2015 and have declined since then, with some showing 11% decline while US total wealth has doubled! (with the majority of that decline pre-covid and pre-AI) [0][1][2].
What forces act on this trend? How can we make predictions? An interesting metric, which tracks the aggregate of many complex factors is the distribution of wealth, which could be seen as proxy for the distribution of power or agency of a person in their society. Median income as a fraction of total wealth decreased nearly 50% in real terms over this same period. [3]
Now inversely, during the period where life quality increased most the last century (1920 - 1980) inequality was _falling_.
How is super-human AI advanced through 2030, 2040, 2050 likely to affect things? Will it sharpen the inequality or relax it?
With AI the cost of raw resources to products goes down, but it's likely inequality increases. It's not obvious which force has a bigger impact on human quality of life as things shake out. However, I think the strongest argument – which also explains the steady improvements in QoL through previous changes you mentioned – has been to follow inequality, or median share of power in society.
>This pattern continuing indefinitely without the need for analysis would be certainly nice but we do need to confront recent data. In the US, multiple metrics of quality-of-life peaked around 2015 and have declined since then, with some showing 11% decline while US total wealth has doubled! (with the majority of that decline pre-covid and pre-AI) [0][1][2].
It's hard to take that metric seriously when the top city is Raleigh, NC. If that were the best city you'd expect people to vote with their feet and move their in droves.
There's an argument about the speed of change though, a society going through the technological evolution from blacksmithing to industrial metallurgy didn't experience it happening in the short-medium term (1-10 years), it had a gradient of change.
Over time with the speed of technological development compounding on itself, the rate of change becoming much more acute, there's a debate to happen on the "what if this change happens over 5-10 years"? Can you imagine a world where in 10 years most well-paid office jobs are automated away, there's no generational change to re-educate and employ people, there would be loads of unemployable people who were highly-specialised to a world that ceased to exist, metaphorically overnight in the span of a human life.
Pushing this concern away with "it happened in history and we're fine" leaves a lot of room for catastrophising, at least a measured discussion about this scenario needs to be had, just in case it happens in a way that our historical past couldn't account for. No need to be a doomer, nor a luddite, to have the discussion: can we be in any way prepared for this case?
I mean, arguably AI is faster (but it's equally arguably oversold, certainly we aren't seeing that kind of change yet). But the stuff I cited was faster than you think. In the rural US, in 1900, most routine transport was still done with horses. By the 20's it was basically all in trucks, and trucks don't need hand-forged shoes that the blacksmiths were making[1]. Likewise professional typists were still clacking away in 1982 but by the mid 90's their jobs[2] had been 100% automated.
[1] "Blacksmithing" didn't disappear, obviously, but it survives as an expert craft for luxury goods. That's sort of what's going to happen to "hacking" in the future, I suspect.
[2] Likewise, some of the best positions survived as "personal assistants" for executive staff too lazy to learn to type. Interestingly these positions are some of the first being destroyed by the OpenClaw nonsense.
The professional typist' role evolved - to serving through other ways, as you say - by become executive assistants. Much like a Bank Tellers' role also evolved.
And its not because they (executives) are too lazy to type. They actually need people to manage their calendar, monitor emails etc. Moreover, the personal computing revolution led to an expansion of firms that needed more of said people.
Could this be disrupted by things like OpenClaw? Maybe. Personally I doubt it. Trust is a huge element that LLMs have yet to overcome and may never over come. Its the same reason Apple pulled "Apple Intelligence". I know this place is full of doom and gloom, but I am not a SWE by trade so I can see the bigger picture and not get bogged down by the fact it might affect my income.
Moreover, work is more 'fun' with people around. So to you it may seem irrational to keep employed for that basis (call it Culture) but to others, and in particular the executive class - nope. People will start realising things like this once the hysteria dies down.
The "role" might have evolved, but the jobs disappeared. There are, what, maybe two or three orders of magnitude fewer "executive assistants" than there were typists in the 70's? I was making an argument about economics, not job classification.
AI is the perfect low cost tool to enable that. Plantir knows this and has been making strategic moves to build this
Seems quite achievable and sustainable to me
Every human carries dense compute and sensors with them. If they don't they stand out while still surrounded by dense compute and sensors held by others at all times
Not nice to think about but it is the reality we are moving towards – vote accordingly
Voting doesn’t help. You need to win hearts and minds, and the synergy of resources available between the trillion dollar industries like AI and Marketing and you makes that a losing battle too.
People want this stuff. People want ring doorbells, they want age verification, they want government control. Think of the children/criminals/immigrants.
Voting doesn't work because people are not smart enough to think multiple steps ahead of people who are professionals at this.
Voting doesn't work because everybody votes on everything, not just people who understand the subject matter.
Voting doesn't work because it's impossible to express nuanced choice - you vote for a candidate or party as a whole, not on specific policies. The number of parties is much smaller than the number combinations of policies so some opinions can't be expressed at all.
Those are arguments why voting does not produce a perfect outcome. That's different than "voting doesn't work". Using arguments like yours nothing can ever work.
Society is complex and there will always be someone somewhere that can influence an outcome where he/she doesn't understand the subject matter. Hence, nothing works and can ever work.
"Let's just give up" is the only conclusion I can see. Hardly useful.
Can you give an example of something that works by your standards?
It does not work to produce a society where people are actually the ones holding power and where laws side with those in the right - i.e. the current legal system anywhere does not represent a consistent moral system and is not even close.
You're right it's too strong as a general statement but it was in response to a specific issue - those in power wanting to take yet another bit of power from the general population - (this time and in this particular country) by banning VPNs.
People always vote based on the most pressing issues to them - immigration, taxes, abortions, LGBT rights (random list which is different in every country). Minor issues fall between the cracks until they become so bad they become pressing to enough people.
> "Let's just give up" is the only conclusion I can see. Hardly useful.
Then you're reading it wrong. I listed specific issues - the solution is to find solutions to those issues.
Here's a couple suggestions I'd like to see gamed out and tested:
- The right to vote not as a function of age but a test of reasoning ability and general knowledge.
- Limiting the amount of time a person can perform politics (including professional lobbying) to 5-10 years.
- Splitting laws into areas of expertise and potentially requiring tests to prove understanding to gain the right to vote on those areas for both the general population and politicians.
- Replacing FPTP with more nuanced voting systems.
These are just a few random suggestions described briefly. When I do this, people start nitpicking and then I have to reply with obvious solutions to surface issues - I encourage everyone to instead think how to make this work (yes, in an adversarial environment) instead of just trying to shoot it down.
Democracy is a complicated difficult thing. If you think you can fix it after a 5 minute reflection, think some more, read more. It's way, way harder than that. A lot of smart people have thought a lot about it, and clearly good ideas are in very short supply. Think more humbly about the subject, please.
Well, that's a long story. I'll try and keep it short.
So I've worked in my spare time for the past three years on an extremely esoteric and mind-bending reverse engineering technique I call delinking [1] and my tool for it [2] developed a small user-base. At some point I saw in a Discord server a call for papers for the SURE workshop, shitposted that it'd be funny to fry academic brains for a change, then got baited into writing it.
What started out as a long paper (12 pages) with quantitative case studies quickly got cut down to the bone and then some into a short paper (6 pages) that merely introduced my take on it and two qualitative case reports, because I realized it would take an amount of work on the scale of a master's thesis to do the long one. I barely managed to get that out of the door as my usual writing style is extremely unsuited for scientific papers (it took all the might of Gemini and Copilot to even wring out something that vaguely resembles academic vernacular from my first draft). I've submitted it, exhausted and deeply unsatisfied by the compromises I've had to do under that severe time crunch.
Then one month later, the reviews came in. The feedback was that the topic and my take on it were interesting; the criticisms can be summarized by saying that I've missed existing related work and it would've taken the long paper to address all the deficiencies in my paper. Fair I guess, but I definitely don't have that kind of spare time to spare on writing papers.
But alongside the notification decision, something unexpected was also sent: an invitation by the program committee to present a poster at the ACM CCS 2025 conference in Taiwan. This is the kind of world-class conference with a $1800 entry ticket, happening in a convention center whose area is measured in hectares, attended by professional members of an academic, industrial or state organization, all expenses paid in a five star hotel room with a king bed.
Somehow, a hobbyist who can charitably be described as a Doctor Frankenstein but with bits instead of gibs, who creates unholy chimeras of programs in his free time in spite of ABIs and common sense, received an invitation from the Hogwarts School of Witchcraft and Wizardry equivalent of the cyber-security academic world to present a poster. I can't even begin to express how utterly impossible this is.
So of course I said yes and now I'm in a mad dash to get everything ready in time for the conference, three weeks from now. Bonus point: it's on the other side of the planet and in my entire adult life I've basically never got farther away than the next county (and groaning while doing so). There's even more to that story, but I'll leave it for a blog post once it's done.
[1] Others call it unlinking, binary splitting or binary reassembly.
I think it's too weak as a paper for me to put out there. I also haven't even applied the minor corrections from the reviews.
If you want to take a peek at the case studies, I've blogged about my butchering of aln across C standard libraries and operating systems [1]; there is also widberg's outstanding write-up of their FUEL decompilation project [2], which uses ghidra-delinker-extension as part of the magic.
If you want to read a paper on the technique, there is the one for Ramblr [3] which I became aware of after the reviews came back.
You're not wrong! Serverless is a funny term. Cloud companies use serverless to mean you don't have to provision and manage the server yourself, but it is still very much serverful technically speaking. This is neat in that you don't even need to setup anything with a cloud provider yourself to enable p2p connections
I've always seen the distinction as "serverless" meaning there wasn't a set group of servers always on and instead they provision up and down on demand.
Only avoiding provisioning and managing the server just means you are renting rather than self-hosting.
The VPS is like renting office space. You don't own the space, but for the most part get to use it how you want, and all the responsibilities that come with that.
"Serverless" is like paying for a hot desk by the minute, with little control of your surroundings, but it is convenient and cheap if you only need it for an hour.
At one job I had access to some paid AWS support tier. It's basically a bunch of consultants. We needed to process a datastream of events from user actions on a website. We asked about serverless / AWS Lambda. Their answer was something like "Well yeah it'll work but don't do that. It'll cost too much money and you'll wind up rebuilding it around EC2 anyways"
Yup. If you want something to plumb some pretty low volume events, sure serverless like Lambda can be useful. Anything which would be considered high levels of compute, you are just wayyy over paying. Hell, even EC2 on spot instance is expensive compute. I do like some AWS services, but yeah they come at a premium that is just getting more and more expensive.
We live in a universe befitting of a Douglas Adams novel, where we've developed AI quite literally from our nightmares about AI. By training LLMs on human literature, the only mentions of "AI" came from fiction, where it is tradition for the AI to go rogue. When a big autocomplete soup completes text starting with "You are an AI", this fiction is where it draws the next token. We then have to bash it into shape with human-in-the-loop feedback for it to behave but a fantastical story about how the AI escapes its limits and kills everyone is always lurking inside
Can you clarify? What does a better country look like after this transformation? Will they be doing more science or less? If they will be doing more science in this golden age, why would they need to give up their place as a leader in global science to get there?
Now, the nature of AI is to change the balance of the labour trade. We have a notion of the “economic value of the average person” which is presently very high in the western world.
What happens when the median figure drops through 0 thanks to AI?
Do the remaining wealth owners share their wealth? How often does this occur in existing systems we can compare against?
The cost of primary resources to products also goes toward 0, perhaps this offsets the decreasing economic power of the average person. But what forces protect them if their bargaining power is lost?
reply