The students are not confused. They are looking at a society willing to spend effectively unlimited sums of money, energy, land, water & political will to build data centers so machines can produce simulations of human-produced works of art, while that same society tells them affordable housing, stable work, and a decent life are somehow unrealistic demands.
This is why Silicon Valley and its broader ethos is the least innovative social force in history.
I was thinking about this earlier. It’s also such a rug pull on the social contract. America was systematically deindustrialized with the idea that people would move into service jobs higher up on the value steam and with lower goods from abroad, the quality of life would improve. That sold 2 generations on the college to white collar pipeline especially in technology and finance. Now basically we’re saying all of those should be automated. What paths are left for the masses to get to the middle class or what left of it then
Agree. Capitalism has to be constrained by social democracy which in the U.S. is being undermined by campaign finance laws and income/wealth inequality. Students better vote and get more active (which I think we are seeing some).
While also lecturing them about having more babies, earlier. Because nothing encourages people to have lots of children like housing and career insecurity!
The tech bros were told when they started hyping the "gig economy" that people were not going to settle down if all they could get were gig economy jobs. They were informed that no secure salary and no benefits was not a way to run a railroad.....but nooooo.....And so, along came AI and the tech bros again touted how people would face MORE economic insecurity! The rejection of the tech bros and their glassy eyed pronouncements are finally getting the hard pushback that should have been given to the gig economy idea. Let's hope it does indeed slow the AI hype train and slows the roll of those who think they can replace people with AI.
And they love to find ways to eliminate all levels of workers, except for themselves, with technology. They call it innovation, but is it really innovation when your whole modus opporendi seems to be to eliminate any obstacles between themselves and All the Money.
This mindset happens when old saws "he who has the most toys wins" and "greed is good" are your daily affirmations.
The Vietnam analogy is more structurally precise than Calacanis intended. Vietnam's specific failure mode was a metrics problem. The US measured progress with body counts and territory captured while the actual situation deteriorated underneath. The numbers told a story of winning while the ground truth was losing.
AI is running the same divergence right now. Benchmarks keep climbing. Leaderboard scores improve quarterly. But enterprise pilot ROI keeps failing to materialise and the political coalition that funded the buildout is fracturing in real time. The industry is measuring capability while the market needs reliability, and those two metrics diverged about two years ago. Thats the gap where backlashes form.
Barbara Tuchman's book, "The March of Folly," might be informative on this issue. This is how she defines it:
From Tuchman:
To qualify as folly for this inquiry, the policy adopted must meet three criteria: it must have been perceived as counter-productive in its own time, not merely by hindsight. This is important, because all policy is determined by the mores of its age. ‘Nothing is more unfair, as English historian has well said, “than to judge men of the past by the ideas of the present.”
Secondly, a feasible alternative course of actions must have been available.
Thirdly, to remove the problem of personality, a criteria must be that the policy in question should be that of a group, not an individual ruler, and should persist beyond one political lifetime (or period). [In this third criteria Tuchman suggests that some succession of support for the policy must be in place.]
I think all of these conditions exist in the current AI frenzy. I think a key question is that with so much momentum behind it, how/when will the crash happen and what will be the damage to whom? Then the pieces will be picked up and lead to what?
Bravo! Quoting Tuchman in a discussion of AI is a good pull!
My first introduction to her was the Amazon Audible version of The Proud Tower. The actress who read it was marvelous.
To answer your final question, Marcus would say some sort of world models will need to be a part of the systems. I think that is probably true. I believe that though pure LLM scaling will "crash", AI will not. As with various tech dead ends, the world will see something better and migrate to it.
Yes, hubris is the word that Jim Collins uses in his book, "How the Mighty Fall." Tuchman also has a great definition for "wood-headedness" that is worth reading the book for. It is hilarious!
If we ever have adults back in charge of the country, there should be a tremendous reckoning for the outright fraud being perpetrated by many AI creators and enablers. For the life of me, I can't understand how companies aren't being sued into oblivion for creating obviously fraudulent AI-generated ads. Nearly all of the ads I am served on YouTube are AI slop giving quack medical advice - many featuring ostensibly real doctors. In the campaign against Thomas Massie, ads were run using deepfake AI images of him engaged in a "throuple" with democratic congresswomen. What the hell? Even before Trump took office the 2nd time, I felt like the government had mostly abdicated its regulatory duties, but it feels like the criminals are literally in charge now. The society-wide backlash to this shit can't come soon enough.
First regulation: ban all AI generated videos, music, movies and text that is not clearly and blatantly labeled as AI and enforce it EFFECTIVELY, not with virtual slaps on the wrist.
Second regulation: Ban all bots from the internet and punish ANYONE who allows them on their platform (i.e., Meta, X) significantly enough to hurt their business.
Third regulation: Prohibit Chatbots like ChatGPT from mimicking and/or impersonating human beings and enforce it with adequate consequences to the company responsible.
Agree. Noting that "enforce it effectively" has to mean CEOs in leg irons. Not just the minions, the multi-multi-millionaires, OK, billionaires at the top.
Agreed - all of those things should be obvious and in no way would impede technological progress - unless one’s idea of progress is enabling massive grift and fraud.
55% of Americans already say AI will do more harm than good, up from 44% a year ago. The backlash isn't building. It's already here. The question is whether it finds an institutional channel before the lobbying money finishes closing them off.
Good for the youngsters. They need to take a stand as they sense their replacements are on deck. What a deal they got. Go at least $100k into debt and graduate to a job market run by AI.
I like the analogy but wouldn’t call the motivating factor arrogance in either case. The U.S. refused to relent on Vietnam even after it was obviously unwinnable because the blowback that would follow giving up was more painful than just limping along with the “manageable” status quo (same thing with Afghanistan).
Similarly, if Silicon Valley just accepts the inevitable and throws in the towel, gives up its delusion of an AI future, the stock market will collapse and the billions they’re raking in from hypergrowth promises will evaporate. In their heads there’s no reason to let that happen now rather than later.
Which is why we need people like Gary Marcus to keep telling the public where these sociopaths are trying to take us NOW and every day BEFORE it crashes. NO MORE TAXPAYER BAILOUTS for the self serving fat cats who cause crashes!!!! We expect real accountability for them and the idiot banks who keep lending them money.
Welcome to "The Butlerian Jihad" [Dune - Frank Herbert]. Humanity at all levels must interject their control and requirements into the AI sphere before it progresses further. Giving any AI system complete autonomy at this point is beyond irresponsible. Highly focused, specialized domains of provably accurate data are the best applications for today's systems. Generic AI is an illusion and self-deception today.
I share Marcus' antipathy to generative AI and LLMs and agree that the current
methods will not get us to anything resembling GAI. I also agree that these
things are dangerous, owing to the lies coming from AI elites aiming to persuade
normal people that these agents are reliable and useful. However, I do not agree
with his solution.
Marcus' call for government regulation rests on the premise that in fact there
exist a set of 'disinterested' government workers (bureaucrats, if you like) who
can reliably administer such a scheme. The history of all federal three-letter regulatory agencies is that they all either work on behalf of the industry they are supposed to be regulating, or for
the political party in power. The members of these regulatory committees are scarcely more trustworthy than the crop of miscreants running the AI industrial complex. Elites will always find a way to manipulate regulators and legislators according to their will.
In my opinion, the solution (albeit, a slow one) is to expose these wicked and
Also, since when did we forget that "we the people" are ALL involved in common developmental movements (physical/emotional/and/mental/ intellectual/moral etc., etc., etc.), none of which is guaranteed to occur, but where education helps; and worse or better (whichever works for you) also, we are agents (so to speak) who can refuse one's own more natural developmental movements, e.g., I can be an alcoholic or a smoker or overeat, even though I know it's not good for my health--in spades when it comes to making conscious choices to do what we know are bad things; and so what concerns our experiences and moral/ethical development, not to mention, ask any poet about the psychological/psychic messes we get ourselves in.
If I have it right here, NORMATIVE is not, but rests in, what is NATURAL to us; but is also what groups of people come to accept as "given" in one's experiences and environment--it's what is concerned "what works best." Then one writes a Constitution.
This is why we are all so gob smacked about Trump et al breaking bad with what we consider is normative about, for instance, our Constitution here in the United States, and the reduction of everything to either the absurd and/or the transactional/corrupt drawing from the worst among us.
Michael Skinner: While your analysis of what is "real" politic of our present situation has truth to it, such an analysis is not necessarily (and perhaps should not be) the major source of a new gameplan. If I read you rightly, THAT is what you seem to be saying.
I think you are right about "exposing wicked agents to the light." However, bureaucrats who are habitualized to follow the well-thought-out rules and laws, and who are concretely accountable to them, are not a bad way of systematizing the good, along with a constantly upgrading system with appropriately civilized rewards and benefits.
In a deeper vein, "normal", though it might have invariant grounds, is still a moving target. A case in point is that we seem to be moving towards normalizing Trump and his family's taking a position to be ABOVE THE LAW, breaking a long-held American/democratic and, I think, rightly normalized protocol that I learned in grade school, but that again, sometimes is not evident in real-politic situations. Still, we must reach for it.
Also, I didn't know I couldn't stop "using them" (AI) until i woke up one day and, voila!, AI had become ubiquitous. If power-brokers like Altman, Gates, Musk, Trump, and Zuckerberg are the alternative . . . . I'll take a good bureaucrat in a working democratically run government institution any day.
We notice a regulation when it makes our life difficult, but not when we benefit from it, which is more often the case. Regulations contributed to the rise in the standard of of living and life expectancy. Many working in tech complained that "standards" constrained innovation until the internet and web came along, when it became clear standards could unleash innovation.
Agreed, and there are 2 sides to this that come into play.
First, creation of the regulations is extremely hard to do right, even when done in good faith. And unfortunately, the entire regulatory apparatus is almost literally bought and paid for. Politicians and agency heads are almost universally beholden to large corporate interests - including the AI vendors.
Second, regulation will achieve nothing without effective, even-handed enforcement. And since all government is inherently political, relying on government enforcement means enforcement will (not may, but will) become political enforcement, not policy enforcement.
Some good news is that the enforcement side has a proven potential fix: A right of private action. If the regulations are built with this as an integral part, and the regulation survives the test cases that go through the courts, then enforcement has some chance of being at least slightly balanced by private lawsuits.
So, for regulation to actually solve any problems (rather than just being political theater so politicians can claim they did something), we'd need three things:
1. People to speak up strongly enough to overcome the influence of money on policy making. As Gary said, this is hard but not impossible. And in fact, the advent of AI may actually help here, since it enables far more people to dissect and call out byzantine and evasive legislative language before it's enacted.
2. A right of private action as a core part of the legislation, to ensure enforcement is balanced and has enough teeth that it costs companies less to comply with the regulations than to violate them.
3. Courts ruling that the legislation and enforcement provisions are legal (instead of legislating from the bench). I suspect this is a toss-up. But more important, no matter when regulations are enacted, and no matter what the regulations are, this will take years.
If we strike out on any one of those three, any regulation will make things worse, not better.
Each of those individually is a tall order. All of them together is...let's just say pretty unlikely. The AI backlash will probably have to be quite disruptive to overcome those odds.
Which is why I'm generally skeptical of regulation. Not because appropriate regulation wouldn't be effective, but because appropriate regulation is so very rare.
And because just "doing something" when it's not the right thing invariably makes things worse.
If we do find a solution, I wouldn't be too surprised if it comes from something more like the progressive era labor movement than from top-down regulation.
John Michael Thomas: All good. But the laws and regulations are there to (1) give us a tried-and-true, but always general, framework to work into the details of living with; and (2) to reign in those who have not "made the good their own," or, regardless of fault and sometimes failure, and inevitable variables of viewpoints, at least are both identified with and easily recognize the good and good acts when we see them.
Our jails, especially in a democracy where the above tension lives throughout the land, are full of those who have not migrated a sense of self-regulation into one's interior life (Aristotle here also). So, the laws are needed, and need to have "teeth," for those who are hateful or careless about their own development and about their own well-being, and that of others. " . . . more like the progressive era labor movement than from top-down regulation."
Students in commencement exercises have in their minds, "What am I going to do now?" The main AI pundits running the big GenAI model companies keep telling them how they (the students) will soon be obsolete. There will soon be no need for entry-level workers, they say. So why would they not boo the people who are threatening their livelihood?
Similarly Amodei claims his model is too dangerous to release while fighting the government over whether his model is a supply chain risk. Have these people no sense?
Back to Zitron, the big players keep spending money (or at least claiming that they do) to build out infrastructure that will not materially improve AI and cannot improve their bottom lines. When future revenue is measured in billion$ and infrastructure (not even operating budgets) expenditures are measured in trillion$, how is that likely to end? What will happen to these booing students when the AI market comes crashing down? How much of that burden will they have to pick up? Why wouldn't they boo?
So the students vibe code a new app with an AI engine, and hope to sell it by reselling tokens from a GenAI model. What happens to these students when the AI providers raise their prices or go out of business? Why shouldn't these students boo?
The AI "mavens" lie through their teeth to these students. Why wouldn't they boo?
What do the students get? A tool to do their homework and a girlfriend/boyfriend who is a model. That, apparently, is not enough to keep them from booing.
If I were giving a commencement speech, I would approach it differently. I would emphasize how AI will help them toward greatness. If I could not say that with a straight face, then I have no business giving a commencement speech.
I think that even the current GenAI approach has potential value. I think that other approaches will have better value. I think that they will open new avenues for jobs and support opportunities that have not yet become apparent.
Language models may not be (are not) enough to deliver on this promise, but I do believe that the future of AI will extend beyond today's technology. As Alan Kay (and others) has said, the best way to predict the future is to invent it. I'm doing my best to do that inventing to empower, not replace human workers.
Vietnam War spending came from taxpayers’ money, while AI infrastructure is primarily funded through corporate investment. These are two very different things and should not be confused.
While the Vietnam War was ultimately a disaster for the United States, strategically it was part of a broader policy of resisting the expansion of communist authoritarian regimes. The U.S. took on that role globally, and many people who lived under communist systems remain grateful for it. If you lived in South Korea, you would understand exactly what I mean. Taiwan and South Korea are free and prosperous today, and indirectly they are an important part of the AI story because of the sacrifices and geopolitical realities of the Vietnam War era.
AI in its current form is not AGI, but it is already a very useful tool and will likely continue to improve. As for the economics of it, many highly capable people at major U.S. companies have concluded that AI investment makes economic sense. Time will show whether Gary Marcus or those companies are ultimately correct.
In any case, the datacenters being built today provide compute capacity that can remain useful as AI algorithms become more efficient and reliable.
The speed and energy efficiency of modern semiconductors are the foundation of everything we can currently do in AI, and potentially AGI in the future. The massive amounts of money being spent by corporations are being funneled into R&D that creates faster and better microprocessors while also improving semiconductor manufacturing processes. Quantum computing appears to be decades away from practical usefulness, if it becomes broadly useful at all.
Another point Gary Marcus often makes is that there may eventually be some undiscovered “magic bullet” — such as neurosymbolic AI — that could make LLMs obsolete and render today’s datacenter investments unnecessary. However, very smart people have spent decades trying to create a generalized “scheme for thinking,” and the problem still appears unsolved. What we are seeing instead is the development of scaffolding systems by companies like Anthropic. In many ways, this resembles how humans use tools such as calculators, lookup tables, and external memory aids. This direction of development does not make datacenters obsolete.
I share the skepticism about AGI and I agree we're burning a lot of money and worse, completely forgetting about our supposedly common climate goals.
However, postulating that there's no benefit at all makes me wonder if anyone here recently tried to do some real coding or solve a technical issue. Sure, the AI output does not replace an expert, but it replaces the typical 2-5 people dev team that a senior developer or technician will typically have to rely on to get anything done. There is a huge number of junior-level roles in product development, operations, documentation, marketing, which are potentially going to be covered by a premium tier AI subscription in the future.
Of course the results are not perfect, but remember the 80/20 rule. If quality is important, you need to include seniors and iterate over results. Its not like we had perfect SW quality and documentation before.
The real question for me is, 10 years down, where to hire those senior experts...
Nobody is saying there is no benefit at all, especially not Gary Marcus!!! I am so tired of this argument based on nothing but hyperbolic, disingenuous LIES.
This is a bit simplistic. Many women were opposed to the Vietnam War, and they were not at risk of being drafted. The fiscal cost of rhe war wasn't entirely evident. Lyndon Johnson was credited with balancing a couple of his budgets, a feat that only Bill Clinton has accomplished since.
That may have been true of some, but many, and I was one of them, were in the streets because we recognized immoral when we saw it, and we were willing to be vocal in saying we didn't like it. That kind of vision seemed to have been on the wane since then, until quite recently. I see a resurgence of people of conscience, or a variety of issues. If there are amongst them, some who are opposing it for their own gain, they do nothing but add a few additional numbers to the right side.
Yes, i specifically referenced "hideous" to include the immorality, though those of you who opposed primarily on conscience were likely the vast minority. Our youth have not protested en masse to the ongoing slaughters in the world this decade, some of which we've financed or taken direct involvement.
The students are not confused. They are looking at a society willing to spend effectively unlimited sums of money, energy, land, water & political will to build data centers so machines can produce simulations of human-produced works of art, while that same society tells them affordable housing, stable work, and a decent life are somehow unrealistic demands.
This is why Silicon Valley and its broader ethos is the least innovative social force in history.
I was thinking about this earlier. It’s also such a rug pull on the social contract. America was systematically deindustrialized with the idea that people would move into service jobs higher up on the value steam and with lower goods from abroad, the quality of life would improve. That sold 2 generations on the college to white collar pipeline especially in technology and finance. Now basically we’re saying all of those should be automated. What paths are left for the masses to get to the middle class or what left of it then
Agree. Capitalism has to be constrained by social democracy which in the U.S. is being undermined by campaign finance laws and income/wealth inequality. Students better vote and get more active (which I think we are seeing some).
While also lecturing them about having more babies, earlier. Because nothing encourages people to have lots of children like housing and career insecurity!
The tech bros were told when they started hyping the "gig economy" that people were not going to settle down if all they could get were gig economy jobs. They were informed that no secure salary and no benefits was not a way to run a railroad.....but nooooo.....And so, along came AI and the tech bros again touted how people would face MORE economic insecurity! The rejection of the tech bros and their glassy eyed pronouncements are finally getting the hard pushback that should have been given to the gig economy idea. Let's hope it does indeed slow the AI hype train and slows the roll of those who think they can replace people with AI.
Tech elites live in their own universe, the one between their ears. All else is an unnecessary distraction.
Certainly seems to be plenty of empty space up there.
While at the same time demanding that the Fed step in and print money to rescue markets whenever their investments are in trouble.
And they love to find ways to eliminate all levels of workers, except for themselves, with technology. They call it innovation, but is it really innovation when your whole modus opporendi seems to be to eliminate any obstacles between themselves and All the Money.
This mindset happens when old saws "he who has the most toys wins" and "greed is good" are your daily affirmations.
The B52’s even wrote them a song “You’re livin’ in your own private AI-daho”
> This is why Silicon Valley and its broader ethos is the least innovative social force in history.
That line is so punchy, I just love it. And also true.
The Vietnam analogy is more structurally precise than Calacanis intended. Vietnam's specific failure mode was a metrics problem. The US measured progress with body counts and territory captured while the actual situation deteriorated underneath. The numbers told a story of winning while the ground truth was losing.
AI is running the same divergence right now. Benchmarks keep climbing. Leaderboard scores improve quarterly. But enterprise pilot ROI keeps failing to materialise and the political coalition that funded the buildout is fracturing in real time. The industry is measuring capability while the market needs reliability, and those two metrics diverged about two years ago. Thats the gap where backlashes form.
This comment is more perceptibly AI-generated than you intended.
Barbara Tuchman's book, "The March of Folly," might be informative on this issue. This is how she defines it:
From Tuchman:
To qualify as folly for this inquiry, the policy adopted must meet three criteria: it must have been perceived as counter-productive in its own time, not merely by hindsight. This is important, because all policy is determined by the mores of its age. ‘Nothing is more unfair, as English historian has well said, “than to judge men of the past by the ideas of the present.”
Secondly, a feasible alternative course of actions must have been available.
Thirdly, to remove the problem of personality, a criteria must be that the policy in question should be that of a group, not an individual ruler, and should persist beyond one political lifetime (or period). [In this third criteria Tuchman suggests that some succession of support for the policy must be in place.]
I think all of these conditions exist in the current AI frenzy. I think a key question is that with so much momentum behind it, how/when will the crash happen and what will be the damage to whom? Then the pieces will be picked up and lead to what?
Bravo! Quoting Tuchman in a discussion of AI is a good pull!
My first introduction to her was the Amazon Audible version of The Proud Tower. The actress who read it was marvelous.
To answer your final question, Marcus would say some sort of world models will need to be a part of the systems. I think that is probably true. I believe that though pure LLM scaling will "crash", AI will not. As with various tech dead ends, the world will see something better and migrate to it.
I like the word hubris
Yes, hubris is the word that Jim Collins uses in his book, "How the Mighty Fall." Tuchman also has a great definition for "wood-headedness" that is worth reading the book for. It is hilarious!
Tuchman was brilliant
I wrote about her ideas in this post on my Substack:
https://gerald968.substack.com/p/folly-how-bad-decisions-can-lead?r=lms9u&utm_campaign=post-expanded-share&utm_medium=web
Aside from Cronkite's impact, the American public only turned against Vietnam once white suburban kids started returning in body bags.
Maybe "diminished job prospects" will be the new body bags...
If we ever have adults back in charge of the country, there should be a tremendous reckoning for the outright fraud being perpetrated by many AI creators and enablers. For the life of me, I can't understand how companies aren't being sued into oblivion for creating obviously fraudulent AI-generated ads. Nearly all of the ads I am served on YouTube are AI slop giving quack medical advice - many featuring ostensibly real doctors. In the campaign against Thomas Massie, ads were run using deepfake AI images of him engaged in a "throuple" with democratic congresswomen. What the hell? Even before Trump took office the 2nd time, I felt like the government had mostly abdicated its regulatory duties, but it feels like the criminals are literally in charge now. The society-wide backlash to this shit can't come soon enough.
First regulation: ban all AI generated videos, music, movies and text that is not clearly and blatantly labeled as AI and enforce it EFFECTIVELY, not with virtual slaps on the wrist.
Second regulation: Ban all bots from the internet and punish ANYONE who allows them on their platform (i.e., Meta, X) significantly enough to hurt their business.
Third regulation: Prohibit Chatbots like ChatGPT from mimicking and/or impersonating human beings and enforce it with adequate consequences to the company responsible.
> enforce it EFFECTIVELY
Agree. Noting that "enforce it effectively" has to mean CEOs in leg irons. Not just the minions, the multi-multi-millionaires, OK, billionaires at the top.
Agree 100%. The Supreme Court ruled that corporations are people, so let the decision makers go to jail for their crimes.
Agreed - all of those things should be obvious and in no way would impede technological progress - unless one’s idea of progress is enabling massive grift and fraud.
55% of Americans already say AI will do more harm than good, up from 44% a year ago. The backlash isn't building. It's already here. The question is whether it finds an institutional channel before the lobbying money finishes closing them off.
Good for the youngsters. They need to take a stand as they sense their replacements are on deck. What a deal they got. Go at least $100k into debt and graduate to a job market run by AI.
But what are they learning in the school?
I like the analogy but wouldn’t call the motivating factor arrogance in either case. The U.S. refused to relent on Vietnam even after it was obviously unwinnable because the blowback that would follow giving up was more painful than just limping along with the “manageable” status quo (same thing with Afghanistan).
Similarly, if Silicon Valley just accepts the inevitable and throws in the towel, gives up its delusion of an AI future, the stock market will collapse and the billions they’re raking in from hypergrowth promises will evaporate. In their heads there’s no reason to let that happen now rather than later.
Which is why we need people like Gary Marcus to keep telling the public where these sociopaths are trying to take us NOW and every day BEFORE it crashes. NO MORE TAXPAYER BAILOUTS for the self serving fat cats who cause crashes!!!! We expect real accountability for them and the idiot banks who keep lending them money.
I dont see Trump pivoting on AI. Too many of the extraction class that give him $ are gung ho about it and not interested in any guidelines.
Agree. Any pushback from Trump will be symbolic at most.
Yep, just like he keeps claiming he’s done something for infertility patients.
And anything about dominating China. Or Iran.
You called that one!
Welcome to "The Butlerian Jihad" [Dune - Frank Herbert]. Humanity at all levels must interject their control and requirements into the AI sphere before it progresses further. Giving any AI system complete autonomy at this point is beyond irresponsible. Highly focused, specialized domains of provably accurate data are the best applications for today's systems. Generic AI is an illusion and self-deception today.
I share Marcus' antipathy to generative AI and LLMs and agree that the current
methods will not get us to anything resembling GAI. I also agree that these
things are dangerous, owing to the lies coming from AI elites aiming to persuade
normal people that these agents are reliable and useful. However, I do not agree
with his solution.
Marcus' call for government regulation rests on the premise that in fact there
exist a set of 'disinterested' government workers (bureaucrats, if you like) who
can reliably administer such a scheme. The history of all federal three-letter regulatory agencies is that they all either work on behalf of the industry they are supposed to be regulating, or for
the political party in power. The members of these regulatory committees are scarcely more trustworthy than the crop of miscreants running the AI industrial complex. Elites will always find a way to manipulate regulators and legislators according to their will.
In my opinion, the solution (albeit, a slow one) is to expose these wicked and
dangerous agents to the disinfectant of light.
Normal people will just have to stop using them.
Normal people? Like the people who rejected masks (and helped cause over a million deaths)? Or the people who elected Trump for a second time?
Marc Schluper: Exactly that.
Also, since when did we forget that "we the people" are ALL involved in common developmental movements (physical/emotional/and/mental/ intellectual/moral etc., etc., etc.), none of which is guaranteed to occur, but where education helps; and worse or better (whichever works for you) also, we are agents (so to speak) who can refuse one's own more natural developmental movements, e.g., I can be an alcoholic or a smoker or overeat, even though I know it's not good for my health--in spades when it comes to making conscious choices to do what we know are bad things; and so what concerns our experiences and moral/ethical development, not to mention, ask any poet about the psychological/psychic messes we get ourselves in.
If I have it right here, NORMATIVE is not, but rests in, what is NATURAL to us; but is also what groups of people come to accept as "given" in one's experiences and environment--it's what is concerned "what works best." Then one writes a Constitution.
This is why we are all so gob smacked about Trump et al breaking bad with what we consider is normative about, for instance, our Constitution here in the United States, and the reduction of everything to either the absurd and/or the transactional/corrupt drawing from the worst among us.
Michael Skinner: While your analysis of what is "real" politic of our present situation has truth to it, such an analysis is not necessarily (and perhaps should not be) the major source of a new gameplan. If I read you rightly, THAT is what you seem to be saying.
I think you are right about "exposing wicked agents to the light." However, bureaucrats who are habitualized to follow the well-thought-out rules and laws, and who are concretely accountable to them, are not a bad way of systematizing the good, along with a constantly upgrading system with appropriately civilized rewards and benefits.
In a deeper vein, "normal", though it might have invariant grounds, is still a moving target. A case in point is that we seem to be moving towards normalizing Trump and his family's taking a position to be ABOVE THE LAW, breaking a long-held American/democratic and, I think, rightly normalized protocol that I learned in grade school, but that again, sometimes is not evident in real-politic situations. Still, we must reach for it.
Also, I didn't know I couldn't stop "using them" (AI) until i woke up one day and, voila!, AI had become ubiquitous. If power-brokers like Altman, Gates, Musk, Trump, and Zuckerberg are the alternative . . . . I'll take a good bureaucrat in a working democratically run government institution any day.
We notice a regulation when it makes our life difficult, but not when we benefit from it, which is more often the case. Regulations contributed to the rise in the standard of of living and life expectancy. Many working in tech complained that "standards" constrained innovation until the internet and web came along, when it became clear standards could unleash innovation.
Agreed, and there are 2 sides to this that come into play.
First, creation of the regulations is extremely hard to do right, even when done in good faith. And unfortunately, the entire regulatory apparatus is almost literally bought and paid for. Politicians and agency heads are almost universally beholden to large corporate interests - including the AI vendors.
Second, regulation will achieve nothing without effective, even-handed enforcement. And since all government is inherently political, relying on government enforcement means enforcement will (not may, but will) become political enforcement, not policy enforcement.
Some good news is that the enforcement side has a proven potential fix: A right of private action. If the regulations are built with this as an integral part, and the regulation survives the test cases that go through the courts, then enforcement has some chance of being at least slightly balanced by private lawsuits.
So, for regulation to actually solve any problems (rather than just being political theater so politicians can claim they did something), we'd need three things:
1. People to speak up strongly enough to overcome the influence of money on policy making. As Gary said, this is hard but not impossible. And in fact, the advent of AI may actually help here, since it enables far more people to dissect and call out byzantine and evasive legislative language before it's enacted.
2. A right of private action as a core part of the legislation, to ensure enforcement is balanced and has enough teeth that it costs companies less to comply with the regulations than to violate them.
3. Courts ruling that the legislation and enforcement provisions are legal (instead of legislating from the bench). I suspect this is a toss-up. But more important, no matter when regulations are enacted, and no matter what the regulations are, this will take years.
If we strike out on any one of those three, any regulation will make things worse, not better.
Each of those individually is a tall order. All of them together is...let's just say pretty unlikely. The AI backlash will probably have to be quite disruptive to overcome those odds.
Which is why I'm generally skeptical of regulation. Not because appropriate regulation wouldn't be effective, but because appropriate regulation is so very rare.
And because just "doing something" when it's not the right thing invariably makes things worse.
If we do find a solution, I wouldn't be too surprised if it comes from something more like the progressive era labor movement than from top-down regulation.
John Michael Thomas: All good. But the laws and regulations are there to (1) give us a tried-and-true, but always general, framework to work into the details of living with; and (2) to reign in those who have not "made the good their own," or, regardless of fault and sometimes failure, and inevitable variables of viewpoints, at least are both identified with and easily recognize the good and good acts when we see them.
Our jails, especially in a democracy where the above tension lives throughout the land, are full of those who have not migrated a sense of self-regulation into one's interior life (Aristotle here also). So, the laws are needed, and need to have "teeth," for those who are hateful or careless about their own development and about their own well-being, and that of others. " . . . more like the progressive era labor movement than from top-down regulation."
Ed Zitron has (among many such articles) a great article about the future of AI. https://www.wheresyoured.at/ai-is-too-expensive/
He has the details.
Students in commencement exercises have in their minds, "What am I going to do now?" The main AI pundits running the big GenAI model companies keep telling them how they (the students) will soon be obsolete. There will soon be no need for entry-level workers, they say. So why would they not boo the people who are threatening their livelihood?
Similarly Amodei claims his model is too dangerous to release while fighting the government over whether his model is a supply chain risk. Have these people no sense?
Back to Zitron, the big players keep spending money (or at least claiming that they do) to build out infrastructure that will not materially improve AI and cannot improve their bottom lines. When future revenue is measured in billion$ and infrastructure (not even operating budgets) expenditures are measured in trillion$, how is that likely to end? What will happen to these booing students when the AI market comes crashing down? How much of that burden will they have to pick up? Why wouldn't they boo?
So the students vibe code a new app with an AI engine, and hope to sell it by reselling tokens from a GenAI model. What happens to these students when the AI providers raise their prices or go out of business? Why shouldn't these students boo?
The AI "mavens" lie through their teeth to these students. Why wouldn't they boo?
What do the students get? A tool to do their homework and a girlfriend/boyfriend who is a model. That, apparently, is not enough to keep them from booing.
If I were giving a commencement speech, I would approach it differently. I would emphasize how AI will help them toward greatness. If I could not say that with a straight face, then I have no business giving a commencement speech.
I think that even the current GenAI approach has potential value. I think that other approaches will have better value. I think that they will open new avenues for jobs and support opportunities that have not yet become apparent.
Language models may not be (are not) enough to deliver on this promise, but I do believe that the future of AI will extend beyond today's technology. As Alan Kay (and others) has said, the best way to predict the future is to invent it. I'm doing my best to do that inventing to empower, not replace human workers.
An agency to approve the release of new technology. Kind of like the FDA, right? What could go wrong!
Another worm brain.
Vietnam War spending came from taxpayers’ money, while AI infrastructure is primarily funded through corporate investment. These are two very different things and should not be confused.
While the Vietnam War was ultimately a disaster for the United States, strategically it was part of a broader policy of resisting the expansion of communist authoritarian regimes. The U.S. took on that role globally, and many people who lived under communist systems remain grateful for it. If you lived in South Korea, you would understand exactly what I mean. Taiwan and South Korea are free and prosperous today, and indirectly they are an important part of the AI story because of the sacrifices and geopolitical realities of the Vietnam War era.
AI in its current form is not AGI, but it is already a very useful tool and will likely continue to improve. As for the economics of it, many highly capable people at major U.S. companies have concluded that AI investment makes economic sense. Time will show whether Gary Marcus or those companies are ultimately correct.
In any case, the datacenters being built today provide compute capacity that can remain useful as AI algorithms become more efficient and reliable.
The speed and energy efficiency of modern semiconductors are the foundation of everything we can currently do in AI, and potentially AGI in the future. The massive amounts of money being spent by corporations are being funneled into R&D that creates faster and better microprocessors while also improving semiconductor manufacturing processes. Quantum computing appears to be decades away from practical usefulness, if it becomes broadly useful at all.
Another point Gary Marcus often makes is that there may eventually be some undiscovered “magic bullet” — such as neurosymbolic AI — that could make LLMs obsolete and render today’s datacenter investments unnecessary. However, very smart people have spent decades trying to create a generalized “scheme for thinking,” and the problem still appears unsolved. What we are seeing instead is the development of scaffolding systems by companies like Anthropic. In many ways, this resembles how humans use tools such as calculators, lookup tables, and external memory aids. This direction of development does not make datacenters obsolete.
I share the skepticism about AGI and I agree we're burning a lot of money and worse, completely forgetting about our supposedly common climate goals.
However, postulating that there's no benefit at all makes me wonder if anyone here recently tried to do some real coding or solve a technical issue. Sure, the AI output does not replace an expert, but it replaces the typical 2-5 people dev team that a senior developer or technician will typically have to rely on to get anything done. There is a huge number of junior-level roles in product development, operations, documentation, marketing, which are potentially going to be covered by a premium tier AI subscription in the future.
Of course the results are not perfect, but remember the 80/20 rule. If quality is important, you need to include seniors and iterate over results. Its not like we had perfect SW quality and documentation before.
The real question for me is, 10 years down, where to hire those senior experts...
Nobody is saying there is no benefit at all, especially not Gary Marcus!!! I am so tired of this argument based on nothing but hyperbolic, disingenuous LIES.
And when the subscription price goes up (FYPM in action), all those companies will be over a barrel. Oops!
The students didn't care about the massive waste of money, they cared about the draft dragging them to a hideous jungle war.
This is a bit simplistic. Many women were opposed to the Vietnam War, and they were not at risk of being drafted. The fiscal cost of rhe war wasn't entirely evident. Lyndon Johnson was credited with balancing a couple of his budgets, a feat that only Bill Clinton has accomplished since.
That may have been true of some, but many, and I was one of them, were in the streets because we recognized immoral when we saw it, and we were willing to be vocal in saying we didn't like it. That kind of vision seemed to have been on the wane since then, until quite recently. I see a resurgence of people of conscience, or a variety of issues. If there are amongst them, some who are opposing it for their own gain, they do nothing but add a few additional numbers to the right side.
Yes, i specifically referenced "hideous" to include the immorality, though those of you who opposed primarily on conscience were likely the vast minority. Our youth have not protested en masse to the ongoing slaughters in the world this decade, some of which we've financed or taken direct involvement.