The sooner people can be broken out of their denial about all this the better, and we can start actually taking it seriously.
That's not what people mean when they say "moving the goalposts". It means that people are adamant that something wasn't important/hard/impressive once the "AI" solves it. And then they come up with another thing that needs to be solved in order to prove it is important/hard/impressive. And once that happens, they do it again. And again. That's what "moving the goalposts" means.
It's also very much not a new phenomenon. It's been happening since the 1980s. As you can see from this quote from GEB by Hofstadter:
> There is a related "Theorem" about progress in AI: once some mental function is programmed, people soon cease to consider it as an essential ingredient of "real thinking". The ineluctable core of intelligence is always in that next thing which hasn't yet been programmed. This "Theorem" was first proposed to me by Larry Tesler, so I call it Tesler's Theorem: "AI is whatever hasn't been done yet."
I don't see how that's any better.
Actually, I'm really an LLM and this whole thing was a clever bluff.
In recent years, I have commonly seen the phrase "you're moving the goalposts" deployed by the "it might be sentient" crowd to shoot down the "it's a stochastic parrot" crowd when the latter respond to a new development with "OK but...". In a well-understood field of inquiry, that would be a clear case of goalpost-moving, in the commonly-understood meaning of the phrase where requirements are retroactively changed in response to them having been met. Thank you OP. 'Artificial Intelligence', and indeed intelligence in general, is very much not a well-understood field of inquiry - in fact we don't even have a common agreement about what 'intelligence' is. We are therefore learning as we go (even after all this time!) but making rapid progress in recent years. When rapid progress is made in a poorly-understood field, then how can our definitions and requirements for success not change? This is arguably one of the most pathological development projects ever - what are the requirements? 'It thinks like a human'? What does that mean? And the answer is we don't know what that means, and we're working it out as we go - moving the goalposts. If we didn't move the goalposts, then by definition we already knew exactly where we were headed at the beginning, and we very clearly did not.
Side note that, in case it's not obvious, none of this detracts from how impressive LLMs are. They're a marvel of the modern age, all the problems notwithstanding. However I reserve the right to stay sceptical about their capabilities.
To me, the goalposts were already defined by the person you were responding to. "The impact of AI is getting undeniable", so, the goalposts are "the impact of AI". Probably something like "the impact of AI is high, or will be soon".
Note that this does not depend on things like AI sentience or defining "intelligence" more rigorously, it just depends on AI impact.
> "The impact of AI is getting undeniable", so, the goalposts are "the impact of AI". Probably something like "the impact of AI is high, or will be soon".
those are not goalposts - that's an incredible vague 'goal'. What are we moving here exactly?
Thus "moving the goalposts" is for the defenders to shift what argument that the other side is trying to disprove.
If the argument is "the impact of AI is getting undeniable", you'd try to show that impact of AI is not that high, and not looking to be that high in the near future.
The other side might move the goalposts by saying something like "well, it doesn't matter if AI is actually that broadly impactful right now, all that matters is whether it's impactful in the field of mathematics". At which point, you might be frustrated because the effort you put into engaging their previous argument is now moot.
PS - yeah, yeah, I overexplained
> If we didn't move the goalposts, then by definition we already knew exactly where we were headed at the beginning, and we very clearly did not.
The criticisms are directed toward people who did clearly act like they knew, not the ones who were honest that they did not know.
It's in how they change, not the fact that they change. The skeptics seem to have secret definitions for intelligence, sentience, consciousness, creativity, etc. that amounts to "a thing only humans have". Often that thing is equivalent to a soul. When yesterday's challenge (LLMs don't have X because they can't do Y!) is met, Y changes but X stays the same. This is not the process by which a field matures, it is a rhetorical technique used by skeptics to avoid honestly stating or confronting their internal definitions. That can be revealed by asking the skeptic the following:
"Forget LLMs. What if we made a completely physically accurate simulation of a human being?"
Many say no, that simulated human being still couldn't have (intelligence, consciousness, sentience, creativity, ...). This reveals that there is a necessary metaphysical component to those attributes, at which point any scientific-minded person will leave the debate.
Totally! Arguments like yours often get lost in the noise of the less interesting / nuanced arguments unfortunately :(
The motte is "AI useful". The bailey is "Singularity is nigh".
But there are people like Ed Zitron, frequently posted and cited here, who disagree even with the former.
"it isn't clear whether generative AI actually provides much business value at all"
"cannot seem to find a product that people will pay for, in part because the results are so mediocre"
"Last week, we got our first real, definitive glimpse of what’s around that corner that future. And boy, was it underwhelming."
"OpenAI claims that o1 “performs similarly to PhD students on challenging benchmark tasks in physics, chemistry, and biology.” Just not in geography, it seems. Or basic elementary-level English language tests. Or math. Or programming. "
"Worse still, it's kind of hard to explain why anybody should give a shit about o1."
"o1 shows that OpenAI is both desperate and out of ideas."
"the software is not becoming more useful"
Honestly, every other line is quotable in this context.
But it seems we have somehow optimized away shame. It wasn’t good for profits, I guess.
Personally I prefer to follow explorers rather than swamp-sitters.
(I'm personally still skeptical about this, but I'm being pulled towards accepting it).
"AI is useful" is too low of a bar, and "singularity is nigh" is too high. "AI is on its way to upending society" is about in the middle, and still vastly contentious among laypeople.
-Alan Turing (allegedly)
The fundamental argument that I've personally made since the early days of this is that LLMs are not reasoning, in the way that word is commonly understood.
There are lots of reasons why that argument needs to evolve that could certainly appear to be "moving the goalposts", but let's take an example.
A lot of AIs were tripped up by the question "Should I walk or drive 50m to the carwash?" Several folks liked to use that as an example that illustrates that LLMs aren't reasoning, but as the models have been trained on that specific example, it's of course less useful. An AI can mostly nail it now.
So a different example is needed. A new demonstration of how these things fail at basic reasoning a child can do.
Did I move the goalposts? I don't think so. The fundamental argument stays the same. It's not hard to find lots of examples that trip up LLMs, because they are what they are: statistical inference machines. Nothing more and nothing less.
Useful, sure. But also commonly misapplied to areas for which they are inappropriate solutions.
Not long ago many folks were saying AI was the same as the crypto bubble. No real useful technology and only hype.
I think AI is clearly both revolutionary and useful. Revolutionary insofar as the job I do has changed almost completely in a year or so span.
I agree with the parent that we need to acknowledge that we're at a turning point in history. I lived through some of them (internet, ubiquitous personal computing). But it's somewhat difficult to comprehend the impact of this one for many people.
I do biomedical research at one of the top European research institutions. We're very well-funded, but I can clearly see the gap between us (say, top-100) and top-10. I also realize this gap is going to get so much wider unless we invest heavily in AI access (and I'm not so sure I can sell anything more expensive than $20 Claude subscription to the leadership).
I think people having 6-7 figure SOTA AI budgets will move exponentially faster than those who don't. That makes me worried.
So, for me, it's not a question of recalibrating expectations. We're way past that.
There is irony here
I ,for one, have read enough history to know that it's never the proles who end up benefiting.
And nowadays with Hong Kong (and probably soon Taiwan) they are proving they are perfectly happy to destroy economic growth as long as it benefits The Party
If your source is AI layoffs, there were plenty of layoffs with the invention of the horseless carriage, but that doesn't mean it made humans worse off overall.
There are reasons to be skeptical about progress and AI and all that but the 'making things worse for most people' thing you mentioned seems yet to be based in any reality
Is the amount of people making money from AI greater than the amount of people who lose money from AI? Is maximizing prosperity across humanity even a goal, at this point? Am I to be expected to believe, without doubting, that the end goal is the benefit of the many?
But it's not clear if LLMs will produce abstractions that humans would find elegant.
https://garymarcus.substack.com/p/two-critical-updates-re-as...
As always, PR hype. Goalposts have not moved.
Guys, please use critical thinking. The haters don't hate by default, we hate because we're gaslit about this stuff every day and it's annoying. Extraordinary claims require proof, and they're not giving us information that would be essential to knowing if this is actually significant or not.
He literally says it's an impressive feat in the second article.
That is the intended purpose of hype.
He's polite, but he's basically saying there's potentially a lot of smoke and mirrors. I agree.
I remember the time when he insisted that diffusion-based image generators trained on Internet scale data will never be able to make an image of a horse riding an astronaut. Today you can generate 4K video of that.
A lot of his predictions have held up very well.
Remember October 2024 Pelicans [1] ? It's been only less than 2 years.
We don't know what will come in the next 2 years. But the progress doesn't seem to stop for now.
[1] https://simonwillison.net/2024/Oct/25/pelicans-on-a-bicycle/
People are skeptical of the announcement because the room include several PHDs in math and physics. The prompts are not published so we can see how generic the starting prompt is.
If I could take one out of context quote from this whole thread as a response to TFA, it would be this one.
I've been analyzing thousands of x86 traces and the AI made sense of them at a scale no previous human or algorithm could do before - I've been reading assembly everyday for half of my life. I'm doing prototypes in minutes what would take hours or days before. AI is answering questions in such a precise way in so many domains, learning something has never been easier.
This is "just ok" for you ? What do you need exactly ? god omniscience ? come on, can't we just appreciate for what it is and be excited about the future ?
The only way that is PR hype is if you're invoking the insane conspiracy that frontier AI labs are just buying off results that would otherwise be career defining for a mathematician, just for marketing.
The posts you linked are urging caution regarding the exaggerated e/acc-esque lies peddled by people like Musk, not that the models haven't proven themselves as having genuine ability to contribute to research in some areas.
They still do things that I find incredibly annoying and “dumb”. And I still have to clean up messes they make quite often.
But on the whole they are clearly smarter than before. No extraordinary claims needed. I just try to learn how the tool works and how to use it effectively.
i can link you likely dozens of comments from people wrong about this replying to me over the last 5 years
We need to figure out how to restructure the global economy. How does UBI work internationally, if the AI companies are taking revenue in the US? What’s the tax base for it? What does that say about international trade and protectionism? Do countries end up splitting into different trading blocks based on their level of access and legality of AI (I assume some will ban it outright)?.
How does intellectual property work in an AI generated future? What about healthcare advances, who gets to own those?
What about meaning, what about purpose? How do we replace the work ethic that tells us we are our jobs and idleness is immoral? How do you replace “What do you do?” As one of the first questions you ask a new person?
That sort of thing.
And it's sad, really, because I think these two groups would make a great pairing if they could stop arguing against one another for a moment. They'll both be impacted about as much and probably have the same ultimate goals (to lead dignified lives).
But it seems these days everyone is more interested in Kayfabe and feeling like they're in the right than working together, so maybe I should just keep quiet rather than attract the ire of both groups...
I don't know if it is fair to say they're in denial. For my part, I don't expect life to get much better for regular people (especially short term), but that doesn't mean we shouldn't work to try to make it happen.
What a lot of people want to do, and I’m not saying that you’re one of them, is to assume that a positive outcome is impossible and either do nothing or loudly yell that the world is ending. Neither is particularly useful.
Or, as I said above, others just deny that there’s anything to see here and try to get people to move along.
Shane Legg (DeepMind co-founder), one of the more intelligent and thoughtful people you'll find in the industry, could only offer "it's a tough problem - we need to think about it" when recently interviewed by Hannah Fry.
On the surface the most likely outcome for AI allowed to replace jobs is extraordinarily negative, especially since it is a general capability technology, not a specific one where displaced workers can just move to another field. Once AI becomes more capable it will be able to do the vast majority of white collar jobs, including any new ones that may appear as a result of AI. As Shane Legg put it, "if your job can be done remotely, sitting in front of a computer, then it can probably be replaced by AI".
Not only does AI threaten to replace ALL the white collar jobs, but it is rapidly going after blue collar (factory jobs, driving jobs) and pink collar ones (Japanese robotics for elder-care) as well.
If a positive outcome (which doesn't include putting displaced workers on welfare - UBI) is possible, then it sure would be nice to hear it, and the silence from the AI companies, and government for that matter, is deafening.
See "Machines of Loving Grace" by Dario Amodei: https://darioamodei.com/essay/machines-of-loving-grace.
"Massive Economic Abundance: Because AI will exponentially grow the total economic pie, overall resource scarcity will diminish. The fundamental challenge shifts from producing wealth to distributing wealth."
So how do we go from everyone out of work, no income to spend on food, or the goods and services that the AI is producing, to "massive economic abundance"?!
It's like the meme:
Step 1: Create AI
Step 2: AI takes all the jobs
Step 3: ???
Step 4: Profit! (massive economic abundance)
What is step 3?
"Massive Economic Abundance" implies massive increase in produced goods. This implies massive deflation, ceteris paribus. So step 3 could be simply printing money to pay for UBI. Deflation from AI productivity increase and inflation from UBI money printing will cancel out.
For money to work it has to represent some real value, something that has some scarcity to it such as potatoes or hours of human labor. Ultimately it is just a decoupler in a barter system, a universally recognized IOU.
Why would someone give me a car in exchange for UBI-scrip when that UBI-scrip has no inherent scarcity or value and can be produced in infinite supply by the government ?
UBI script will have some value, I am proposing printing enough money just to combat AI productivity-induced deflation, not infinite UBI money.
Eventually UBI will be the norm, and if the living standards of a person on UBI is as good as yours or mine today, that will be an enormous win for everyone. It’s like pensions, once these were only for the elderly poor, now they’re a right for everyone in most developed countries.
It’s also interesting that for most of human history leisure time was the point of life, and only in recent modernity has work come to be the meaning of someone’s existence.
UBI has to be commensurate with production being automated. That’s a big logistical problem, if you think building datacenters is a challenge try bringing about radical abundance, but even so it’s not insurmountable. It just needs to be taken on as project and not seen as an impossibility.
So much of this is not about what is possible so much as what people believe is possible. We can do anything if we try.
I see comments like this tossed around a lot, but what makes you say this? Don't you think its more likely that most people end up in poverty?
why do they spend the money they do on the things they do?
they are not altruists and they never will be.
they could change millions of lives today, but most do not.
pure naivete.
In any case historical trends are really irrelevant to the discussion, which is about the impact of AI, which threatens to take away almost ALL the jobs (this is what Dario's essay is assuming), which has no historical precedent. Some people like to bring up previous waves of job automation such as the industrial revolution, but the difference there is that automation took some jobs but created others. In the case of AI, AI will also be taking the new jobs that it creates.
If AI takes all the jobs - which is what the people like Dario who are creating it assume will happen - then it seems "UBI" is indeed the logical conclusion (other than those able to make a living working for themself, or via self-sufficiency), but this is not going to be utopia where we are all idle rich practicing our hobbies. What it really means is a welfare state, where formerly proud people capable of supporting themself become dependent on government handouts. What comes to mind is the movie "Soylent Green", not utopia.
I'm not sure how anyone imagines this would actually work. Is there any private enterprise left at all, or are all the means of production (AI datacenters and robotic factories) all controlled by the state. Instead of distributing Soylent Green, the state distributes UBI-scrip, essentially food-stamps, that can be exchanged at state stores for provisions?
Seriously, how could this actually work ?
An alternate future, no more optimistic, at least in the short term, but perhaps more realistic, is that in the fairly near future when unemployment and public pain becomes high enough, we reach a tipping point, and the pitchforks come out. Eventually the government concedes that AI is no more conducive to the public good than nuclear weapons, and heavily regulates it, banning the use of AI to replace jobs. This may sound extreme, but surely not a fraction as extreme at the UBI-based welfare state that Dario Amodei is fantasizing about as the best possible outcome (that is compatible with himself becoming enormously wealthy).
This *seems* does a lot of heavy lifting here. That's not what data shows.
> The middle class is being gutted and descending into survival mode if not poverty, while the top few percent become obscenely rich at their expense.
The middle class is getting smaller because people are getting richer and start to qualify to higher class, which is opposite of what you just said.
The US has the 2nd highest poverty level of all OECD countries, second only to Costa Rica.
What about other countries in the world where people... live in poverty?
> Capitalism allows people to escape poverty by personal effort
Isn't it possible that there is a future where AI makes the average value of human labor (or, "personal effort") plummet? Perhaps capitalism will lose some of its edge against a technology like this.
My own $0.02 on the economics piece - every country should have a sovereign wealth fund. Governments should block market access from automated[0] companies until those companies provide equity contributions to the wealth fund for that country. This aligns regulator and corporate interests. Dividends flow into the sovereign wealth funds and then can be allocated locally from there - UBI, job programs, etc. Let different jurisdictions explore different ways to structure a post-labor society.
On the broader social front - I think a lot of lack of meaning discussion boils down to the overemphasis we have on your job as your self-worth. We need to realign our societal expectations - and people need to spend more time with their families.
[0] for this to work, I think we would need well accepted metrics for 'how automated' a company is - and that probably needs a 3rd party auditing industry.
Is there any government that has gotten socialism correct for its citizens? I'd point to UAE/Qatar if they didn't depend on human servitude and inequality.
I'm only suggesting (eventual) negotiations between countries and automated corporations - to operate here, you contribute equity.
The only way to win is to wield the AI.
If you are correct, I expect corporations to reap massive profits while most Americans try to find a way to survive in a world where they are obsolete.
> In the near term handling the transition. Jobs will be lost, careers ended, people won’t be able to reskill quickly enough. At the same time AI is an enormous opportunity to uplift living standards, but nobody has the logistics of this figured out.
> We need to figure out how to restructure the global economy. How does UBI work internationally, if the AI companies are taking revenue in the US? What’s the tax base for it? What does that say about international trade and protectionism? Do countries end up splitting into different trading blocks based on their level of access and legality of AI (I assume some will ban it outright)?.
UBI in the United States is never going to happen in time. If it happens at all. We don’t even get universal healthcare. I think people who think AI will be a net positive for humanity are also in some sort of denial.
In a different US political climate I would entertain it. If these frontier labs weren’t so clearly going after the money, I would entertain it.
LLMs are clearly a step up for capitalists so I just can’t see any inclusion of LLMs move towards more progressive ideologies.
> How do we replace the work ethic that tells us we are our jobs and idleness is immoral?
For many people it has nothing to do with morality, it's hardwired into their instincts. They want to work, and they will work.
For a lot of us who are not excited about this future it's that no one is trying to answer all the questions you laid out. Instead we have the disgusting people at the helm purposefully spreading doomerism and saying, "We'll figure it out." I think it's pretty problematic (to say the least) to care more about technological advancement than how that advancement is actually shaping up to effect people in the short term. But I know many people don't care, especially those who believe they won't be among the affected.
Humanity survives (but we reading this probably don't), the AI treats the living humans like the Emperor's favorite pets (probably a pretty good life), and then the AI does whatever else it deems important.
In my experience modern models are better at all tasks than models from two years ago, especially complex multi-step tasks.
It wasn't "better" it was better at kissing your ass which matches what a lot of people want in a partner.
I suspect GPT 5.6 would be even better at it, if given the same sycophantic system prompt and lack of guardrails.
If you want creative writings, use the API and play with the sliders.
For example, every day people teach teenagers how to drive and with only dozens of hours of practice, they are on the road.
Another interesting task would be to take the AI in a robot body into a vegetable garden and teach it to pull weeds. This is another task that lots of children help out with.
I think most are actually worth, as agentic harnesses seem to optimize for solving poorly described problems rather than following complex procedures as written. In other words, instruction following maximizing models seem to make worse free-form agents, but they're really all that some domains need.
You can do many more things, when stuff is cheaper, even if the stuff were otherwise unchanged.
What? GPT-4.1 was not a small model! And why wouldn't you use reasoning?
You're of course going to see poor results when you restrict yourself to small non-reasoning models, but why would you?
In voice interactions, ttfat is actually relatively important. If you look at models with a <1s ttfat you eliminate almost every reasoning model, less some of the diffusion models and more obscure ddtree/dflash like speculative decoding implementations.
GPT-Live, which is coming to the API soon, responds instantly while reasoning in the background. So it can say "Hold on, I'll look that up for you" and continue to respond to the user conversationally while running an asynchronous reasoning task in the background.
It's not in the API yet, but it should be in the coming weeks. You'll see an enormous improvement compared to GPT-4.1.
That’s not a credible position, but there isn’t anything that I or anyone else can say to someone who simply doesn’t want to believe something.
I don't know if I agree with that but it doesn't seem like an irrational claim and does seem credible to me.
so far there is no end to this progress in sight so it's full steam ahead on this singular domain. once it plateaus you should expect to see the greatest disruptions in human endeavors ever as all the training flops will start flowing to other domains to disrupt and dominate.
I heard that Gary Kasparov was impacted by AI chess, but at least he still seems to have a job, so don't give up.
My point being that AI math is a narrow skill just like AI chess and implies nothing about generality (AGI).
Sarcasm begats sarcasm.
not sure how many will get this reference but "AI" for science and math is like super-shoes for runners
at first we are blown away by the impossible improvements including sub-2-hour realworld marathon and every other PR/CR/WR is dialed down
but then the improvements slow and reach a stall point because of the limit of technology and the source of the achievement
ie. sub-2-hour marathon yes, sub-1-hour never happening (rollerblade inline-skate record is 1-hour marathon)
The fact we see a lift is not the same as evidence that the lift is unbounded.
The lift being finite is supported by the fact improvements have come at the edges: improvements from human feedback, improvements in harnesses, improvements on model compatibility with harnesses, improvements in inference efficiency with new architectures, etc. If we were just training better models from scratch that would be one thing, but we are just making better use of a tool we've developed.
basically everything Benjamin Franklin did was trial and error because no-one understood what electricity was in the slightest
almost everything Edison did was trial and error too, he had his lab try thousands of materials for his long lasting lightbulb filament
even the most advanced "AI" today is just machine-learning going through everything already known trying to piece together previously discovered facts, admittedly at levels and detail impossible by human hands
but that means there are limits and it's not really "AI"
I have seen it produce tentative genetics for experiments, just like a scientist does. Then the data comes in and it can evaluate the data from the experiment just as well. One just had to give it a lab budget.
As a programmer, I am mostly interested in whether my role is sustainable long-term and whether the models will get better. I don't feel in jeopardy yet, but two more years like this and the calculus of hiring software engineers could shift even further. QAs are already overwhelmed with work
but with super-shoes more and more runners are qualifying for boston marathon and even olympic trials marathon where it would have been impossible for them previously
and that's what "AI" currently does, it allows average people to immediately "pick the brain" of every expert in every field, in every scientific paper, without previously reading a single other google result, something that would have been impossible for them previously (super-shoes for the brain? too far?)
but "AI" isn't creating new knowledge, it's just stitching together existing knowledge from patterns that would have taken years by human hand if even possible at all, it's going to "hit the wall" eventually (in its current form)