> Around COVID times many top universities experimented with removing test requirements from admissions, under an argument largely related to equity. It's been a failure everywhere, with many, if not most, universities already reversing it.
It's the universities that have failed. They've restricted admissions to a set of people who would learn no matter what the schools did, which is what makes them lazy.
When confronted with a set of students who haven't been provided with an enormous amount of childhood reading material, and the time, encouragement and social acceptance to indulge in it (the most faithful test predictor is childhood pleasure reading, the next best is parental income), they fail horribly.
The purpose of elite colleges for students is credentialism and networking, the purpose for the schools themselves is to force cultural conformity onto smart or extremely pressured students. They generally just tell you to go learn things by yourself. They have no particular insight into teaching, because they are supplied with students who don't need to be taught.
Can you cite a source for the claim about "most faithful test predictor"? I'm genuinely curious. I would think high school GPA would be more predictive.
High school grades are not evenly applied, and sometimes heavily inflated. Eliminating that variable is the whole point behind taking a standardized test.
Pretty sure that's what NSO Group (https://en.wikipedia.org/wiki/NSO_Group) is. Israeli intelligence could also just insert vulnerabilities in cheap garbage (or even more expensive garbage like this) for NSO or NSO-like Israeli orgs to take advantage of. We know they sell pagers.
> Should not there be a tax to offset the "frictional" unemployment?
Absolutely not. That's like taxing shovels because people were digging with their hands. The result is just more people having to dig with their hands, the fact that 7/8ths of people who now dig with their hands will starve because shovels have been introduced is a choice that we are making. We are choosing not to feed them.
Creating an unnecessary pretense so that they can suffer before they are fed is psychotic. Pay them to go to school or to show up to healthcare appointments, not to do work that can be automated away.
> the fact that 7/8ths of people who now dig with their hands will starve because shovels have been introduced is a choice that we are making. We are choosing not to feed them
The unfortunate truth of western economic history is that capital does not willingly share its profits unless forced to (by government or labor).
The core point here is about power.
Assuming AI takes off and automates large sections of the economy, who gets to have control over those entities?
It seems a bit premature to identify the current major AI labs as inevitably being the ones to benefit.
But I am sympathetic to the idea that having the public as a (mostly) silent partner, both in profits and control, is prudent.
Where it gets dangerous is how "the public's" equity share is represented and by whom.
F.ex. I'd be vociferously opposed to the current kleptomaniacal US administration being able to wield 50% control
> And the ones spanning generations were completely fair game.
No they weren't. Copyright at this point covers things for at least half a dozen generations back, and is intentionally made annoying enough that it is difficult to find out what is covered and what isn't. LLM companies didn't bother with any of that (they just pirated like your average online 13 year old), meanwhile archive.org got sued for pulling the music off ancient wax cylinders.
Clean up and dramatically shorten, restrict, or even eliminate copyright, and we can start talking about what's fair game or not. People were afraid to sing "Happy Birthday" in movies for probably 80 years, and the corporations that own all IP made it very clear at the time that they preferred for the status of "Happy Birthday" to remain unclear, and would send you a scary letter if you used it.
> probably going to depend on whether AI is a transformative use.
It's probably going to be entirely political, and decided through corruption. It's obviously a mechanical transformation. If rap DJ's got sued for cramming songs full of 80 extremely manipulated samples that you'd need a forensics expert to trace, and all sampled music had to revert to a form where they'd license a single song and re-release it with somebody rapping over it, LLMs are a violation. DJ's were doing an absolutely creative translation, and LLMs are not creative, they are Δ-following pinball machines.
I am repulsed by this because it will obviously be the vehicle through which tax money will be directed into Altman & Co's pockets, but I also understand that they will get bailed out whether the government gets a share or not.
As long as they are voting shares, I don't see an increase in the harm. I'd like to see a legislative framework about how that ownership is handled that allows Congress and regulatory agencies to make decisions restricting how these companies will operate, but without any regard to the constitutional rights of the corporate persons or their owners.
I'm sick of the government arguing with monopolies, then taking dives. I want it to be abundantly clear that government has the ability to restrict these AI utility companies freely (such as their ability to feed on their customers), while still limiting the rights that the state has over the personal use of AI by private individuals. Partial state ownership will make that possible. Hell, let half their boards be publicly elected.
Seems like LLMs are that. A bunch of most probable word associations is a network, and you can build a physical model of a network, or build a network that allows you to reason about a physical model. Whether it's just a flowchart or workflow diagram, or an X-dimensional matrix with vectors moving through it.
But the only way to map the network in an LLM is experimentally. You have to prompt it, and see how the coefficients fall in order to construct your most likely walk through the training data.
I think that LLMs can and do come up with novel things through exhaustion, just by applying the relationships between some set of entities to entirely different sets of entities because an accumulation of earlier context pushed the probability of those entities being mentioned, and they were able to easily replace a selection of entities that were more associated with those nearer connective, relationship words.
I think that as such LLMs are good at generating metaphors, and a lot of innovation comes from going "What if As worked like Bs?" Just go through all the As and Bs, toss the ones that don't make any sense and test the ones that seem like they might.
Would you say the same about any other tool, like where is the revenue caused by Susan in accounting having a computer, shouldn't we take away her computer if she can't prove a benefit?
The benefits of having a computer that we can now interact with in plain natural language, that can extract intent from vague questions/statements, and that can piece together answers is obvious.
The link talks directly about the disconnect between the supposed productivity benefits of a technology and the measured productivity benefits of it in practice. And provides historical context about why the “obvious” benefits of a computer did not materialize when it was introduced; business and their processes had to be rebuilt around the computer before real gains were seen.
Nobody is talking about hand waving. Look at progress in models between the original ChatGPT release and what came out this year. The progress is incredible, both in the frontier models and in the smaller ones that can be run in high end laptops (<100GB of ram).
There are new architectures, paper, and harnesses coming out weekly that improve performance, accuracy, and/or performance efficiency.
Whether they can do better or not than a secretary is a poorly defined metric. But by objective metric they are already producing less buggy code than $30k developers, and doing it faster.
Only if by "incredible" you mean it takes vastly more resources to do the same thing slightly better but with an even larger chance of completely fouling the bed.
But by objective metric they are already producing less buggy code than $30k developers, and doing it faster.
That's like bragging that a $1 trillion tank can go faster than a cheap budget car. For $1 trillion being slightly better isn't good enough.
First off, the state of the art models are perusing a strategy of doing more with more, but there are lots of small models that can be run locally on high end pcs. Those small models are more capable, are cheaper, and are faster than the state of the art of 3 years ago.
If you consider the change in what’s possible in June 2026 vs 2025 (to say nothing of 2024) “incremental” than you and I are living in different realities and I don’t know what to tell you.
not sure one would expect huge revenue increases from these internal tools, but maybe dramatic cost savings? Surely a lot of corporate processes could be automated?
That's been the dream for the 40 years I've been paying attention. And in that time, I've seen plenty of incremental changes but never the kind of sudden sea change that the hype machine anticipates.
The perennial reality is that automation is inherently inflexible, so there's only so much of it that you can do before you've committed a huge strategic blunder by making your business resistant to change and severely curtailing its ability to cope with situations that don't cleanly fit the mold. So then we need to hack in ways to deal with the exceptions, but, since they're hacked in, they're often painful and time consuming. Sometimes so much so that after the new process stabilizes it turns out to be even more cumbersome and require more manual effort than the system it replaced.
When anyone other than a technologist suggests doing that kind of thing, we call it "bureaucracy", and we hate it. I think maybe what we have trouble seeing is that there's actually a pretty fundamental difference between automating purely technical processes like server deployment, and automating processes that are fundamentally about mediating human interactions.
I worked for the state and had to participate in procurement. Not only did they do this, but when we were purchasing commonly available things, we purchased them from weird insider vendors whose catalogs were literally photocopies of other catalogs with the prices blacked out.
When I left, people were advising me to become a vendor, which was what a lot of people did as a retirement plan. You'd go to the secret portal and fill out the inscrutable forms, then give them to someone who you probably knew at the special time. Then the state would order things from you, and you would simply order them from Amazon or Uline or whatever. There was also a trade in minority and female figureheads (to white male businesses) to get you prioritized.
Money is how you test whether someone is currently in privileged circumstances (be it their own doing or not), not whether they are good at argumentation or decision-making.
It's the universities that have failed. They've restricted admissions to a set of people who would learn no matter what the schools did, which is what makes them lazy.
When confronted with a set of students who haven't been provided with an enormous amount of childhood reading material, and the time, encouragement and social acceptance to indulge in it (the most faithful test predictor is childhood pleasure reading, the next best is parental income), they fail horribly.
The purpose of elite colleges for students is credentialism and networking, the purpose for the schools themselves is to force cultural conformity onto smart or extremely pressured students. They generally just tell you to go learn things by yourself. They have no particular insight into teaching, because they are supplied with students who don't need to be taught.
reply