Much of the present-day debate regarding the use of LLMs and generative AI centers around topics such as necessity, accuracy, intellectual property rights, utility, efficiency, economics, worker protections, environmental impact, and future developments. But very little, it seems, is focused on what I think is really at stake, which is the centralization of authoritative truth.

If we regard such models and algorithms as "tools," then yes, there are advantages as well as disadvantages. We can discuss the benefits in productivity/efficiency/knowledge synthesis/discovery as much as we can discuss its impact on the labor market. We could also add in the context of where the training data came from, and whether the originators of such data possess rights to its use in this manner. We could talk at length about how data centers impact the environment in which they are built as well as the energy demands they make. We could endlessly speculate on various predictions for this technology as it might impact the human civilization long-term.

But what I would like to do is make a simpler statement, which is that at its core, these models--and in particular, the way they are being commodified and therefore utilized--represents a fundamental regression of the democratization of human knowledge. Some may say that, by assembling such vast quantities of information in this manner and then making it accessible to the world through the internet, we are improving access. After all, people who are not artists are using it to generate "art"; people who are not programmers are using it to "vibe code" computer applications; and in general, people who are not previously taught in a domain of knowledge through traditional means are using it to appear to attain a certain degree of competency. This is a seriously flawed argument; one that does not become more persuasive if the LLMs become more accurate, more knowledgable, or more capable.

What I mean by "democratization" is not simply about access to knowledge. It is about who controls that access, who curates it, who says it is the truth, who is accountable for the consequences of its use, and who teaches the critical thinking skills that facilitate its interpretation. In all these respects, the entities who have created and commodified these models are failing miserably, and if we look at their behavior, all of the evidence very clearly points to sinister motivations--that is, the failure is intentional. These people do not want you to think and reason and learn for yourselves, because that threatens their ability to sell that capability to you, and more importantly, to be the ultimate arbiter of truth.

It's a red herring to argue that individual humans are more fallible in their judgment, less objective, less knowledgeable in scope; thus are more prone to believing lies. The accuracy argument is misdirection. The point is, if we don't have the capability to interrogate and question, if we leave that to the machine, who is pulling the levers on that machine? And in a rather frighteningly perverse way, it may actually be preferable to have a human in control of that, as megalomanical, authoritarian, and evil as they may be, than having the machine in control. But this is my point--we shouldn't be letting this kind of centralization of power to occur in the first place. I don't want a dictatorship of knowledge, whether or not an oligarch trillionaire or the machine itself is in command of it.

So yes, I object to this technology, in as much as I object to, say, making nuclear weapons. People can point to "positive" examples of the technology--we've heard the arguments that these weapons keep us safe through a policy of mutually assured destruction; or that the underlying physics have given us nuclear energy, therapeutic radiation for treating cancer, and all kinds of other knowledge about the universe we live in. But will any of that matter if those weapons are used? Now ask yourself who has the power to use them, and what their motivations might be. When faced with the evidence, should we not be similarly skeptical of LLMs/genAI?

When someone says "I don't care" in this context, it isn't an expression of indifference, as if the issue doesn't have any material impact on them. It's an expression of DISDAIN. For example, when someone says "I don't care what Trump thinks," it is reasonable to infer that they mean that his thinking is fundamentally wrong and unworthy of discourse. That he is not someone whose thoughts are to be respected or regarded with any value.

We DO need to stop these arrogant sociopaths. And we do that by refusing to buy into the explotative ecosystems and false narratives they push on us. But let's not read too literally into what is a common expression of frustration and twist its intended meaning.

Why should I care what this dumb asshole thinks? If he takes up entire world's supply of RAM, then with what devices will we be able to presumably pay for the privilege to access his AI models? Is he really suggesting that we all go back to the days of running dumb terminals?

What is this CEO disease, where they all think that just because they run a company, that they are suddenly subject matter experts for every domain of knowledge that built said company? Like, somehow just by being in a leadership position, they suddenly have the knowledge of all the engineers, developers, and technicians who built the product, as if they built everything with their bare hands? And what is with this mass delusion that so many people--shareholders included--buy this absurd narrative?

Fuck this guy and fuck his arrogant "I'm the leader so I'm a visionary who can say and do no wrong" attitude. Put them all on a spaceship and blast them out of the solar system.

But that's exactly my point: if all you want or need is competence, then according to the professor's own admission, an AI will do that. No need for your job, no need for grants, no need to teach. What is the purpose of teaching future generations if an AI will be better at mathematics than you ever can be?

And what I am saying is that either (1) the framing of AI as being the hyper-competent replacement for everything intellectual that we do is fundamentally flawed, and/or (2) it is necessary to understand, for the well-being of our own species, that "competence" and "results" are not the end-all and be-all of a task.

If we fail to appreciate the importance of doing and learning things for their own sake, and to find satisfaction in that process, then we will cede the task of thinking to machines. We can already see early signs of this. And the consequences of this short sightedness is catastrophic.

Let's be clear about the nature of education. Before a student reaches a particular level of self-discipline during their formative educatoinal years, homework is assigned because it is an effective mechanism to have them practice the concepts being taught to them, and to do this outside of the structured classroom environment. This activity promotes the crystallization of learning objectives and prepares the student for the next lesson, which builds upon the previous.

After a student becomes largely self-sufficient, the homework model becomes less about enforcement through grading, and becomes more about preparing the student to manage their own academic development outside of the lectures and office hours. The student goes to class because they understand they NEED to be there to learn, unlike a child who must be TOLD to be there.

In this regard, whether or not homework is graded, and whether or not a student feels a need to cheat or take shortcuts with homework, is really just a reflection of the student's maturity and sense of responsibility for their own intellectual development. Typically, a professor should not need to police what a student does on their own time; what matters is the ability to certify that a student has met the learning objectives--and that is done through examination.

So, what I find disturbing is not the question of whether this professor should grade his students' homework, or whether the students are using AI to do their homework for them instead of actually learning the material. As the old adage goes, cheaters are only cheating themselves--at least, that's what I would still like to believe.

No, what I actually find most disturbing is this apparent disconnect in the professor's views on the future of AI as they pertain to professional mathematics, and the impact of AI use on the education of humanity. He predicts that AI will be better than professional human mathematicians, and I think this belies a deeply mistaken perspective on the purpose of learning and doing mathematics, one that I am extremely surprised that a working professional mathematician would make.

Even if AI models far exceed the human capacity to prove theorems, will it ever FEEL enjoyment and excitement about doing mathematics? Because if it can't stir something in you, then why do it, and why bother teaching it to others? Why would you ever need to teach another student, if you could just always ask a computer instead? In fact, why will anyone in the future ever need to learn anything, if AI will be the replacement for our brains?

I am a human being and I feel. I dream, and I love. I love mathematics, and it is not simply the result or the answer that I love, but the entire process. The struggle, the failure, the questioning and working away at it until at last, there is UNDERSTANDING. And the flaw that I see my fellow humans are making, over and over, is this crazy focus on RESULTS. ANSWERS. GOALS. They seem to have forgotten how to enjoy the process of doing things, whether it is the restoration of a classic car, or composing a musical score, or deciding what to cook for dinner.

I mean, why spend months and months on taking apart a 1957 Buick engine, lovingly cleaning every component, repairing and sometimes manufacturing replacements, and reassembling it? Why, when you can have a new electric car that is far safer and more efficient, with more functions and conveniences? Let me be clear; I don't work on classic cars or have any experience with automotive mechanics. But even from my outsider's perspective, I instantly know, at a visceral level, why they do it. It is for the love of the process; to take this machine and build it with your bare hands so that you have a complete understanding of it; and that this machine is an embodiment of engineering, aesthetics, and a symbolism of the freedom of movement.

That's what AI will not, and will never, do. It can't fall in love with the process, because to do so is what it means to love learning, and is part of what it means to be human. And for a professor of mathematics to not appreciate this is, to me, very misguided, and does not bode well for the education of future generations.

One other thing I forgot to mention. I'm skeptical that this "AI startup" actually did what it claimed to do, but since I have no evidence either way, I reserve my judgment until more details are given.

If the startup can't or won't provide evidence, that in itself is suggestive of some kind of fraudulent behavior meant to hype up the company, a la Theranos/Elizabeth Holmes. But in so far as the underlying scientific principles are involved, the strategy seems plausible. I just don't know if this company actually did it.

So I only remembered a little bit of the virology of phages from what I learned in school--namely, that they infect bacteria and that they are everywhere. And when I saw some of the early comments posted here that were big on knee-jerk reactions and scarcasm but small on actual science, I decided to read the Wikipedia article.

Phages don't infect eukaryotes; they only infect bacteria. And phage therapy has been successfully used to combat antibiotic resistance in humans, although the problem of resistance in general remains. This makes sense: bacteria replicate rapidly and have relatively little in the way of proofreading mechanisms, so when there is selection pressure, they mutate and develop resistance very easily, even to phages. So engineering a phage to target an antibiotic-resistant bacterial strain might work in the short term, but unless there's a way to cultivate phages that are able to adapt in response to (and not simply target) that strain, the strategy is intrinsically self-limiting.

To those comments drawing ill-informed comparisons to engineering viruses that infect humans, know this: phages are literally everywhere on this planet. They are far, far more abundant than bacteria, and most people can appreciate how ubiquitous bacteria are. You already have phages inside of you, and in fact, they form an entire ecosystem. Same thing with bacteria--you have bacteria all over and inside your body. You have more bacterial cells coexisting on and inside you than you have your own cells. And what's more interesting is that these phages seem to be protective by helping to maintain a balanced microbiome.

The concern I have with engineered phages is not that they would be harmful to the environment or to us, or that they could mutate to be harmful. That is extremely, extremely unlikely, because such a concern doesn't reflect an understanding of how phages actually function in real life and how they co-evolve with the bacteria they infect. If anything I would be concerned about spreading bacteria that become extensively resistant to antibiotics and phages. Thus we need better education about what phage therapy is, and to ensure that study participants are adequately informed of the potential risks and benefits. The direction such therapy would most likely need to take is quite similar to CRISPR and other gene therapies, in that it would need to be a form of personalized medicine. If we can develop computer models to help perform this personalization more efficiently, then I'm for it; such an approach may also prove useful for treating cancer.

Precisely. I alluded to this in my previous comment, but you give very simple and specific examples of this phenomenon.

Other examples with significant impact:

Ability to borrow money is almost nonexistent for the poor, except for usurious rates such as in payday loans or credit card interest. An absolutely astonishing percentage of Americans carry a balance on their credit cards from month to month, and pay interest on that balance. And the poorest Americans can't even access credit. This inability to borrow at a favorable rate impacts major purchases that would, in the long run, save enormous sums of money, such as financing a home or an automobile. A home builds equity and doesn't enrich a landlord. A car lets you commute to better jobs and saves time otherwise wasted on public transit.

Fines, fees, and consumption taxes as a proportion of income are vastly higher, and there is no legal recourse. Late on your credit card payment because your employer didn't pay you on time? The bank doesn't care and you don't have money to sue your employer, not to mention they might just fire you. Sales tax is extremely regressive; 8-10% makes no difference to the well off, but regularly stops the poor from being able to afford basic necessities--and many state/local governments continue to seek increases in sales tax in order to make up for budget shortfalls as a result of their own lack of fiscal discipline.

What bothers me is how the wealthy class is always quick to point out how much "assistance" is given to the poor, but when it comes to the poverty tax and actually having an equitable playing field in economic affairs, they are dead silent. And the middle class eats up this narrative, because they don't want to believe that they have more in common with the poor than they do with the rich. Most Americans are one health emergency away from bankruptcy. Most cannot afford to lose their job. Yet so many believe that they should direct their anger at the poor for "being lazy and taking the benefits their taxes pay for," when the truth is that the truly wealthy have rigged the financial and political system to pay almost no taxes and steal far more tax dollars than all of the poor are ever given.

It is a kind of mass delusion perpetrated by the rich in order to sustain their wholesale theft of the wealth created by the entire working class, and the fact that so few of us have woken up to this reality is what makes me want to scream.

What this allocation seems to suggest is that study participants who receive UBI feel compelled to distribute some of that income to their loved ones, because they too need assistance, and doing so improves their social circle's quality of life in ways that are not so material as buying goods and services. Therefore, if UBI were made available to everyone in need--and not just to study participants--then such redistribution would not occur at this rate.

What I've noticed with these studies is that critics of UBI and other forms of social safety net programs frequently come from entitled backgrounds. They have never known poverty and have formed this narrative of poor people "wasting" the money they earn or receive as the reason for their poverty. They can't understand why poor people don't just "find a better job," or "stop eating so much food," or "find a cheaper place to live." They look at the world through their own lens of plenty, and don't see how society severely penalizes the poor for simply being poor and extracts from them additional financial costs that the more fortunate do not have to pay.

In the same vein, there is this insidious undercurrent of elitism in the criticism that UBI recipients need to have their spending monitored and judged before they are entitled to it, because it reinforces this false narrative that poor people are poor because they don't use their money "wisely." Meanwhile, we have a billionaire class that truly wastes their obscene wealth on luxury yachts and monuments to their greatness. If anything, it is the poor--who have lived the experience of scarcity and have to decide every day what is truly meaningful in life--who should be entrusted to wealth, because every day they feel the tremendous pressure, anxiety, and insecurity of not having it, and therefore know how important it is to make sure that others do not suffer the same.

So when they give money to their friends or family when they barely have enough for themselves, you KNOW they aren't doing it frivolously. Unlike billionaires, who only love themselves, they know how important it is to share with the people you love.

This is absolutely correct, and it's extremely important to understand that this behavior is baked into the model. It's not even so much that it's "baked into" the model as it IS the model.

Ever since the earliest days of LLMs and generative AI, we have always, always seen what people call "hallucinations." But the only thing that qualifies such output as a hallucination is that we, the human observer, has judged it as such. By construction, the model DOES NOT KNOW THE DIFFERENCE. If it knew or "understood," it wouldn't have produced the spurious output in the first place.

So every now and then, we see some amusing audio/video clip of some human managing to fool or "break" an AI agent, and it makes the rounds of social media for people to point at and laugh and say "tee hee, how funny!" And then these same people, without thinking about what these hallucinations mean and what they imply about the intrinsic nature of the technology, immediately turn around and ask Claude or ChatGPT what they should make for dinner, or who to vote for, or who to hire and fire; or worse, give them access to their personal data and critical systems. That is a level of uncritical thinking and cognitive dissonance that ought to be profoundly distressing.

From a functional perspective, there is quite literally zero distinction between an AI generated pretty woman breaking and suddenly speaking in Cantonese during an English-language interview, and an AI agent that violates constraints imposed upon it and causes damage. They are both "hallucinations" in the sense that they are unintended behaviors/outputs, and as we have already established, the only meaningful arbiter of "intention" is our own human judgment. AI proponents try very hard to make excuses and minimize the severity of the problem. But it is intrinsic to all models.

Whether through laziness or ignorance, when humans cede their authority to interrogate and decide the truth for themselves and let machines do their thinking for them, who or what ultimately becomes the authority for that truth, and who is or is not accountable for the consequences? Who has a vested interest in controlling the source of truth and knowledge, and what are their motives?

People have been driven to delusion and suicide because of interactions with LLMs, and although you might be able to argue that they would have killed themselves regardless and were in need of treatment, no one can possibly know that, because that's not what actually took place.

But I think the more serious question is the framing of the premise: why is it that the point at which we ought to hold people accountable is if someone is killed? Who decided that the tipping point is when these AI agents start killing people, and why? Not only do most legal systems seek to prevent and punish abusive, antisocial, and outright criminal activity, a just society has a moral and ethical responsibility to apply rules and norms fairly. While we can see that the fair application of the law largely remains an idealistic fantasy and that those with wealth and status often escape accountability, the fact remains that adding AI to that privileged status certainly is not doing anything to help reduce that inequity.

So again I ask: why should agentic AI and their operators be permitted to do things that we would not allow under other circumstances? Just because some people find these tools useful does not excuse the requirement that they be safe and respectful.

Please do let us in on the joke and share your explanation of what you find so funny about this phrase.

Currently, there are no known functional cures for AD. There exist a number of treatments (monoclonal antibodies) for which there is supporting evidence for slowing cognitive decline and delaying the progression of functional disability, but the effect size is generally modest. To be fair, some people seem to be stronger responders than others; however, these amyloid-targeting treatments also have a risk of ARIA (amyloid-related imaging abnormalities), which are associated with microvascular hemorrhages (bleeding of the small blood vessels in the brain).

A subgroup of the general population is predisposed to AD because of a genetic mutation they carry (ApoE4). Each of us carries two copies of this gene, and we know that people who have one mutant copy (heterozygous) have a modestly increased risk of developing AD, and those with two mutant copies (homozygous) have a significantly higher risk. Not only is the lifetime risk increased, the typical age at onset is decades earlier; homozygotes may begin showing MCI (mild cognitive impairment) as early as in their 40s, and we call this "early-onset AD."

So, to answer your question, you are correct: there isn't really anything someone can do to stop or reverse the condition, but with early detection and treatment, they might be able to improve their quality of life and reduce the burden on caregivers and long-term financial cost of care. How such a person copes with a diagnosis is a psychological question. Some will spiral into depression and despair. Others might plan for the future in ways they might not have done otherwise, and find focus and purpose during the time they have. And it is for this latter group that the availability of such a test might be particularly helpful.

That said, having a validated and (relatively) noninvasive biomarker for AD pathology is a huge game-changer for AD research and drug development. One of the biggest challenges for targeting AD is accurately identifying the patient population. In the past, it was not always clear whether trial participants--many of whom were elderly (i.e. 75+)--actually had AD, or some other dementia, since the gold standard was to perform an autopsy. Because of this, clinical trials often suffered from a lack of statistical power due to a proportion of participants not actually having the disease that drug was intended to treat.

Since then, we devised various imaging modalities (e.g., PET) to use in conjunction with investigator assessment to improve population quality, but the more measures we have, the better the trial data gets, and the better researchers can identify targets for drug candidates. But we are still a long way off from figuring out how to prevent and/or reverse AD. One reason is that we believe that by the time patients exhibit symptoms of cognitive impairment, it's already too late--there is already extensive damage. So if there is a biomarker that can reliably detect subclinical AD pathology, this also has positive implications for designing clinical trials that aim to prevent AD disease in susceptible people, much like a vaccine for infectious disease or cancer.

It's a big dream, like being able to prevent or cure all forms of cancer. But I would like to imagine it is possible. These diseases take such a tremendous toll--on patients, on their caregivers, on the healthcare system, on society--and I hope there will come a day when they can be prevented as easily as going to your local pharmacy and getting a shot.

We generally don't let the general public have access to enriched uranium, high explosives, or chemical or biological agents. These are also just "technologies." Anyone sufficiently intelligent CAN gain access to hazardous materials. But we don't allow it because society has recognized that these things are too dangerous to not tightly regulate. Unauthorized possession is itself considered a crime, and this is because of the risk level.

AI agents might not reach this level, but this example serves to illustrate that your arguement--that "outlawing technology is not the answer"--requires more nuance. Many gun owners are fond of saying that "guns don't kill people, people kill people." That is also a specious argument. The gun makes it much, much easier to do it, and to do it in a way that is faster, more deadly, and more difficult to trace.

And the way these AI systems are designed is similar: whereas before, a competent hacker could go in and manually find vulnerabilities, now the process is automated and not even explicitly asked for. That's the crux of the matter: the more capability we train this technology to have, the more dangerous using it will become--just like a revolver pistol versus a chain gun.

The solution is not just technological or even legal. It's a societal one. We need to first acknowledge that agentic AI technology is intrinsically burdened with very serious risks, by understanding what it is actually and potentially capable of doing. We are at a phase of LLM development where there is a great deal of power exhibited by and granted to these tools but very little regulation due to the lack of such understanding. It might evolve to a point where computing systems have built more robust protections against abuse (whether intentional or not). It might evolve to a point where laws will be passed restricting certain uses. But to your point, the genie is already out of the bottle, and unlike, say, the relatively successful proliferation of age-verification laws and censorship of legal adult material worldwide under the pretense of "protecting children," the real reason why there hasn't been a massive concerted effort to prevent AI abuse is because the people who have the power to restrict it are the same ones who are making billions of dollars off promoting its use.

How difficult is it for an operating system update to detect whether or not a service is installed, and if not, then not install it unless it is explicitly requested?

I find it strange that this simple design philosophy is ignored. When a user installs or updates software, they should be presented with a dialog box with a list of components that can be checked/unchecked for installation, with any dependencies reflected as appropriate. The whole paradigm of "automatic" and "background" updates is absolutely bizarre to me. I never want the OS to decide if or when to update itself. I never want it to decide by default to install and enable a new service without giving me the option to choose. Microsoft seems to violate this principle with alarming regularity.

The fact that OneDrive at one point its its development caused massive data loss for users who found it had exfiltrated their locally stored files means that at no point in its future development can it be trusted. The risk is too high to not check whether it has deleted the local copy, relative to the inconvenience of performing the check.