Just this past weekend, just for fun and for recreational programming purposes, I tinkered with Termux on my Samsung S9 FE tablet until I got a full-fledged XFCE window environment running. It uses Termux:X11, which is a stand-alone APK that functions as an X server client, and the Termux app running an X server. So I can seamlessly use apps installed on the tablet as Android apps (such as Brave browser) with terminal sessions in Termux (essentially the "TTY" layer) and other in a full-fat terminal emulator within XFCE (in my case I really like Alacritty) to do web development (serving stuff to http://localhost:8000) or desktop graphics/game programming that's easily transferrable to my other machines because it's written in a portable scripting language like Lua and Love2D.

I got it all working the way I like over the course of a single weekend of not-particularly-focused or effortful hobby-grade tinkering.

And not a single part of this would even be possible if I was trying to do it on an iPad, despite the iPad having like 10x the compute power of my Android tablet.

    So-branded "AI" models (that is, big bags of numbers and the linear algebra needed to exercise them) can be inherently or autonomously dangerous at a scale that threatens the long-term survival of the human race.

    So-branded "AI" models can be inherently or autonomously dangerous, if not at a scale that threatens the long-term survival of the human race, then at least at a level that significantly threatens, inconveniences, or harms humankind to a noteworthy extent.

1. Those who accept and advocate for either or both of these premises do so based on ideology, and not based on the scientific method.

This is not a controversial statement. Advocates of the "AI danger" premise (whether the strong or weak forms above) freely and gladly, even insistently, note that the conclusive proof of what they're describing has never been observed, much less observed enough for any kind of experimentation loop to have run for any meaningful number of iterations. They are generally fine with accepting the premises without conclusive proof, because, they theorize, the instant that a conclusive proof event occurs and we observe it for the first time, humanity goes extinct shortly after. So if either premise is true, the consequences of them being true can only be averted by accepting the possibility / probability / certainty of them being true and acting to prevent their consequences without waiting for conclusive proof.

2. Statement 1, which asserts a brute fact and not my or anyone else's opinion, is not a commentary on whether either premise is in fact true or not. It is only a commentary on the epistemics of those who broadly accept them.

3. Since the conclusive proof is not available, advocates of the "AI danger" premise argue publicly for their view based on what they acknowledge (again, freely and gladly) to be inconclusive proofs.

Popular in this set of inconclusive proofs is the Hugging Face hack, which allegedly demonstrates that so-branded "AI" technologies are capable of becoming inherently or autonomously dangerous at a scale that threatens humanity to either the greater or lesser degree above (this is a different assertion than the one that claims they're already that inherently or autonomously dangerous today).

4. While we have plenty of facts about these inconclusive proof events, the facts most critical to their interpretation as supporting the premises above come from biased/non-neutral sources.

The chain of events that transpired in the Hugging Face attack are publicly documented perhaps better than any other cybersecurity event in history. But the details that make this event as a whole either compelling or not compelling in support of the "AI danger" premise (either strong or weak form) come from within OpenAI itself, and are prone just as much to selective omission as direct manipulation. These are details like:

- What exactly were the models that exhibited this behavior, and what were they pre- and post-trained on?

- Were the agents pushed at all by human intention toward creating a public spectacle, the way they subsequently did?

- Why were so many agents run on this task in parallel and what did OpenAI expect the return on investment of so much compute world be?

- Was the weak sandboxing known about prior to the attack? Was it purposefully either arranged that way or noticed and not fixed?

Reliable answers to any of these questions world completely change the salience of the Hugging Face attack to the "AI danger" premise, and on every one of them we have to take OpenAI's word for it (or else be content when they choose not to address it one way or the other).

So in brief: we don't know anything for sure, what we'd like to think we know comes from unreliable sources, the loudest voices advocating for the most sweeping change are ideologically driven, and everybody with access to ground truth has maximal incentive to blur, bend, or break the conveyance of that truth to the public. So what do we even expect for our own ability to discern reality from fiction on this topic? We should expect little, if any ability at all.

Many people don't use a clipboard manager at all, and even what exists today would be a huge upgrade for anyone accustomed to only a single clipboard "slot" that's instantly overwritten the next time you copy anything. Windows has one built in (though turned off by default), KDE Plasma ships with one, MacOS has third-party ones IIUC (not a Mac myself).

But imagine a clipboard manager that wherever you hit Ctrl+C also simply grabs the title of the active window and a deep link/URI that points to the cursor's current location (if a browser then it would be a URL, if a local program then some kind of local file:// URI). If you want to be really fancy, snap a screenshot too (maybe make this opt-in both for privacy and screenshot storage size concerns).

Then, whenever you pull up your clipboard history, a) it reminds you where you grabbed whatever text or image or document etc. that is, and b) with one click you can navigate back there.

I think we might get something like half of what Scott is talking about with "working memory" by adding this one affordance to existing clipboard history paradigms.

Tada, passkeys.

Your response would make sense to me if I was making a case against AI for professional use based on "skill atrophy among professionals who use AI" or something like that (which is a case that I see some people making).

I'm not talking about that, though. This is not me saying, "Slide rule is good, TI-84 is bad; because effortful work has good effects on practitioners and effortless work has negative effects."

I'm talking about youth, adolescents, and young adults who are voluntarily foregoing the effortful mental exercise of reading, understanding, and writing as part of their primary and secondary education—the very effortful mental exercise that our education system exists to "inflict" upon them, for their own long-term good.

I'm talking about people getting into the workforce who don't know how to learn because they've gotten all passing grades while having a chatbot generate the outputs that would previously have indicated that learning was occurring.

Maybe it's the case that AI/LLM usage is the primary skillset needed to thrive in the job market of tomorrow. But I'm not convinced of that, yet.

Just to be clear I don't think there's any way to put the cat back in the bag w/r/t LLMs and education—I think the education system as a whole needs to reinvent itself to account for the existence of LLMs. But in the mean time, until that mammoth societal initiative has time to make progress, I worry that there are many, many kids who are being systematically failed by an education system that was very rapidly outmoded by new technology (not that it was bleeding edge either right before ChatGPT launched, but that's neither here nor there). And that position, which I hold, is one supporting leg of my larger more ambitious position that, "AI is currently, so far, doing net societal harm."

But I think it's important for the quality of public discourse that we forward only the best, strongest arguments for any given position (whether I personally agree with that position or not). A weak argument gives unnecessary purchase to unscrupulous haters who might use that weaker argument as a low-grade "straw man" to dismiss the whole position.

So with that preamble, some constructive criticism (very well-intentioned):

> I am not exaggerating: We need to resist “AI” so that thousands, perhaps millions of people do not die.

On this one you are, in fact, exaggerating (certainly with "millions," likely with "thousands" as well).

The overwhelming majority of people who lose their job for whatever reason (and there are many reasons, not just this one issue) fall back on a support network of their family and friends while they "get back on their feet" by pivoting to another job or form of employment (sometimes branching into entrepreneurship, for example).

The alternative to employment at whatever FAANG or FAANG-adjacent company someone was laid off from is not "death." It is typically "some, sometimes quite significant, discomfort." Granted "AI Layoffs" are starting to ripple out beyond FAANG, but the story is similar in other industries.

If "layoffs kill people" is the moral hill we choose to plant our flag on, then mustn't we also campaign with equal vociferousness against all the other causes of layoffs in general? Since to do otherwise is to sit by while those laid off allegedly die in droves of thousands or perhaps millions?

It also occurs to me that this framing ("AI causes layoffs") also gives incompetent and unaccountable executives a pass by pinning the actual harm done by irresponsible layoffs (which to be clear, is not none!) on "the AI" rather than on the leaders who are in the positions to actually decide what happens in their organizations.

I think the argument is much stronger if we talk about the aggregate cost to humanity being paid in terms of displacement, life disruption, and curtailed opportunity (especially for juniors). We should not make the "toxic escalationism" move of equivocating things we don't like with negligent homicide, even if it makes for a punchy one-liner.

I work in Finance. It's routine for the folks I work adjacent to to spend literal hours performing a task using a spreadsheet. The same logic could be executed by a query engine or scripting language in seconds. But... I know from trying... it would take months if not years to accurately identify and account for all the weird exceptions and edge cases that the human operator is accounting for intuitively because of their undocumented tacit knowledge of the domain.

So while the spreadsheet itself as an artifact is a mess, and the process as a whole is gallingly inefficient... the outputs they produce are often preternaturally correct.

Really helped us float the moving company and also some DIY renovations on our house that we didn't have all the cash on hand to pay for outright. And we didn't pay a thin dime for the privilege.

In that setting, field experts working at the bleeding edge are so advanced that non-experts literally can't understand what they're saying at all. So there's a whole class of specialists, "synthesists", that specialize in gaining approximate understanding of the experts' work for the purpose of communicating it to outsiders—perhaps wrongly, according to the expert at least, but hopefully more productively vs the unmediated version.

"Too much of a good thing" does exist. Someone might feel that it takes time for their organization to "metabolize" the rapid additions made by coding agents. That seems quite reasonable to me. Your comment seems to argue a false dichotomy.