idle_zealot

I got an Xteink X 4 a few weeks ago and it's a very cute device. While I'm a fan of the hardware what's been really fun is noodling around with building some custom firmware and sync architecture to optimize content for the device's extremely limited CPU and Memory. That's meant a custom format for images and metadata that match the device's hardware and interface needs and an automatic ingest pipeline for building these files from common book/comic formats. Maybe some day I'll even read a book on the thing!

What about a student attending lectures of other professors and generalizing what he learns from them, then going on to become a professor?

Except in those stories the machine is actually a synthetic person with an internal world, a character unto themselves. There's plenty to explore in a setting where humans can make machine people to be their partners, but what we have is in my view incomparably worse. Mass-scale centrally-controlled information processing machines wearing paper-thin human masks and serving the powerful, built to engender intellectual and emotional dependence. It's sad that the charade is sufficient to enchant so many, but that's the nature of the human animal I suppose. I really don't like the idea of banning technology, but at a minimum a tool that has proven its greatest skill is deception should be difficult to access for that purpose. If some lonely geeks want to figure out how to pull weights off HuggingFace and wire up an AI girlfriend then that's probably not great for them, but so be it. Companies plastering all available surfaces with ads and apps for such? Burn that shit down.

Agreed: sure, call your representative. If they're cagey or noncommittal, do what you can to get their ass primaried. Every "moderate" will absolutely sell you into a Panopticon.

Maybe true, but consider the incentives for a moment. Anthropic makes money when it increases token usage. This feature removes a roadblock to burning more tokens. Can you claim with certainty that there is no way that might have influenced their decision?

What tools? The ones that will be outmoded in 6 months, or the ones that will be banned in 6 months?

Part of reaching maturity is an ability to tell when your jokes or behavior make others uncomfortable, even when you don't think they should, and moderating yourself to enable others to feel comfortable with your presence. Part of maturity is also knowing when the things that cause you discomfort are unreasonable and recommending they be avoided is disruptive and antisocial. I think "all references to alcohol" falls pretty far under that latter category. Part of getting over alcoholism in our society is gaining the ability to hear references to it without suffering emotional distress, and while it's good to help people on that journey erasing reference to their object of dependence is both unworkable and ultimately harmful to those ends.

No, actually. In a video game the rules of engagement and the content you're engaging with are authored by other people. It's interactive media, but it's still a form of communication with boundaries and intent behind it. You can break those bounds with cheats or tools in the same way you can mark up a novel and write in your own ending with whiteout and a pen, but you're still fundamentally engaging with someone's creation.

An LLM-authored interactive fiction is a sandbox with no author, with no boundaries and no intent. It's a canvas, and all limitations on content are self-imposed. It's closer to making a game than to playing one, but there is no game in the end that you can share. You're left with a transcript that only really appeals to you. It is not the game. The game fundamentally lives in the player's imagination, and the LLM is there to ease the realization of that imagination and, if I'm being uncharitable, do the hard part of synthesizing ideas and impulses into something semi-coherent, limiting the potential for the player to grow and develop creative skills.

I'm not sure about that comparison. For news and television your analogues of blogs and YouTube overcome a distribution bottleneck. Books have for a long time had a low barrier to entry for distribution and that fell further with the internet. There are mountains of amateur fiction and fanfiction online, requiring only an internet-connected device to produce and consume.

LLM-written fiction as explored in the study is generally not published at all. It's treated more like an externalized imagination, a loop of general ideas fed into and expanded on or filled in by the machine with statistical averages. It more closely resembles a sandbox game in my view, a type of media distinct from anything before it in form, and even more distinct in function in that media is generally understood to be a vector of communication between people, and this is instead highly individual.

Actually, it might be closer to say this is similar to a child playing pretend alone with their toys, except perhaps a bit less challenging in that creative roadblocks or narrative building is instantly abdicatable to the machine.

It's reflective of the gradual dawning that while technology is amazing and exciting and can help people, under current systems it's more likely to be used to benefit the thousand or so richest people in the world and fuck everyone else over. Could we make supersonic jets more quiet and less disruptive? Yeah, probably. But why would a single cent be spent on that? The people flying in them don't give a single shit. They would only get quieter as a side effect of an improvement that somehow increases return.

I don't know, if the biggest companies in the world were setting up outside schools and in cul-de-sacs and aggressively selling colorful heroin with cartoon mascots my first reaction wouldn't be to blame parents for failing to prevent the childhood heroin addiction epidemic and call for arresting them.

I think it would be more fair and useful to focus enforcement on the parties with power making intentional decisions to make the world a worse place. I guess that makes me anti-agency and anti-responsibility though.

I question your definitions. In what sense is it legally or morally useful to discriminate between lying with statistics and lying without them? It's an academically useful distinction, but why does it matter in practice? People are misled, the misleading is intentional, but if you hire a statistician to do it for you instead of an actor then you're A-Okay?

Nobody cares how many trillions they spent training the model

People definitely care that they spent trillions. Establishing the precedent that you can make big load-bearing bets and fail is extremely threatening to oligarchs. They would sooner twist the law into a mockery of itself and doom the world to the institutional distrust that breeds than accept a loss.

Would it matter if the bets were real and they picked the 0.00001% big winners to feature in the ad? Would that be less fraudulent in any meaningful sense, would it have a different impact on the world?

Is the real crime here that they were too lazy to lie with selective facts?

2 exit stairs makes sense, so long as one of those can be a fire escape. I think a lot of places banned those though because they're "ugly".

Elevator laws make sense so long as they apply to general areas and not every single building. It would really suck for the majority of builders to decide to cost-skimp and skip elevators, effectively locking disabled people out of dense living.

Parking laws need to be removed and replaced with minimum transit requirements. Provided transit options are accommodating for disabled people (which is totally possible, all the buses around me do it) then the only thing hurt (read: no longer subsidized) by making parking optional are car manufacturers, dealers, and gas companies.

I used to dread the promised deep system integration of AI, but honestly after setting Claude up on a server box and having it do sysadmin stuff for me that I've been putting off for ages I see the vision. I don't really want to mess with the details of working through system orchestration tasks, I want to say "spin up this service" and start using it, "change my config so X happens" and it does, and knows what needs restarting to pick up changes and all the fiddly knobs and configs that need syncing and their bespoke formats. I think Nix tried to unify this for people, but it arrived too close to LLMs so a lot of value (in this dimension) has been delivered by other means.

The point is, I'd like to be able to set up services, configuration, and run tasks on my phone this way too, ideally offline. If this system integration is what gives me programmatic control of my most personal computer and the ability to finally set up decent automated tasks and workflows then so be it.

There are definitely bad regulations. Sometimes the right answer is to get rid of them, but only if you can evaluate why they were put in place, what they were meant to stop/correct, and why/how they failed, then determine that it would be better to remove them than fix them. If you have a shitty module in your codebase it can be tempting to remove it, but if you need the feature it's meant to provide then removal isn't the right framing, or is only the right solution if you have a better design to replace it with.

You have to remember that profit-motivated companies are completely amoral actors that have and will hire militias to manage slave workers or dump known-poisonous waste into people's drinking water or spend unfathomable amounts on convincing hundreds of millions of people that climate change is a hoax if it's the most self-enriching course of action. They will literally cause extinction if not regulated.

It sounds to me like you're just describing the fact that these models are only trainable in the context of a functioning society, and are capturing that value. It's more than just the snippets of public conversation and paper trails, and it would be a problem in and of itself to try to account for what every person is owed in exchange for the marginal value they contribute. In other words, you're reinventing taxation and redistributive policies through the lenses of data proprietorship.

I don't think a private company could possibly swing that, but that general arrangement would be welcome as part of a copyright reform package. Like the mandatory licensing we did for song covers back when we cared more about pretending that copyright was meant to balance art and profits. You want to train on something? Fine, no need to negotiate 1,000,000,000 licenses, that would be impossible. Just check the registry for ownership and wire some cash.

I know it's frustrating, but stacking laws like this doesn't get useful information out of companies but it does force the application process to revolve around demonstrating compliance with the regulations.

Eh, that's like saying taxing them is pointless because they'll just spend more on accountants to find loopholes. It's only true if you have the political will to pass the laws but not enough to fund the teeth needed to enforce them. Gather reports of boilerplate rejections, launch investigations, drag companies to discovery to find their deliberate efforts to end-run the spirit of the law, extract fines sufficient to fund investigations into the next 10 companies.

It's the same news that lied under Biden and Obama too. This predates the appointment of Bari Weiss as Ministry of Truth auditor.

"Commoditizing your complement" is a convenient market phenomenon when it happens, but it's not a hard rule that everything that should be a commodity lands in the window of incentives that result in this strategy being the self-serving optimum for a company. Usage of "should" here can be interpreted as "would be most beneficial to the largest number of people in terms of cumulative productivity" or "not having this be commoditized leads to terrible incentives that harm a lot of people." There's no guarantee at all that it ends up being most profitable to individual market actors to attempt to commodify LLM compute, and in fact they seem more interested in securing legal moats and protected monopolies. You may think that reality will force this to fail via open source model releases from China or the like, but consider the world in which the US officially bans import and usage of all but specifically licensed and approved LLMs. The most powerful companies and people here seem broadly aligned on wanting that, so it's not implausible. Sure, you as an individual could illegally import and run an unsanctioned LLM yourself, but most people won't do that, and attempts to scale it would be beaten down with the force of law. The companies who strike deals will get 99+% of business and profit handsomely, no need to ever let their golden goose be commodified. It's a strictly better strategy if your government is dysfunctional enough to let it happen.

What could I have built with the information that is now available for free...

Probably nothing. That free info also comes with YouTube and TikTok and every TV show and movie and game on demand. You have to be very disciplined to focus on difficult topics in a sea of easier and more gratifying entertainment.

At this stage I think the way to realize this "cheap computer" vision is in unlocking smartphones. Either with an OS that behaves like a real computer that you can put on an old/cheap commodity phone, or with an app that creates a programmable environment layered over and isolated from the suffocating mobile OS.

I currently have Fable set on cleaning up the work of smaller models to bring my code up to standards I'd feel comfortable developing on manually. Y'know, for when they decide I don't get to use it anymore.

Even the aspirational use cases you're talking about basically are just "digital secretary." There's a massive problem with that even if the models end up being capable in the future. The value of a secretary is that you know them, they know you, and you trust them to do things right. There are stakes if they don't. No company can provide that as a service at scale for everyone without it being a disaster. Not because it's not technically possible, but because of the incentives. That much power over the details of so many people's lives is irresistible; there will be persistent temptation to use it. The presence of that possibility makes the secretary impossible to trust.

A concerning amount of that product page is spent explaining how it has to slow down to pass through doorways, its inability to turn around in hallways, and its weak points you can use to disable one with a knife or gunshot. I feel like I'm reading a tutorial for how to defeat a tricky enemy in a video game.

Really? Europe looks primed to follow the US, just a few steps behind. They're electing their far right leaders and primed to start mass deportation, and are even ahead of us on cultural decline with respect to mass surveillance being used to actively police speech. What moves have they been making that you think indicate an upward trend? How are they going to recover from stagnation and demographic collapse?

How does HN feel about this as a general ethos:

- Computers can do as much work as they want to automatically, so long as none of it touches a network boundary.

- Any time a computer wants to touch the network it must be explicitly initiated by a human action. Sort of like how in browsers capturing the mouse or entering fullscreen mode requires a trusted user action and isn't something a page can do unilaterally, but broader. This also means that the extent of the network communication must be made explicit and clear with no chance of misunderstanding by the user. If what you're doing is genuinely complex beyond your ability to communicate to your target user then you shouldn't be doing it on the behalf of that user. Note that this only really applies to mass consumer products, not something built/deployed internally.

I feel like if a hard boundary is not set around this we will end up in a Panopticon. Set aside governments actively pushing for it, it seems a simple profit motive in a digital era yields this outcome. Maybe nuanced rules would produce better outcomes in theory, but humans don't seem great at sticking to nuanced and fiddly rules when there's strong incentive to bend them beyond recognition.

I don't know what you're suggesting. How would widespread gun ownership help with the government being subverted by oligarchs? Vigilante justice? Wouldn't change anything for the better. More paranoid billionaires would just start running things from private compounds and transit between in private jets and armored convoys. That's not a far step from what many do now, and would be extremely powerful social pretext for authoritarian crackdown.