@VoxelVexillologist's banner p

VoxelVexillologist

πŸ‡ΊπŸ‡Έ Multidimensional Radical Centrist

1 follower   follows 0 users  
joined 2022 September 04 18:24:54 UTC

				

User ID: 64

VoxelVexillologist

πŸ‡ΊπŸ‡Έ Multidimensional Radical Centrist

1 follower   follows 0 users   joined 2022 September 04 18:24:54 UTC

					

No bio...


					

User ID: 64

How does one waste four months of investigation on this?

I'm not familiar with how law enforcement works here. Does "four months of investigation" imply more than "this was theoretically on someone's desk for four months before officially closing the case, but actual time may or may not have been spent on it"?

Imagine if we'd held a plebiscite on whether universities should continue spending money on this "Internet" thing in friggin' 1973

If it helps, if you held this retroactively today, I'm still not sure how I'd personally vote. I get a lot out of the Internet, but I do sometimes wonder if it's worth all the downsides.

Partly it's happening, just slower than you think

This is always the case with technology: it can take a surprising amount of time to replace existing infrastructure even when new technology is clearly better. Existing equipment was expensive and still works, teaching staff new workflows has overhead, and the impetus to actually change things is often low. OPM was still processing paper records for employee retirement until very recently. How many industrial machines are still running DOS out there?

I can imagine AI being both transformative and seemingly-ubiquitous, but still not reaching all the dark corners of the workforce for a generation or two.

You may grumble at an auto factory, but you do, at least, understand why cars are necessary. Even if you personally don't own a car and rely on public transport to get everywhere, it's obvious that there's real public demand for cars, on the basis of their obvious usefulness. You might not want a factory near you, but you understand why there need to be factories somewhere. LLMs have not cleared that hurdle.

I don't think you're completely wrong here, but in this case the comparison to make if you think AI will be extremely useful would be if the anti-car urbanists started showing up to protest your car factory. Sure, you might not want it in your neighborhood, but you buy a new (to you, at least) car every so often and don't want them to become scarce resources, and people show up to protest every single car factory on the basis of "induced demand" traffic claims, "just walk, bruh", accident statistics, and a demand that you commute by never-to-be-finished passenger rail instead. They are convinced cars have no real use, and it'd be rather frustrating to normies if they doubled car prices without delivering any of the promised better alternatives first.

How to tell which framing is more relevant seems difficult from inside the situation.

Texas is an interesting case there because ERCOT has long-term (2032) forecasts expecting peak grid demand to rise from 91GW (last month) to 370GW. More than 4x! I'm frankly not sure where they expect energy to come from, but solar (and battery storage), wind, and natural gas have been growing pretty fast for years now. But probably more important is that someone is at least thinking about the distribution infrastructure to do so.

IIRC YouTube spends a surprising amount of processing power transcoding uploaded video, to the point where Google has developed custom ASIC hardware to do so.

I guess I don't know how AWS lays out their physical locations, but "all EC2 hardware in one building" would extremely computation heavy, whereas S3 would not be. But perhaps they're intermingled for some reason. AWS and Google also have a lot more institutional experience building these things, which probably smooths out some of the wrinkles.

Surely there's a town with a data center they can go ask how good it was for them?

Are these new datacenters really that different from the ones we've been building for literally decades? AWS and Google have had giant server farms longer than anyone was really looking at machine learning like today. Heck, I remember people complaining about the NSA's data center in Utah slurping up all our emails, not its water usage. Complaining about this seems a new idea, but I don't recall too many complaints about the facilities previously. On the other hand, maybe it's like fracking, which wasn't new a decade ago, but really took off at scale.

On the gripping hand, I'm really not sure who's paying for all these new data centers, how they intend to pay the bills if AI tokens are fully commoditized, or why they're being put close to neighborhoods. AWS has existed in multiple locations (they don't disclose exact addresses, but it's probably not a single building) for a long time, and doesn't seem to have garnered quite the same level of complaints.

You have to be rich to afford a nice enough place to cook for yourself in New York. Even giving away food for free will not solve this problem.

I seem to recall reading that expectations of home cooking vary a lot, and that in big Asian cities (Singapore, Taipei, Tokyo) in particular home cooking isn't the norm. But those cities have relatively affordable, quick restaurant options. But last I checked, that isn't really an affordable option in NYC (uncertain about Europe). It's at least worth considering if that could be an option in the West.

But in America, complaints about the rule of law are strongly correlated with β€œsilent majority” populism.

I think there is a strain of populist thought, which I don't think is completely wrong, that technocratic liberalism as-implemented has so overweighted seemingly sympathetic cases as to be obviously, if indirectly, negative to the majority. "Harm reduction" sounds like a nice idea, but in practice means giving addicts clean needles to use and leave on playgrounds, and while it might reduce acute harm to addicts is not reducing harm to playing kids: the diffuse cost was ignored or swept aside. Asylum laws sound nice and probably do help the people claiming it, but the state seems to assume the side effects on the rest of us are negligible: if you believe that I have an election to sell you. Ditto for soft-on-crime policies and police abolition. Across the board bureaucrats seem to have a clear blind spot for the diffuse side effects of policies on the rest of us outside of a few groups that get special hyper-focus on disparate impacts and treatments.

There are areas of law which are technical and complex for good reasons

The right to a jury trial can be waived in the US system, which I understand is not uncommon in areas where lay jurors' understanding could make results unpredictable or unsympathetic. But it's a choice the defending party gets to make (well, there is also arbitration law, but that's separate).

And using radioactive decay is excessive and unnecessary.

There are very real concerns about even seemingly-good sources of hardware randomness. There remain reasons why Linux maintains CONFIG_RANDOM_TRUST_CPU: if your concern is nation-state level actors, you might not trust Intel's pinky swear that the random bits it generated have nothing up its sleeve (see DUAL_EC_DRBG for the sorts of concerns you'd want to be aware of).

Radioactive decay is at least something you can build relatively easily that has no obvious metaphorical sleeves to hide weaknesses. But you're probably right it's overkill for most users.

Key generated from low entropy PRNG" is a failure mode we had before.

The rather infamous Debian SSH key generation vulnerability was like this, and was way back in 2008. Ironically, that one was introduced by following the guidance of an automated security checking tool, and impacted Debian and related distributions.

I think you're right that legal activists aren't really limited by ideas today. But as an example, how will the APA notice and comment regime survive AI-generated slop comments? My understanding is that a lot of it today ends up as boilerplate copy-and-paste letter campaigns: if every commenter writes a unique letter, review will presumably be more involved. Or actively using AI to generate as many novel arguments for/against a proposed rule as possible to exhaust the review process. I'm sure the government side will start using AI to review comments, but I wouldn't expect them to be ahead of the public on their processes.

It looks like Morocco is in the US camp

Morocco was the first country to recognize the United States of America (1777), and has the longest continuous treaty relationship with the US (since 1836). Of Arabic-speaking nations, it's historically one of the most US-friendly in international relations.

The future of the right may not be building a vast apparatus of company men, but tightening the OODA loop, getting faster and leaner, and doing more with fewer people.

I have been wondering: for all the drama going on with frontier AI models in the world of digital security, I have to think there are some interesting red-team/blue-team ways to apply those to laws and regulations. Some of the big lawfare shifts we've seen in the last decade or so ("did you know we could just have all the border crossers say the shibboleth asylum and we have to let them in indefinitely until we can give them a hearing?") seem to have come from novel readings or applications of existing rules. It'd be at least interesting (and I haven't tried) to prompt those same models for what corners of the tax code could be abused --- or in reverse, to provide feedback to Congress and regulatory rulemakers --- to promote less-direct outcomes. It's not hard to imagine "hey Claude, let's figure out the most pathological case to bring to trial to make SCOTUS rule my desired direction" actually working.

How this same attitude to separation would justify an invasion of.. oh lets say Hawaii is left unsaid.

Isn't this basically the same argument Argentina has made for Las Malvinas The Falklands?

Asking for a friend. Well, literally, asking for advice in helping out a friend, although I am being a bit loose on the details for privacy reasons.

I have a longstanding friend from childhood who is still part of my regular friend group. We play games regularly, and despite having scattered to the winds after high school, we still see each other a few times a year in person. The rest of us have gotten jobs and established ourselves, but this friend is probably best described as somewhat Bohemian in comparison. I know he's used alcohol (not unusual) and marijuana (I'm less familiar with, personally) off and on, but nothing too much harder to my knowledge.

There have been a couple of minor incidents in the last couple years that I've considered cause for concern. I know he's been checked into a hospital for mental health reasons (I haven't asked for specific details), and I've heard him make some comments that seem, at least from my perspective, somewhat paranoid: it's not impossible he rubs people the wrong way sometimes, but I don't think a bartender somewhere he's not a regular has it out for him.

Given my own (IMO quite fortunate) lack of experience with modern mental health care, what should I be doing as a friend to encourage positive outcomes here? This isn't an acute concern with a specific incident, just a recognition that I'm pretty far outside of my expertise here and I value my friends' long-term success and happiness.

Wojtek served in the Western Polish forces, not the Red Army. Although that's probably not as good a story.

I didn't think the OG markdown required a blank line between list items. But there are lots of variants: even Reddit doesn't render it consistently across its frontends.

Yes, I meant to say the destructive ones are faster.

It does seem odd that a large fraction of recent top level posts have resulted in bans. I don't think we have a good balance of required effort and culture warring. Or it's a weird tradition to disallow users that start conversations from participating in them.

One of my biggest criticisms of HTML and Markdown is how it renders h1 and the other headings. I really wish it could scale based on the depth of headings, such that a lone h1 rendered as the smallest heading, only getting larger when there were subheadings.

Now I wonder if CSS could do that...

Information should be free! except for distilling.

I largely expect that US courts will eventually find model weights to be uncopyrightable: they're the product of an expensive "sweat of the brow" operation like a phone book, and purely functional in nature. The information in phone books isn't covered, nor are purely functional inventions. The process just doesn't strike me as creative at all.

Although I wouldn't bet huge money on it, and there is a nonzero chance Congress decides to modify the law in this instance.

So even though Anthropic didn't have to destroy their books, at least they apparently haven't destroyed anything truly rare.

I'm pretty sure the non-destructive scanners are faster and cheaper, and I'm not sure the expense is justified for scanning mass-market "slop" books. The "rare" books they're trying to invoke here (first printings, etc) are almost exclusively out of copyright (and this likely already scanned and available on Project Gutenberg), and I doubt they're paying extra for anything that's currently in print.

Destroying books for many evokes (or is even the same as) book-burning. And many consider destroying books to be one of the greatest evils imaginable.

Are you familiar with how libraries dispose of old, unread, and poor-condition texts already? They aren't usually building new storage space to go with the new books they buy regularly. Unsold books at bookstores have historically been "stripped" (cover removed and returned to publisher for refund) and destroyed. I can appreciate a general reverence for books, but there was no shortage of books getting recycled before AI.

At least some people are using distillation or reverse osmosis for desalination. I'd do that too if my only water supply was the ocean.