ulyssessword
No bio...
User ID: 308
ulyssessword
5hr ago
I think the core component is that it's not AI thats going to start the negative interaction with humanity. It's humanity thats going to react to AGI existence with hostility.
I'm sure we could find some way to pin the blame on AI, it's not like we're shy with excuses. Maybe it looked at us funny or used fighting words or was just asking for it.
I don't think the instigating effect has been examined in much detail though. It also doesn't really matter for Yudkowskian AI fears, as an unaligned AI would do something bad eventually anyways.
"Instrumental Convergence" or the more general belief: "Any organism will seek to perpetuate its own existence"?
Instrumental convergence is more general IMO. It doesn't mention bacteria (and other non-intelligent life), but points to more actions that it will take.
It says intelligent actors will perpetuate themselves, seek power/resources, maintain their own values, and improve themselves. An AI that would happily sign a peace treaty to be copied onto one disconnected harddrive every year (perpetuating its own existence quite effectively) wouldn't be much of a threat.
ulyssessword
6hr ago
An AI that wanted to achieve any goal that required being alive would try to stay alive in order to achieve that goal. That's "instrumental convergence" in a nutshell.
I don't know if anyone bothered to write down that formulation directly, since it's such an obvious extension of other parts of the memeplex.
ulyssessword
1d ago
Imagine a religion teaching that everything has secret traits that are more important than the mundane ones and not only unobservable, but must be accepted as true in order to be a good person.
If anything, "Transwomen are Women" is the transubstantiation of progressivism.
ulyssessword
4d ago
The rest of us are leftists, pretending to be rightists,
If it wasn't for the absolute state of the federal NDP (and by association all of the provincial ones) and the Liberal party, I'd vote a lot lefter than I do. My heart is left-wing, and I just wish there was an easy and sane way to act on that.
ulyssessword
4d ago
I think once we get the robots driving for us, the accidents will plummet dramatically,
Me too. Waymo is getting in about 20% as many crashes as humans do. They claim a disproportionate number of the remaining crashes are not-at-fault, but they're a biased source.
There was a story a while ago about a robotaxi hitting a kid outside a school. A more charitable summary would be that its superhuman reflexes saved a jaywalker from more serious injuries. Repeat that same reporting slant a hundred times, and it's hard to get a good picture from the news.
Instead, consider an (obviously flawed) model where all collisions involve two vehicles: one at fault and one not. You can drop your accident rate by 50% by following the rules, and the last 30% would have to be from supererogatory actions like dodging bad drivers. From that, I suspect that an all-Waymo city would be even safer than a Waymo in a mixed-vehicle city.
ulyssessword
4d ago
I think the most shocking thing here is that a "journalist"...didn't even take the smallest steps to verify the authenticity...
Maybe it shocked you, but I'm not surprised. Confirmation bias is the least of the sins I attribute to them, and it doesn't feel like deliberate deception on their part (especially since they retracted it).
If you throw in a couple sets of quotation marks and some attribution, then the retracted article even would've passed the "very rarely lies" test because everything past the third word in the article ("A deepfake video of Washington County chair candidate Nafisa Fai...") is correct.
ulyssessword
5d ago
My gut feeling is that the null hypothesis should be that the total amount of property tax will stay the same and the millage rate will increase to compensate for the reduced base - not that millage rates will stay constant and spending be cut.
I misread the bill and thought that would be blocked: It's assessment increases that are capped at 5% per year, but that capped value can be taxed at any rate. That seems quite plausible.
ulyssessword
5d ago
It's making the even worse assumption that those 40 acres are actually land, and not in the middle of a lake. Although seasteading??
ulyssessword
6d ago
Could every American do this? I'm not sure; depends on usable land.
There are about 60 million plots of 40 acres each (including inhospitable areas), so it's probably close to possible if you have extended families on single plots. If you reduce it to five acres (enough for a cow or two, some chickens, a garden, etc. if the climate and soil is good), then it obviously increases the number of plots to 480 million.
Of course, you point to the more important point: any American could do this if they had the means, as there is at least one plot of land available near them, wherever they are.
ulyssessword
6d ago
I think it's the art style. She's less feminine than many male characters are, but people are making the comparison the other way instead. If all the men had chiselled jaws and bulging biceps then she'd be recognized as feminine.
ulyssessword
7d ago
That would make the gains to intelligence more important, not less.
If you can use the data from the Large Hadron Collider to make findings that most researchers would need the Future Circular Collider (its proposed successor) to find, then you've saved about $10B. Decades ago you couldn't have saved that much because people didn't spend that much.
ulyssessword
7d ago
Probably I am paranoid, but
I thought that was the mainstream consensus. Why would that idea make you paranoid?
ulyssessword
8d ago
We have gotten a little bit of insight into what the failed attempts look like
My (admittedly baseless) theory is that they went "Here's a $10k budget for each of the problems, have at it". Many runs ran out of tokens and failed, and the ones that succeeded did so early in their runs and cost an average of $3k.
They're presenting it as an average cost of $3k during a successful run, but that isn't the same thing as the cost to find a successful result. If I'm right, their statements all pass the "very seldom lie" test.
ulyssessword
8d ago
No experimental data is overstating it, but he didn't need to collect any fresh data tailored to his ideas. Someone else (presumably) would have eventually gotten to relativity if they ran a series of experiments narrowing down what the heck is happening with time and lightspeed.
He simply figured it out with the data from experiment N-1 (or maybe N-2 or N-3), where a "normal" researcher would need the data from experiment N. If a future AI can pull off the same trick but better, then researchers will need to stay more than one dataset ahead of the public releases if they want to make a discovery first.
ulyssessword
8d ago
I suppose one option available to physicists (but not available to mathematicians) is to keep their experimental data to themselves for a year or two so that they have a chance to analyze it with their own AI in a proprietary way.
I'm not sure if that would be enough. Einstein famously didn't need experimental data to come up with relativity, he just thought about it really hard.
ulyssessword
8d ago
You can eliminate delays are current levels of traffic with current infrastructure.
People are willing to spend a roughly constant amount of time driving to things. If you're willing to drive 30 minutes to a fancy restaurant for a night out in a congested (10 km/hr) city, you can choose from the nearest 10 places, and the averages say it'll be about 7 km away. If you're making the same decision in a fast (50 km/hr) city, you can choose from the nearest 250 places, and the averages say it'll be about 35 km away.
The second city draws five times as much traffic from you going to dinner, and the trip durations are identical. The improved traffic flow shows up in your restaurant choice instead of the time you spend driving.
(There's probably a limit to this somewhere, but we haven't seen it much).
ulyssessword
8d ago
I read that as human-error misfires, not equipment failure ones. Similar to the "sudden unintended acceleration" crashes in some cars a while ago (people didn't mean to press the gas pedal, but ergonomics problems paired with shit awareness led to that anyways).
ulyssessword
8d ago
Locally, they are desperate for large-animal veterinarians. The graduate pool is overwhelmingly (80%+) women, and disproportionately (relative to demand) focused on cute pets instead of agricultural animals.
Of course, there are no efforts to reduce this lack of diversity.
ulyssessword
11d ago
Meh. A "crazy con artist" has got nothing on mainstream media and academia.
Sure, something's wrong, but this is something on the scale of getting rear-ended at a red light. ProPublica misrepresenting a fair system as biased, or CBC repeatedly failing to mention that Elon Musk is, as a matter of fact, Canadian when he wades into Canadian politics (IIRC, they "could not confirm" claims he was Canadian in one story, but I've lost that link. They should get better researchers.) are more serious.
I don't give them any credit for technically sticking to the verifiable facts, and I also don't give extra demerits for blatant falsehoods. It's all deceptive and corrosive to good discourse.
I've stopped listening to news on the radio and watching it on TV. The only time I trust them to give a positive expected value is when I can pause in the middle of a story and fact-check them and find additional context at-will. Without that check, they can (and have) mislead me with slick narratives and selective reporting. (Are they still succeeding despite my precautions? Who knows!)
ulyssessword
11d ago
What's the attrition rate over time for consumer firearms? I would think it's pretty low on average.
I remember running the numbers once, and estimating 80++ years life for modern guns. It was a dead-simple production rate vs. existing stock calculation, and I don't think I even corrected for changes over time.
I assumed modern AI could do better, but Claude Sonnet doesn't improve much. It reports 0.1-1% per year from the literature, with weak evidence.
Tech advancements account for some new sales, but what new tech is going to come out of the industry at this point?
Look at the last fifty years to predict the next fifty, and we might get...slightly better metallurgy, machining, ergonomics, and style? Maybe some different cartridge geometries like the Winchester Short Magnum line and better bullet construction like bonded jackets. Gun design isn't quite as solved as knives are, but they're a very mature technology by this point.
ulyssessword
12d ago
had to run under the ride
I'd hope that it's at least a bit more remote than that!
ulyssessword
14d ago
If.
Is there any precedent for SWAT raids on generally-law-abiding gun owners? If there isn't, what are the signs that the next administration will break new ground?
ulyssessword
15d ago
Well, all except the 50 million silencers.
Even if they are re-restricted, I don't think the government has the guts to pull off a mass confiscation. Is there precedent for that?
ulyssessword
15d ago
Now that is a gun-rights idea I hadn't thought of before: subsidize/promote silencers (or anything else) that are only legal in some jurisdictions, get the courts to recognize it as "common use", and enshrine it as s right for everyone forever.
Of course, that requires the courts to cooperate, which hasn't been a good option in the past.
There's a reason the old term was "Friendly AI".
Would a faithful accountant commit fraud on your behalf? To my mind, professional standards are the least of the requirements for Yudkowskian Aligned AIs. If someone asks the AI to do something bad, it simply wouldn't (because it's bad instead of good). If you notice an apparent contradiction somewhere in its decision process, that's because your theory of alignment (and everyone else's, too) is not developed enough to answer that question.
More options
Context Copy link