How do you guys think about AI-ruin-reducing actions?
Most of the time, I trust my intuitive inner-sim much more than symbolic reasoning, and use it to sanity check my actions. I'll come up with some plan, verify that it doesn't break any obvious rules, then pass it to my black-box-inner-sim, conditioning on my views on AI risk being basically correct, and my black-box-inner-sim returns "You die".
Now the obvious interpretation is that we are going to die, which is fine from an epistemic perspective. Unfortunately, it makes it very difficult to properly think about positive-EV actions. I can run my black-box-inner-sim with queries like "How much honour/dignity/virtue will I die with?" but I don't think this query is properly converting tiny amounts of +EV into honour/dignity/virtue.
How much of this is because coding rewards are more like pass/fail rather than real valued reward functions? Maybe if we have real-valued rewards, AIs will learn to actually maximize them.
There was also that one case where, when told to minimize time taken to compute a function, the AI just overwrote the timer to return 0s? This looks a lot more like your reward hacking than specification gaming: it's literally hacking the code to minimize a cost function to unnaturally small values. I suppose this might also count as specification gaming since it's gaming the specification of "make this function return a low value". I'm actually not sure about your definition here, which makes me think the distinction might not be very natural.
I think this definitely can happen. The Liberal Democrats in the UK (third party, colloquially the Lib Dems) gained support in the 2010 election by pledging not to vote for increases in tuition fees for UK universities. They were able to form a coalition government with the Conservative party after the 2015 election. The Lib Dems hadn't been in power before (though their predecessor, the Liberal party, hadn't been in power since the 1920s).
The coalition government increased tuition fees by a factor of 3. There were widespread protests at universities, mostly peaceful though one bloke did lob a fire extinguisher off a roof. The Liberal Democrats never recovered, going from 57 seats to 8 at the next election in 2015. I was a member of the party around 2017, and many people I spoke to said "I used to vote for them but then they increased tuition fees".
This doesn't prove a causal link (people might have just been annoyed at the coalition government's austerity policies, and used tuition fees as a totem of that) but people do literally say "I liked Nick Clegg but he betrayed his position on tuition fees and I'll never vote Lib Dem again."
UK council tax is based on overall property value, not unimproved land value.
Did they do whole properties or was there an even weirder system where one bureaucrat does the land and the other does the property?
Strongly agree. Sell-price tax with forced sales sounds like something a cryptocurrency would implement. It might work there, since if a malicious bidder tried to buy your TOKEN at above-market price, you could automatically buy a new one within the same block, at actual-market price. This could also work for fungible but rapidly-transferrable assets like cloud GPU time.
If taxing physical goods (like infrastructure or even land) which is where a lot of value in the world lies, this does just open up companies for extortion. E.g. what if I demand to buy one square inch of land under Bill Gates' house, then demand he remodel his house around the square inch. Or suppose his summer house is in the middle of the Pennsylvania woodland, where the land is actually quite cheap, so he's either forced to pay extremely high tax on his land for, or he's open to extortion in the same way.
I think implementing a Georgist land tax basically requires that we trust some government bureaucrats to approximately accurately determine the value of plots of land. This isn't an unreasonable level of trust for a first-world country; in the UK we trust government bureaucrats to draw the boundaries for our election districts, and that works alright.
That's a good point, given basically every respiratory illness you'd come across in the first world is viral.
If everybody is wearing clothes (which I expect is the case for at least 2/3 of the events organized by LessWrong users) then UV exposure will be limited to face, neck, hands, arms, and lower legs.
I expect that hands, neck, arms, and legs will be rapidly re-colonized by bacteria from the torso, upper legs, feet, etc, just from normal walking around. The face is the main area I'd be worried about, since I'd expect it to have a slightly different microbiome than the rest of the skin (I think it's oilier, hence acne) and it's going to be pretty maximally exposed to the UV light. Having thought about the problems I'm less worried than I was before.
I'd keep a small eye out for acne/eczema/dry skin on people's faces after being exposed to this, just in case.
(Of course the ideal method is to have the UVC light internal to your air conditioner/heater unit, which is already circulating the air, so you can blast everything passing through that with enough UVC to annihilate any and all pathogens in the air, but that requires retro-fitting to AC units and stuff. Still, would be cool to see Aerolamp partner with some AC/heater company in the future.)
I donated last year and I'll donate again this year. I do hope to get to visit Lighthaven at some point before it/the world ends. It's likely that if Lighthaven continues to exist for another year I'll be able to visit it. I would be extremely sad if LessWrong stops existing as a forum, though I think the online community would persist somewhere (Substack?) albeit in a diminished form.
Is there a quantification of the effect of this on skin microbiome as of now? I would not like to kill all of the bacteria on my skin.
MondSemmel is correct but if you don't want to use the menu, type "> " at the start of a new line and it will begin a quote block (you can also use >! for spoiler tags).
This is the place where I can most reliably just communicate. Basically anywhere else I either have to limit myself to particularly simple thoughts (e.g. Twitter) or to spend time extensively running ideas past people in order to figure out what context they're lacking or why my explanations aren't working (e.g. Real Life).
Here I have a decent chance of just sitting down and writing a 1-2k word blogpost which gets my point across successfully in one shot, without needing further questions.