All of Dirichlet-to-Neumann's Comments + Replies

Open Thread - Jan 2022 [Vote Experiment!]

I'd be interested although I'm in France, UCT+1 so it may be a bit difficult to arrange a meeting twice a day.
I'm a PhD student in mathematics.

Open Thread - Jan 2022 [Vote Experiment!]

I kind of feel like there should be a funny/not funny axis. Sometimes I read a good joke or a fun take in a comment, and I would like to signal I liked it, but the overall karma does not seems like a good way to signal that.

Also true and hits the mark do not seem orthogonal to me. Can something be false and still get the point ?
 

How Bad Is QWERTY, Really? A Review of the Literature, such as It Is

Anecdotally, French users use an Azerty setup, which is very similar to Qwerty for most letters, but has more room for all those nice diacritics like é, ù, à, è, ê... Typing French on a qwerty keyboard is really super annoying.
However in the scramble for room in the keyboard, many symbols like the full stop, the numbers, and things like |, \, [, { are locked behind the shift or alt keys. It really makes it slower to type in LaTex (and from my experience increase the risks of RSI too, but that's probably mainly due to my awful desk set-up).
Mac's Azerty set-... (read more)

Uncontroversially good legislation

I think in most timelines this ends up as just an added layer of bureaucracy, but in the few of them when it does not it's great.

 

Uncontroversially good legislation

I feel unreasonably validated by this too to be honest.

How to think about and deal with OpenAI

Trust between partners do not happen overnight. You don't suddenly begin sharing information with concurrents when the prize is in sight. We need a history of shared information to build upon, and now - when, as you said, AGI is not really close - is the good time to build it.
Because if you don't trust someone with GPT-3, you are certainly not going to trust them with an AGI.

 

2Daniel Kokotajlo3dChoosing to not release GPT-3's weights to the whole world doesn't imply that you don't trust DeepMind or Anthropic or whoever. It just implies that there exists at least one person in the world you don't trust. I agree that releasing everything publicly would make it easier/more likely to release crucial things to key competitors when the time comes. Alas, the harms are big enough to outweigh this benefit, I think.
Has anyone had weird experiences with Alcor?

It seems to me a simple solution for your problem would be to allow preservation in any case, and then let the decision to put you into long-term storage into the hands of your wife, when she has the time to decide what to do.
While you would need to eat up the cost of the cryopreservation part, you may be able to recover the money that goes into long term storage ? 

Uncontroversially good legislation

I think this one is totally uncontroversial and I'm voting for whoever has this in their platform.

2Randomized, Controlled5dI feel unreasonably validated by this
Uncontroversially good legislation

They would not necessarily fight its creation - although a department those aim is to fight bureaucracy may very well seem threatening to people in position of power in the diverse bureaucratic departments. But they would definitely fight its action, i.e. any attempt at changing the way they run their services. 

 

2ChristianKl5dI do agree that the job of the department won't be easy. To the extent that you can create such a department without too much resistance, I think it makes sense to count the idea of creating the department as "uncontroversially good legislation".
Uncontroversially good legislation

The most powerful of all : the bureaucrats themselves. Also Moloch and entropy, so we are fighting powerful forces here.

2ChristianKl7dThe new department would employ some bureaucrats, it's not clear to me why existing bureaucrats would feel a need to fight its creation. That is why it's a job that needs a specialized team to combat it and can't just be done effectively in the existing structures.
How to think about and deal with OpenAI

It seems to me that being open about what you are working on, and having a proven record of publishing/sharing critical informations, including weights, is a very good way to fight the arm race.

If you don't know where your concurrent are, it is much more difficult to stop to think about alignment than to rush toward capacity first. If you know where your concurrent are, and if you know that you will be at worst a couple weeks or months late because they always publish and you will thus be able to catch up, you have much more slack to pursue alignment (or s... (read more)

Yes, when we are getting really close to AGI it will be good for the leading contenders to share info with each other. Even then it won't be a good idea for the leading contenders to publish publicly, because then there'll be way more contenders! And now, when we are not really close to AGI, public publication accelerates research in general and thus shortens timelines, while also bringing more actors into the race.

Uncontroversially good legislation

None of the answer posted yet seems uncontroversial to me, except maybe the one on price transparency.

1Randomized, Controlled5dI posted mine [https://www.lesswrong.com/posts/9fK9jSgTWJhcGYjbh/uncontroversially-good-legislation?commentId=cKCu5sjrkNSTt4EMf] a day after this comment, do you still feel that way?
Why did Europe conquer the world?

The main point of Diamond books seemed to me to be that European dominated the world because of sheer geographic luck and emphatically NOT because they were more intelligent than Asian, American or African. Accusing him of latent racism seems really disingenuous to me.

Why did Europe conquer the world?

Europe 1300 A.D. was more advanced both technologically and scientifically than the Roman empire in a lot of ways. Europe 1000 AD is more debatable and Europe 700 AD surely was not more advanced.

 

Experiments with a random clock

I've wanted to test that for a long time !

Sadly I can't stand to wear a wrist watch...

"Acquisition of Chess Knowledge in AlphaZero": probing AZ over time

Although the technical details are way to difficult for me, as a chess player I found the article really interesting. 
When it was first release, AlphaZero seemed to play more human-like than traditional engine such as Stockfish. Do your analysis support this conclusion ?

What would we do if alignment were futile?

If you are really convinced that 
1) AGI is coming really fast.
2) Work on alignment has basically no chances to break through in time.
3) Un aligned AGI result in quick and complete annihilation for humankind
4) You firmly believe in a utilitarism/consequentialism. 
which seems to me to be Eliezer's model,
then you should focus your efforts to launch an all out nuclear war between the USA and China, which would be very unlikely to destroy humanity.
You could even move MIRI to New-Zealand or something so that work on alignment can continue after the nuclear blast.

4Jackson Wagner2moSee the similar comment here [https://www.lesswrong.com/posts/CpvyhFy9WvCNsifkY/discussion-with-eliezer-yudkowsky-on-agi-interventions?commentId=QDX2bRmKvGxgpJcKN] . Personally, I think that we can do better than starting a nuclear war (which, after all, just delays the problem, and probably leaves civilization in an even WORSE place to solve alignment when the problem eventually rears its head again -- although your idea about disaster-proofing MIRI and other AI safety orgs is interesting), as I said in a reply to that comment. Trying to reduce Earth's supply of compute (including through military means), and do other things to slow down the field of AI (up to and including the kind of stuff that we'd need to do to stop the proliferation of Nick Bostrom's "easy nukes" [https://www.lesswrong.com/posts/ioueJyKE3CaHQtZHk/nuclear-strategy-in-a-semi-vulnerable-world] ) seems promising. Then with the extra time that buys, we can make differential progress in other areas: * Alignment research, including searching for whole new AGI paradigms that are easier to align. * Human enhancement via genetic engineering, BCIs, brain emulation, cloning John Von Neumann, or whatever. * Better governance tech (prediction markets, voting systems, etc), so that the world can be governed more wisely on issues of AI risk and everything else. But just as I said in that comment thread, "I'm not sure if MIRI / LessWrong / etc want to encourage lots of public speculation about potentially divisive AGI 'nonpharmaceutical interventions' like fomenting nuclear war. I think it's an understandably sensitive area, which people would prefer to discuss privately."
-5CronoDAS2mo
What if we should use more energy, not less?

"The decline in energy use per capita started in the 70s and largely coincides with the spread of the counter-cultures of the 60s, including the green movement. Of course, not all regulations are bad or unnecessary, but it's not very clear that strict regulation actually makes us much safer and healthier overall, and in fact, the opposite may be true:"

 

Or maybe, you know, there happened to be an oil crisis in 1973 when oil prices increased by 400%, and energy usage per capita has declined ever since because energy went from super cheap to increasingly expensive ?

Just take out coal/oil and a stable technological level seems possible. Also I'm not sure those stable fantasy worlds really exists in literature, most examples I can think of have (sometimes magical) technological growth or decline.

Tolkien Middle Earth is very young - a few thousands years. This means no coal, no oil, and no possibility of an industrial revolution. Technology would still slowly progress to 18th century level but I can see it happening slow enough to make the state of technology we see in the LOTR acceptable. On the other hand magical tech... (read more)

How do you learn to take more beautiful pictures with a camera?

I'd say that the first step is to work on composition/subjects. 

Professional have better materiel and skills, and there is also a lot of work editing the pictures, but if your subject is not interesting and your composition is boring the pictures will be bad no matter what. 

Look at pictures or paintings* that you like and try to understand how the different objects are placed - the strength lines they form, where they are in the picture...

When you take a photo of a foreground subject (like a person, an animal), try to have a nice background (and ... (read more)

I wanted to interview Eliezer Yudkowsky but he's busy so I simulated him instead

I'm putting this here rather than in the collapsed thread, but I really think the initial post (before the edit) was at the very least careless. There is a widespread habit in tech publications, especially in IA, to pretend results are better than what they actually are - I would hope that Lesswrong, with its commitment to truth-seeking and distrust of medias, would do better...

So, the edit says "However the Yudkowsky lines were also cherry picked. I ran several iterations, sometimes modifying my prompts, until I got good responses.". So, how were they che... (read more)

7Ben Pace4moThis is true inasmuch as posts written with help from GPT-3 are meant to be evidence about the capabilities of GPT-3. Sometimes posts are primarily intended to be fun, and success is measured by how fun they are, and then I don't care how much iteration you put into it, I just want it to be fun. I guess this was a combo, because it's about simulation? So your question is reasonable. FYI from having played with GPT-3 myself, I assumed something like Lsusr had run multipl (3-15) iterations and partial iterations on each segment and thrown bits out and thrown whole other segments out. That said it was probably clearer to me because I've written with GPT-3 myself, and someone who hasn't could've been under the impression this was just the first pass.
9lsusr4moI wasn't counting. Rarely more than 10. Sometimes the first answer just worked. Never did I come anywhere close to 100. More important than throwing out answers was how often I changed prompts. Some prompts prompt much better input than others. No. Though I did sometimes keep half an answer I liked and then autocompleted from there instead of rerolling the dice from scratch. Sometimes the answer kept going and going and I truncated it early. There were lots of edge cases. In one instance, Robin Hanson butted in, which I edited out. I didn't keep the answers I threw out. The uncurated answers were created specially for that comment.
The Myth of the Myth of the Lone Genius

I had a look at my copy of Simon Singh's "Fermat's last theorem" (amazing book by the way) and three things are pretty clear :

  1. Wiles' proof makes extensive use of papers published in the years 1986-1993 while he was working on his proof, so he was certainly not isolated during this time.

  2. he was unable to find the error in his proof, and had help from Taylor to correct the error.

So if he had not been so obsessed with "being the one to prove Fermat's last theorem", the proof would have been finished a couple years sooner.

So yeah, the lonely genius is a myth and a dangerous one. Long live the collaborative genius who works with his fellow geniuses !

The Myth of the Myth of the Lone Genius

Perelman had ethical objections to scientific prizes, but during his productive period he was not particularly solitary and was involved in the normal academic live - although he was on the lower end of academic interaction.

The Myth of the Myth of the Lone Genius

Ramanujan and Galois are textbook exemples of mathematicians who would have had a much bigger impact if they had not been as isolated. (And Ramanujan most productive period was when he was working with Hardy and Littlewood).

Also, Erdos.

Re: Competent Elites

Also coming up with a clever idea is much more difficult than evaluating if a clever idea is good. For example It's hard to find a proof for a theorem, but easy to check if a proof is correct. Likewise you can evaluate someone's intelligence even if he is way more intelligent than you.

Don't feel bad about not knowing basic things

It was, a little. Truth is I know the basic definition but I've yet to build up enough knowledge and intuition around them to really use them in my research. Thinks analytic vs bounded semigroup, L^\infty calculus, angular sectors and so on.

What's your peace plan?

The Holy Sepulchre has been administered by a Muslim family for centuries because the Ottomans empire was fed up with the different Christian denominations endless fights to pray in the sanctuary.

What's your peace plan?

"I don't know to what degree this represents their current views of Hamas. But whatever Israeli Jews think of Fatah, I believe they would much rather see Fatah in control of Palestine than Hamas. Any external organization facilitating negotiations would be wise to account for this. "

That's probably part of the problem too. If I remember correctly Hamas won the elections a decade or so ago and Israel+the Western governments helped Fatah to stay in power... meaning that right now there is not even a real government to negotiate with.

What's your peace plan?

It's exactly the problem, and the fact that the Hamas started the war just as Netanyahu was losing his grip to power is highly unlikely to be a coincidence. 

The fact that Israel lead the best vaccine effort in the world but refused the easy PR move to give vaccine to Palestine is also significant.

[Prediction] What war between the USA and China would look like in 2050

I'm not sure the victims of, say, Pinochet really appreciate the "we are not expansionist, we are just helping overthrow your democratically elected government in order to maintain our influence on your county's international and economic policy." defense.  
 

5ChristianKl8moI don't think that's how most control goes. If you forbid our cigarette companies from selling in your country you will have to pay hundreds in millions for a case at the WTO or if you don't don't sell your state owned businesses you won't get IMF loans is more the dynamic I'm pointing towards. There's a reason why there are currently no Chinese villians in Hollywood movies. It's just not a smart move if you want to raise capital for another movie. If current trends continue Australia will end up in a state where you can't have a political career if you are openly critical of China. They don't need to overthrow any Australian government to get what they want.
[Prediction] What war between the USA and China would look like in 2050

A few selected thoughts :

  1. The number on the US invasion of Afghanistan may explain why it was such a failure. With 10^6 soldiers instead of 10^5 they would probably have been able to really secure the country.
  2. I think a protracted war is unlikely because I expect the US to just give up on Taiwan.
  3. Expansionism is something that grows on you. After China invades Taiwan, I guess they may become interested in securing some form of control over say Singapore, or Vietnam. The fact that they are not expansionist now does not mean they will not be expansionist then.
  4. I think a Chinese invasion of Taiwan is likely to happen way before 2050.
5ChristianKl8moThat depends on the amount of trouble you have administering the new expansions. I guess they may become interested in securing some form of control over say Singapore, or Vietnam. The US has some form of control over many states without needing to formally control their territory. China's actions in Australia or Hollywood for that matter, indicates the like having some control but that's very different then needing to incorporate everything.
Sherrinford's Shortform

My model is that what is called "left of center" in the USA is "far right, at least economically"* in Europe (and what the USA call "socialism" is "what everyone agrees with".

*"economically" does a fair bit of work here - on issues like immigration for example the left right divide is the same as in the US.

Don't feel bad about not knowing basic things

That's a good layman description - a semi group is basically the exponential of some linear operator. The problem is that I'm supposed to be a bit more than a layman.

2johnlawrenceaspden7moThen perhaps this was a little hyperbolic?
Don't feel bad about not knowing basic things

I'm three years into my PhD. Still don't really understand what a semi-group is (something something exponential). I can't remember the formula for the exposants in Sobolev inequality but here it is just my fault for never learning it correctly.

2johnlawrenceaspden8moSo the idea is that if you've got a linear differential equation, like dx/dt=2x, then all the solutions look like x=x0*exp(2t). And you could imagine that there's an operator which says "Where will this point end up in 1 second?", call it O(1), which looks like x0->x*exp(2). I.e. it just gets multiplied. If you know that operator, then you know the operator that represents "Where will the point end up in 2 seconds", because you can say "where will it end up one second later" then "where will that point end up in another second. So this gives you a law that O(2) is O(1) done twice. And O(3) is O(2) followed by O(1). So these operators are a group (the identity is 'where will it be after I run the equation for no seconds'). In the one-dimensional ODE case, this is all so trivial that it's not worth mentioning. Almost a pun. But it generalises nicely to dx/dt=Ax, where x is a vector and A is a matrix, and that gives us a way of defining exp(At), i.e., how to exponentiate a matrix, by saying, "well what if we run this matrix equation for a second and see where it puts all the vectors?" And in fact we can generalize it further, to differential equations on linear operators on infinite-dimensional spaces. Which is another way of looking at partial differential equations. (A function is an infinite dimensional vector, a vector is a function on a finite set). So you can talk by analogy about how to exponentiate the diffusion equation, and a time delay operator that takes wiggly functions to less wiggly functions. But you can't always run pdes backwards, e.g. the diffusion equation won't run in reverse, so you don't always get a full group. I think that's the intuition for semigroups, I hope that's what you were talking about! (And sorry if it's not clear or just wrong, I haven't been a mathematician for thirty years.) If you think about this for long enough, you should suddenly understand why e to the i pi is minus one. In fact it's just a really obvious thing tha
What will 2040 probably look like assuming no singularity?

True but putting them in Northern part of the world may also be a good idea. Right now looking for example at Google's data centers map there seems to be a very small trend toward northern locations (at least in Europe), but it may just be a flux due to local financial incentives being more favorable in some countries.

2ChristianKl8moGoogle seems to have one datacenter in Ireland, Denmark, Belgium, Finnland and two in the Netherlands. Those seem to be countries that currently have cheap industrial electricity: https://www.statista.com/statistics/1046605/industry-electricity-prices-european-union-country/
What will 2040 probably look like assuming no singularity?

I'm saying a war between China and the USA is a possible scenario (with <50% but >10% probability). The intensity of destruction would depend on how the war is fought, but conditionally on a drawn out war nuclear mutual destruction does not seem implausible to me.

It's hopefully not the only scenario that prevent singularity though.

What will 2040 probably look like assuming no singularity?

Well it was mostly meant as a joke. But here are my probabilities.
- China occupies Taiwan 90% - this is clearly a mid term goal of the current Chinese government.
- The USA defend Taiwan 30% (maybe 50% if the invasion happens in the near future). The USA may or may not hold their defense agreement - I think the later the invasion happens, the more likely it is that the USA just let Taiwan be invaded.

The war could be a quick win for China (they have home advantage). If it keep going for a few years, I still think China will win (the coronavirus crisis is str... (read more)

2Daniel Kokotajlo8moDoes the chinese government have a 90% success record at achieving their mid term goals? Do you have a 100% success record at judging what the chinese government thinks is a mid-term goal?
What will 2040 probably look like assuming no singularity?

Is refrigeration a big part of data centers energy costs ? This would mean the best places for solar energy are also the worst places for data centers...

2ChristianKl8moIf refrigeration becomes a major part of the energy cost it's worth noting that there's thinking about putting data centers under water where they can be cooled more easily. At the moment that's not viable but it might be in 2040.
2Daniel Kokotajlo8moI've thought a bit about this, but haven't done any calculations. My guess is that it would be overall cheaper to have datacenters in sunny regions than to try to get solar panels in cold regions. Your refrigeration (and heat management more generally) electricity bill will be higher, but not that much higher, but your electricity costs will be much lower.
What will 2040 probably look like assuming no singularity?

An all out war between China and the USA over Taiwan has crippled the whole world. 
Good news is AI alignment is not an issue anymore.

2ChristianKl8moAre you saying that's the only scenario that would prevent singularity or are you saying that it's generally a probable scenario.
2Daniel Kokotajlo8moBold claim! Perhaps you should make a post (or shortform, or even just separate answer to this question) where you lay out your reasoning & evidence? I'd be interested in that. If you think it's infohazardous, maybe just a gdoc?
Why Democracy?

There are many different ways to look at democracy - one important historical point of view is that the point of democracy is not really to allow coordination to implement policies, but rather to make it impossible to coordinate to implement (unwanted) policies. It's not a bug, it's a feature ! 
In this sense democracy is not really defined by itself, but rather in opposition to what it is not (i.e. some kind of dictature).

An other interesting point of view (probably historically the first) comes from Aristotle. I'll try to explain it but I'm not an ex... (read more)

Your Dog is Even Smarter Than You Think

If that were the case, don't you think that animals that actually run faster than humans in dense forest would be more intelligent ?

2ChristianKl8moWhich animals are you thinking about?
Sexual Dimorphism in Yudkowsky's Sequences, in Relation to My Gender Problems

I doubt this factor is really significant. A quick google search only yielded results for breast cancers, and glancing over at the abstracts I'm skeptical those studies are strong enough to prove a real effect for size (they seem to find an effect for BMI, but BMI affect more than just the number of cells in the body so it is not really significant for our discussion).

Also in the context of health benefit associated with transition this is irrelevant because transitioning will not change your body size...

2tailcalled8moSee e.g. this [https://royalsocietypublishing.org/doi/10.1098/rspb.2018.1743]. Yeah I know, I could've pointed out the body size effect to nim too.
Sexual Dimorphism in Yudkowsky's Sequences, in Relation to My Gender Problems

Cancer and heart disease kill most people but kill mainly old people, so their impact on life expectancy is less significant per death than suicide or unintended injuries.

I don't have stats but the link you give seems to be supporting my point : suicide and unintentional injuries amount to 10% of males death and at most 5% of female deaths. 
Quick back of the envelope calculation : assuming average age of death by suicide/accident for males is 30, while "all other causes" average age of death is 80, you'd get a life expectation of $0.9\times 80 + 0.1\t... (read more)

Sexual Dimorphism in Yudkowsky's Sequences, in Relation to My Gender Problems

A big part of the difference comes from human males taking part in riskier activities, having a higher suicide rate and generally a less healthy lifestyle. Those are statistically significant differences, but they don't affect individual person. If you are a male, do some reasonable physical exercise and have good food habit while avoiding things that are obviously risky, a big part of the difference vannishes.

1nim8moStats like https://www.cdc.gov/women/lcod/2017/all-races-origins/index.htm [https://www.cdc.gov/women/lcod/2017/all-races-origins/index.htm] vs https://www.cdc.gov/healthequity/lcod/men/2017/all-races-origins/index.htm [https://www.cdc.gov/healthequity/lcod/men/2017/all-races-origins/index.htm] lead me to believe that heart disease and cancer are more significant killers than risky activities or suicide. Could you share where you've found stats broken down by activity riskiness, suicide rate, and lifestyle healthiness? And even if lifestyle factors account for a big part of the difference, it sounds like you believe there's still some difference, and it seems like any difference would be relevant to someone seeking to maximize their longevity.
1tailcalled8moAlso, bigger bodies = more cells that can end up turning into tumors and such.
What books are for: a response to "Why books don't work."

Books are best seen as a way to keep knowledge than a way to share it. Well, that's what they were designed for (that's what writing is for in fact).

Recently I bought a new laptop

I'd write the conclusion a bit differently : "So thank you to all the people who pick laptops from the top of a spreadsheet and use legible metrics of laptop picking to keep manufacturers honest so I can just go to the computer store nearest to me, buy whatever they advise me, and still end up with something pretty good.

i'm pretty sure I'm the one free-riding here.

Covid 4/9: Another Vaccine Passport Objection

Did you have a look at the https://covidtracker.fr/covidtracker-world/ project ? It does not give the amount of details you used to have for the USA (as it is a French site) but it seems to do what you want country by country.

The Holy Algorithm

In practice it's pretty unlikely (I actually had a look at it recently when I was looking for years where both the feast of St Joseph and the Annunciation fall during the Holy Week and must be pushed back to after Easter).

The Holy Algorithm

Main problem seems to be that the Orthodox church, seeing that the sun was agreeing with the pope, proceeded to declare the sun to be heretic and anathema and to keep its traditional computation.

But the "WCoC" and "simple" solutions seems to be better Schelling points that "do as Catholic do cause the pope has the biggest hat".

The Holy Algorithm

As far as I am aware the computation is in fact a rather simple "first Sunday after the first full moon after the Spring equinox" for the Catholic church, so anybody can just look at the sky and compute the date. 

3Viliam9moI imagine that once in a while there may be a problem with precision, when two things appear approximately at the same time. Did the full moon happen at the same time as the equinox, or did it actually happen a few minutes sooner, therefore we need to wait for the next one? Did the full moon happen slightly before or slightly after the Sunday midnight? I have no idea how often this happens in practice, and what precision is used to determine "equinox" and "full moon". But I guess that's what the 140-page Latin documentation is for.
Load More