Fan Hui is 2p, so very skilled, but the ranking system goes up to 9p. To give a sense of how large a gap that is, there is and has only ever been one Westerner to achieve that rank, Michael Redmond. The article states they plan to face off against Lee Sedol 9p, and if they beat him in a no-handicap game, that will be as impressive as Deep Blue against Kasparov.
You'll want to watch this year's Computer Go UEC Cup[0] in March, with Zen and CrazyStone being the typical victors. (CrazyStone in particular has sustained a 6d rating on KGS, a popular Go server, but lately has been at 5d. In the past it has beaten a 9p with a 4 stone handicap, which is impressive, but that handicap is huge and it hasn't yet won with 3 stones.) Of interest this year is that Facebook is competing, and AFAIK they seem to take a similar approach as Google by training the AI using deep learning techniques and then strengthening it further with MCTS. In their public disclosures they claim to beat Pachi pretty often, which puts their bot around 4d-6d, it'll be interesting to see how it fairs against Zen and CrazyStone in the Cup and if it wins against a 9p.
This is not correct. In the professional ranks 1p is often no weaker than 9p because often the young players start out at 1p and are very strong because selection pressure is much greater. The 9p rank in Japan also came partly out of just playing a lot for a lot of years so that accomplishment was not as great as it seems. In any case, active professional players for the most part are not dramatically stronger than one another. The range for the most part is about 2 stones, maybe 3.
Even though Fan Hui is not an active professional in the traditional sense this is an absolutely huge accomplishment and leap in playing ability by computers. BTW, Michael Redmond is not particularly strong by professional standards.
(Edit: The correlation between strength and rank now as noted below is due to promotions more often coming from acheivements: If you win at X you get promoted to 7p immediately, if you win at Y you get promoted to 9p. You cannot win a big tournament and keep your low rank. Here is an example of someone who went from 3p to 9p in one match: https://en.wikipedia.org/wiki/Fan_Tingyu. But professionals cannot give each other 6 stone handicaps when one is 9p and the other is 3p)
This is not quite correct either. 1p is of course generally much stronger than an arbitrary amateur dan (even some (many? most?) 9d amateurs) and you can advance up the pro ranks quickly by winning certain games, but you can see the histograms yourself that while there's a clump of 9ps everywhere but China there's still a distribution. http://senseis.xmp.net/?ProfessionalRankHistograms Another problem is you don't lose your 9p rank once you earn it. It would be nice if there was an international Elo system tracking all the 9ps of various countries to rank them properly... Maybe someone's tried to calculate rankings independently? Still, I think it's pretty uncontroversial that someone who's been a lower-rank pro for longer than a few years is going to be significantly weaker than their higher-rank peers.
There are two major attempts at international ratings today. Dr. Bae Taeil does ratings for Korea, and Remi Coulom produces ratings independently, based on the database at go4go.net (http://goratings.org is the site).
Taeil's method seems unusual, though it may well be justified . I think he's using a relatively complete database of games, but I don't know for sure. Coulom has a very well regarded mathematical model, but we know that there are some gaps in the database, which a) may skew international comparisons, and b) may result in inaccurate ratings for players with few games in the database (but those players are usually not top players in the world).
See my comment below: there are very few 1p players near the top, unsurprisingly.
I feel like this is mostly misleading. Look at a list of top players (goratings.org). The top fifty is mostly players 5p and up. There are a few 3p or 4p Chinese players running around, and apparently even 1 1p (Li Qincheng) but by and large, there is a relation.
Yes. He's a case where you'd make exactly the right assumption if just told his rank.
Though I should warn: as a low level pro who moved to Europe, this database has very little data on him. His European results indicate that he's stronger than Pavol Lisy, Alexander Dinerstein, and Mateusz Surma, who all rate higher than him here.
I don't think it's overblown at all. They make it clear he is the European champion and never imply he is the best player in Go, just obviously very good. Looks like he has won the last 3 European championships in a row (https://en.wikipedia.org/wiki/European_Go_Championship).
Overall I think it was a well balanced article highlighting an impressive achievement, not sure why you felt the need to diminish it as not significant?
The version of AlphaGo used in the paper beats CrazyStone about 99% of the time. Even if we give CrazyStone 4 stones handicap, AlphaGo still beats it 80% of the time.
Very nice. How come you guys didn't enter the UEC Cup? (Edit: rereading this it kind of sounds snarky, not meant that way. Really impressive if you can kick CrazyStone's teeth in...)
From the paper, no that's with "single machine AlphaGo" which is 48 CPUs and 8 GPUs. Distributed AlphaGo beats Single Machine AlphaGo 77% of the time. (Distributed AlphaGo being 1202 CPUs and 176 GPUs.)
It's worth clarifying your use of the "p" rating scale alongside the "dan" rating scale. Essentially, there's an "amateur dan" scale which is what's measured on public servers like KGS. Under this scale essentially all professional Go players rank at the maximum rank, 9 dan.
In parallel there's the professional ranking system which also uses the title "dan" but is bestowed by the professional Go associations of every country. These rankings are symbolic instead of quantified, although generally higher professional dan ranks cannot be anything but corresponding to higher skill.
So, a "European 2d" is a professional rank which may or may not have a good translation into a quantitative scale like ELO or "KGS dan" (I actually don't know). Generally, my understanding is that European professional rankings lag behind Asian ones as well by some amount.
You're generally accurate, except that Fan Hui is a Chinese 2p from the Chinese Pro Association who has played in Europe for several years. He's still generally a bit stronger than the homegrown European professionals.
> In the past it has beaten a 9p with a 4 stone handicap, which is impressive, but that handicap is huge and it hasn't yet won with 3 stones
4 stones is a huge handicap? 4 stones is a huge handicap if he was playing against a 6p maybe. But with 4 stones handicap anything near 2-4p would still have a hard time.
The only one who knows if the input is needed anymore is the caller, not all callers will agree, so it's bad form to let the called function decide automatically unless it's very clearly marked as a mutating, list-destroying function.
Besides, all the ARIA stuff isn't free. It only benefits screen readers and bots that aren't Google's. The rejection from lack of accessibility is as strange as rejection from lack of functionality in IE 8 or less (which has perhaps 2-4x as much market share as that of all screen readers). For any website or webapp you should know your market and make the call whether you need to spend time upfront, burn the time now, or delay indefinitely or forever, on supporting these low market share alternatives.
Posts like this give me the sads. Equating disabled users with a browser that has been EOLed? Wow, just wow. Also aside from being the right thing to do, depending on who you are and where you are, it may be the law...
There already are tax and welfare benefits to having children, and indeed it does incentivize having them when you really can't afford them. It doesn't seem sustainable long term, but we'll see. I don't think it's a problem with basic income any more than the current system. The bigger issue I find with basic income is that proposals typically don't include any mention of price control on the seller side and purchase control on the buyer side. Poor people are often trapped there because every local actor selling them stuff is trying to maximize rent extraction, and they continually buy things that are expensive and bad for them (case in point the recent cigarette article). Landlords get a lot of money from rent extraction, since they're paid in rent directly. It's even worse with universal, unconditional basic income. If as a landlord I hear everyone in my apartment complex is now getting an unconditional basic income of $20k/yr, guess what, next year rent prices are going up by around $20k, with lowered adjustments to account for other actors (like local food suppliers) also competing for this extra $20k/person/yr. This is basically the same thing that has happened to student loans. Without authoritarian measures to restrict economic freedom, which in addition to being authoritarian must also be good measures so as to not backfire as every other centrally planned economy has, basic income has no real chance at a large scale. In America in particular, authoritarian anything is a no-go.
A landlord may try to increase rent knowing that the tenants now have a basic income, however that creates an opportunity for someone else to provide housing at a cheaper price - it's no different from any other market, as long as you don't have massive supply-side restrictions like in some major cities.
The best way of fixing this is to make sure that there is an over-supply of housing (located NEAR the jobs) and to make it as easy as possible to move.
Economic incentives should encourage local, suitable (even for apartments) housing in an area to be approximately equal to the number of jobs in that area (then pad up slightly for BIG if that's also an 'employer').
What is this total inequality you talk about? It seems like you're picking the inequality of 0 and infinity. But we live in a finite world, this total inequality is impossible. Furthermore, whether we introduce an infinite inequality or a finite inequality, there's no reason the lower bound should be at 0 (where people starve and die). Suppose we have a society whose inequality range generally falls within 1 and 1,000,000. 1 is a sustenance standard of living, any unfortunate event could totally wipe you out, but it's pretty much maintainable with little effort provided that unfortunate event doesn't happen. Somewhere around 3 being the poverty line and somewhere around 8 being a decent standard of living so long as you're working to maintain it, and somewhere around 50-100 is enough for most people to effectively have that same decent standard of living as 8 but without having to maintain it so hard e.g. through a job. What's wrong with this lower bound increasing by 1 every year (with its associated low level of upkeep remaining the same) while the upper bound increases by 2 every year? The inequality is growing with each year, but in 2 years there is no one below the poverty line, and in 7 years everyone has a decent standard of living but they don't really have to put in effort for it. Why is inequality the issue here?
The entire thrust of our medical regulatory system, from the Flexner Report to today, is the belief that it's better for 1000 patients to die of neglect, than 1 from quackery. Until this irrational fear of quack medicine is cured, there will be no real progress in the field.[1]
They'll probably succeed for the most part, too. Who's going to spend their time playing cat and mouse when they can just unsubscribe and torrent? If you're not seeding, you're breaking the same law of unlicensed access to content either way.
If you include thinking about work, then even work-from-home people could reach the 50+ figure relatively easily since you include weekends and trying to get to sleep. But then you might easily get into low numbers of hours per week, 15-30, if you subtract things like spending some large % of your time doing non-productive non-work (happy hours, lunches not at your desk, watercooler talk, browsing HN) even though you're at work.
Work time should generally only be the time between clock-in and clock-out. For professionals that are freed from clock-in-clock-out, I think they would still clump around that 40 hour figure if you made them clock-in-clock-out for the data rather than for the pay.
If we limit 'free action' to mean 'ability to charge your credit card' then you are pretty much described beeminder.com (a company I highly recommend and of which I am a very happy user).
I also recommend them to everyone, but between all the praise they're getting, I think I should also give a fair word of warning - while great, it may not work for everyone.
I used the service once; I tried to discipline myself out of a poor financial situation using Beeminder - very quickly I found myself consistently slipping, had to pay the money, and I ended up in a worse financial situation and stressed to the point I could barely function.
Now don't treat it as a negative - instead, like a warning on the razor box: this thing is sharp, do not use when your hands are shaking. It's great as a motivator, but be sure you're in a good enough mental shape to handle both your commitments and the eventual slip-ups.
That said, I also have to praise Beeminder for having a great tool with lots of little nice features (I'm particularly fond of the graph aesthetics), amazing and responsive support (and proactive - I screwed something up once, they fixed it for me pretty much immediately and sent tips on how to avoid this happening again), and wrote a lot of interesting things on akrasia.
EDIT:
To 'dreeves - long time ago I shared this experience over at SlateStarCodex, and I missed your reply then. Referring to it, and that you're "extremely averse to people paying money to Beeminder that they don’t feel was worth it" - no, I feel it was totally worth it, even if it contributed a bit to one of my biggest stress episodes in my lifetime. I'm not scared of Beeminder, I'm scared of myself. I treat it as a combat scar in battle against akrasia :). So thanks for your kind refund offer, but a) you're totally entitled to the money, and b) it was long time ago :).
Oh, and thanks for supporting SSC :).
And, to everyone else - did I tell you already that Beeminder people are really great, friendly and radiate honesty in a way that's pretty much unseen in this industry? I do very much recommend the service despite the fact that, by my own fault, I managed to cut myself with it.
Holy cow, you made my day! <blush> Thank you! Fascinating what you said in another comment about the catch 22 of remembering and sticking to plans when you're feeling good. Actually, I'm realizing I'm confused about this. I'm so everything-looks-like-a-nail that it sounds eminently beemindable so it would probably be good for me to keep trying to understand what I'm missing there. dreeves@beeminder.com if an email thread seems easier.
Thanks so much for saying so! Maybe I'll also take this excuse to highlight the footnote in the article:
> Thanks to my friends at Beeminder for some of the ideas I mention here. You can see their full article [http://blog.beeminder.com/akrasia] on commitment devices to overcome akrasia for more ideas.
I remember looking at Beeminder, probably around the time that article was written unless you promoted/were promoted on LW before then. How much would you say you've improved the service since then? It definitely looks better than I remember... I think I'll test it out this weekend since I can see it being pretty helpful for those long term yet easily measurable goals, two of which I have under "lose x lbs" and "read y books this year". I'm not so sure it would be that helpful for the smaller goals, though, especially ones that require more subjective measurement ("you've been on your cell phone too much today") or ones that require frequent measurement updates from the individual ("out of this list of whitelisted sites that are directly work related your fraction of time spent on other sites over the last time period is x%, which happens to be too much"). Anyway I wanted to ask if you're familiar with any research on akrasia correlated with childhood. I have no idea if or how much things like divorce, loose parenting, strict parenting, or a change in style as the child transitions into teenage years affects anything. It was just an idea I had this morning when I realized my above comment is really just asking for a strict parent who in the process of looking out for my own best interests (that as an adult I can now agree with are such, or at least rationally argue otherwise and be taken seriously) can do such a wide variety of things like taking away my phone for a set period (while still letting me answer any important calls that might come in), or periodically poking me with a stick if I'm staring at my screen seemingly frozen and there isn't a code editor open.
OMG yes, that article is from before we actually launched. We've made, let's see, 1789 user-visible improvements since then -- http://beeminder.com/meta/uvi
It's still confusing and nerd-centric and we're working on making it more intelligible to newbees. Would especially love to hear your thoughts, having looked it years ago and now coming back to it.
As for beeminding things like time spent on your phone, you'd be surprised what's possible if you're willing to nerd out a bit. For iPhone I think Apple makes this impossible but on Android you can connect RescueTime and Tasker to Beeminder to automatically measure and report time spent. Or you could just have Tasker count the number of times you turn on your phone's display and beemind that. I'd love the excuse to better document such things so email me, dreeves@beeminder.com, if that sounds intriguing.
Akrasia and childhood: I don't know of research other than the Stanford Marshmallow experiment which I used to cite as supportive of Beeminder's philosophy (kids who employed tricks to distract themselves from the temptation did better) but later research makes it much less clear what's really going on there (maybe kids from unstable homes just don't trust the researcher to keep their promise which makes it rational, in expectation, to grab the marshmallow while the grabbing's good).
I do, personally, view the holy grail of Beeminder to be a nannybot that tells you minute by minute what the optimal thing for you to be doing is. Actually we just made a Beeminder Slack bot -- http://slackminder.com -- that may point us in that direction, though still very primitive now.
Whence the high performance claim? I tried finding some benchmarks, and apparently they used to be on the home page since this was on HN about a year ago, but I didn't find any recent ones. It would be nice to see it compared to, say, Rust, Nim, Haskell, OCaml, C...
But I do see where the claim is coming from. Pony is aimed at the same kind of compile-to-bare-metal that we do with C, C++, Fortran, etc. Furthermore, Pony's design explicitly deals with many of the issues that get in the way of aggressive optimization in other PLs: pointer aliasing, dynamic type checking, data races, etc.
So Pony was clearly designed with high performance in mind. And it seems likely to me that the efforts made in that direction are practical, workable ones.
But that is not the same as saying, "We can write a Pony compiler that generates fast code." And it is certainly not the same as, "We have written a Pony compiler that generates fast code."
Well, it's faster than C++ with OpenMP so it's definitely the fastest safe language around. For sure faster and safer than Rust, and it looks better to the eye also.
What I'm missing are FFI finalizers and GC triggers as with libsigsegv. Long running hungry actors can force unhandled out of memory.
Fan Hui is 2p, so very skilled, but the ranking system goes up to 9p. To give a sense of how large a gap that is, there is and has only ever been one Westerner to achieve that rank, Michael Redmond. The article states they plan to face off against Lee Sedol 9p, and if they beat him in a no-handicap game, that will be as impressive as Deep Blue against Kasparov.
You'll want to watch this year's Computer Go UEC Cup[0] in March, with Zen and CrazyStone being the typical victors. (CrazyStone in particular has sustained a 6d rating on KGS, a popular Go server, but lately has been at 5d. In the past it has beaten a 9p with a 4 stone handicap, which is impressive, but that handicap is huge and it hasn't yet won with 3 stones.) Of interest this year is that Facebook is competing, and AFAIK they seem to take a similar approach as Google by training the AI using deep learning techniques and then strengthening it further with MCTS. In their public disclosures they claim to beat Pachi pretty often, which puts their bot around 4d-6d, it'll be interesting to see how it fairs against Zen and CrazyStone in the Cup and if it wins against a 9p.
[0] http://jsb.cs.uec.ac.jp/~igo/eng/