In other news today...
-
The amount of poo that involuntarily escaped bowels must have been enormous.
-
Continuing the theme of vehicles in unusual places.
-
“This is not a situation that happens every day.”
-
@hungrier said in In other news today...:
“This is not a situation that happens every day.”
Thank the l lucky marbles flying cars never took off...
-
@chozang That driver was really high.
-
@chozang said in In other news today...:
Continuing the theme of vehicles in unusual places.
I wonder if the insurance covers car accidents on the first and second floor.
-
@tsaukpaetra said in In other news today...:
flying cars never took off
-
@dcon said in In other news today...:
Good thing it's not email.
Saw that over the weekend. My response (as admin of that particular list) was to write to them privately and tell them that they're being idiotic if they think that registering for a mailing list would not result in them getting email, and then pointing out that every message had a link in the footer that would take them to where they could unsubscribe if they chose.
I then added them to my personal auto-junk mail filter rules.
-
@dkf Although most new phones sold in NL are pre-configured for the national emergency cell broadcast channel, it is possible to opt out.
-
@dkf said in In other news today...:
@dcon said in In other news today...:
Good thing it's not email.
Saw that over the weekend. My response (as admin of that particular list) was to write to them privately and tell them that they're being idiotic if they think that registering for a mailing list would not result in them getting email, and then pointing out that every message had a link in the footer that would take them to where they could unsubscribe if they chose.
I have a friend who manages a lot of dog shows, so she uses a mailing list provider. Which people then unsubscribe from and then complain that they don't get any information about the show. She points out (very nicely, I wouldn't be able to hold back the sarcasm) that they caused the problem and only they could fix it. They look at her blankly (this is at the show where I'm helping out).
Idiots. Idiots everywhere.
-
Normal weekday. Russians testing West-European air defence.
-
@pleegwat Why didn't they fly straight over the Finnish sea? And why are they using that stupid fake map with a country every right thinking person knows to be inaccurate?
-
@jaloopa said in In other news today...:
@pleegwat Why didn't they fly straight over the Finnish sea?
I suspect there is no international airspace between Denmark and Sweden?
And why are they using that stupid fake map with a country every right thinking person knows to be inaccurate?
One of the mysteries of life, I guess.
-
@scholrlea said in In other news today...:
Kind of surprised no one has mentioned this here yet.
Filed Under: Replies leading to this topic being Jeffed to the Garage starting in 3... 2... 1...
-
@pjh said in In other news today...:
@scholrlea said in In other news today...:
Kind of surprised no one has mentioned this here yet.
Filed Under: Replies leading to this topic being Jeffed to the Garage starting in 3... 2... 1...
So what you're saying is, they scrolled over the dropdown and it changed the value? Feasible.
-
So, ran across this over the New Year's lull, and have seen a couple more breathless AI hype links on ./ that finally triggered me into posting the rant that I had when originally reading this article. Though my 5 seconds worth of searching couldn't find a really good thread for ranting about breathless AI hype, so you get it here instead because
Unlike other top programs, which receive extensive input and fine-tuning from programmers and chess masters, drawing on the wealth of accumulated human chess knowledge, AlphaZero is exclusively self-taught. It learned to play solely by playing against itself, over and over and over — 44 million games. It kept track of what strategies led to a win, favoring those, and which didn’t, casting those aside.
In some aesthetic ways, though, AlphaZero represents a computer shift toward the human approach to chess. Stockfish evaluated 70 million positions per second, a brute-force number suitable to hardware, while AlphaZero evaluated only 80,000, relying on its “intuition,” like a human grandmaster would.
No, AlphaZero didn't gain intuition or some mystic ability to divine future results of positions as it evaluated them more slowly... IT BUILT A FREAKING CACHE! Instead of brute-force "rollouts" in real time of millions of future positions like normal chess engines do, AlphaZero did all those rollouts during its "self-training phase" and CACHED THE RESULTS, in a "positional score", like a lot of Backgammon engines use for positional evaluation. So it could search "fewer moves per second" because it had a synthetic, cached score for each of its possible move trees, instead of having to brute force all the way to the end of the game to evaluate a given next move.
:wharrrrrgarbl:
-
@pjh said in In other news today...:
@scholrlea said in In other news today...:
Kind of surprised no one has mentioned this here yet.
Filed Under: Replies leading to this topic being Jeffed to the Garage starting in 3... 2... 1...
I blame poor UI design.
-
@dcon Or in modern UI design:
-
@izzion said in In other news today...:
No, AlphaZero didn't gain intuition or some mystic ability to divine future results of positions as it evaluated them more slowly... IT BUILT A FREAKING CACHE!
They can't say something like that to the press, otherwise the majority of the press conference would be spent trying to explain that no, the computer isn't going to get arrested for counterfeiting...
Because homophones.
Filed Under: And then they'd have to field questions about why they made the computer homophonic.
-
ObDrivingAntiPatterns:
-
@boomzilla said in In other news today...:
ObDrivingAntiPatterns:
"Where are the brakes. Oh fuckfuckfuckfuck... Hell, I need a drink, that supermarket will do..."
edit: Damn, I nailed it (well the last part). Wrote that without reading the article.
-
@scholrlea
And here I thought it was so they could get moreKickscammergovernment cheese funding
-
@boomzilla said in In other news today...:
ObDrivingAntiPatterns:
At least he was on the right tracks.
-
@boomzilla said in In other news today...:
ObDrivingAntiPatterns:
At least no one was hurt, tank goodness.
-
@izzion said in In other news today...:
IT BUILT A FREAKING CACHE!
I don't think a cache is a useful way of thinking about AlphaZero. Even if the evaluations of a trillion unique board positions per game were saved, that would only amount to ~10^21 cached results. The estimated number of sensible chess games is estimated to be on the order of 10^40 (to say nothing of 10^120 possible games). How useful is a cache that covers such a tiny amount of the state space? AlphaZero has to be evaluating positions, not just performing lookups from an inordinately sized database.
-
@mzh
While I will admit that the entirety of my knowledge of AlphaZero was gained solely from that 538 article, it seemed like a pretty obvious inference to me:It learned to play solely by playing against itself, over and over and over — 44 million games. It kept track of what strategies led to a win, favoring those, and which didn’t, casting those aside.
So, it started playing moves at random, recorded the results of the game after each move, and started winnowing down the tree of possible moves for the next game by scoring which moves led to wins or losses. Seems reasonable to me to infer it's building up a "position score" cache, like what GnuBG presents to you with the list of possible backgammon moves for a given position and roll.
Though yeah, 88 million games (since it was playing itself) isn't much out of a universe of 10^120 possible games, so maybe I'm over inferring?
-
@izzion The paper written by the AlphaZero team is pretty interesting:
-
This neural network takes the board position s as an input and outputs a vector of move probabilities p with components pa = P r(a|s) for each action a, and a scalar value v estimating the expected outcome z from position s, v ≈ E[z|s]. AlphaZero learns these move probabilities and value estimates entirely from selfplay; these are then used to guide its search
The parameters θ of the deep neural network in AlphaZero are trained by self-play reinforcement learning, starting from randomly initialised parameters θ. Games are played by selecting moves for both players by MCTS, at ∼ πt. At the end of the game, the terminal position sT is scored according to the rules of the game to compute the game outcome z: −1 for a loss, 0 for a draw, and +1 for a win. The neural network parameters θ are updated so as to minimise the error between the predicted outcome vt and the game outcome z, and to maximise the similarity of the policy vector pt to the search probabilities πt ... The updated parameters are used in subsequent games of self-play.
So, yeah, it's building a cached "positional value" score and then using that in future games, starting from assigning random scores to each position and then adjusting the scores based on actual observed results when they get played.
-
PlayStation developer Quantic Dream (Heavy Rain, Beyond: Two Souls, the upcoming Detroit: Become Human)
If the accusations are that they made employees play Beyond: Two Souls then they're 100% warranted
-
@izzion that's a lot like what Max Deutsch tried to do. He challenged world champion chess player Magnus Carlsen with something like a month of practice. His algorithm wasn't quite done in time for the match, unfortunately, so he had to play on his own intuition and what he'd learned about chess so far. His algorithm was supposed to provide him with a way of determining, in his head, how good or bad potential moves were, without having to exhaustively simulate every possible outcome.
-
Well, that's just fucking horrible.
-
@polygeekery said in In other news today...:
Well, that's just fucking horrible.
I am normally an alleged and presumption of innocence type of person in addition to being against cruel and unusual punishments, however in this case...castrate them both and throw them in the hole in a prison in one of the worst developing countries that we can find.
-
@izzion said in In other news today...:
Though yeah, 88 million games (since it was playing itself) isn't much out of a universe of 10^120 possible games, so maybe I'm over inferring?
The whole point of AI is over-inferring! The number of games isn't very important; the number of reachable positions is much more important (as there are a whole host of positions which can be reached by many possible games, yet have effectively the same valuation). But with AI, what you do is you take a bunch of data (which could be nearly randomly generated to start with) and over-fit the fuck out of it with a high-dimensional hyper-surface. Keep that up and you get something that works tremendously well on its training data, and which crumbles nastily as soon as something unexpected turns up…
So it's just like a lot of people. It's not artificial intelligence, so much as artificial management…
-
@karla said in In other news today...:
I am normally an alleged and presumption of innocence type of person in addition to being against cruel and unusual punishments, however in this case...castrate them both and throw them in the hole in a prison in one of the worst developing countries that we can find.
So… Southern California?
-
@pleegwat
Hey look the air defense did a thing. ( and share the task of scaring off Russians from their borders.)
-
Due to the urgency of the situation, the F-16s were allowed to break the sound barrier above The Netherlands. It is reported that a sonic boom was audible in the North East of The Netherlands.
So usually they ask nicely?
-
@jbert
No, usually they don't. It's like flying with the handbrake on or something over 95% of the territory.
-
@luhmann said in In other news today...:
@pleegwat
Hey look the air defense did a thing. ( and share the task of scaring off Russians from their borders.)The was at it too, I was leaving for work when I saw a tanker take off and head North at maximum speed to support our Typhoons:
-
@cursorkeys said in In other news today...:
The was at it too
But that is no surprise ... showing up on the other hand ...
-
@izzion said in In other news today...:
Though yeah, 88 million games (since it was playing itself) isn't much out of a universe of 10^120 possible games, so maybe I'm over inferring?
No. Shannon was simplistic in that estimate - he presumed 30 legal moves for 80 plys (3080). Clearly at the start and end of real games you don't normally have 30 legal moves.
Godfrey Hardy came up with 101050 based on more rigorous assumptions, but that also includes nonsensical games (you could win the next move, but do something else instead.)
A more sensible estimate, ignoring silly games, is around 1040. More:
-
-
-
@polygeekery Fortunately, it's not Arnie in action down there or we'd have to start looking out for murderous aliens as well…
-
@dkf said in In other news today...:
@karla said in In other news today...:
I am normally an alleged and presumption of innocence type of person in addition to being against cruel and unusual punishments, however in this case...castrate them both and throw them in the hole in a prison in one of the worst developing countries that we can find.
So… Southern California?
That's funny...but even I know Southern California is not the worst of the worst.
-
-
@dkf said in In other news today...:
@karla said in In other news today...:
I am normally an alleged and presumption of innocence type of person in addition to being against cruel and unusual punishments, however in this case...castrate them both and throw them in the hole in a prison in one of the worst developing countries that we can find.
So… Southern California?
Based on both the real estate prices, and the amount of silicone on display, I would called Southern California overdeveloped.
-
@pjh
Meh, I grabbed the bigger of the two numbers mrz's post had linked, but it's not like 44M or 88M are even a tithe of 10^40, either, so the point of "AlphaZero didn't have a cache of every possible game" is still valid.@anotherusername said in In other news today...:
@izzion that's a lot like what Max Deutsch tried to do. He challenged world champion chess player Magnus Carlsen with something like a month of practice
Buried around the end of October in his Medium retro-diary, there was an interesting note: he got his algorithm together, mathed it down to the most efficient version he could come up with in the time limit he had, and then tried to run it on his computer... and the computer rolled over and died. So he uplifted it to the cloud, got it running, and then figured out the model he was using was too simplistic and over-fitted, and only really worked for the test data; once it tried to start "learning" it got worse at fitting the test data and didn't do any better in the real world. So he shifted to a different model, and the model just died on the conventional cloud platform stuff he had access to (even "machine learning" optimized cloud hardware).
Big Blue, Deeper Blue, AlphaZero and other chess computer "breakthroughs" have been achieved by throwing impractically large amounts of hardware at the problem. Now, maybe AlphaZero is an advancement in a way to efficiently cache results so you don't have to brute force rollout every time, so it can help with other machine learning models. But to attribute it "intelligence" or "intuition" is just scratching your God complex.
-
@luhmann said in In other news today...:
@cursorkeys said in In other news today...:
The was at it too
But that is no surprise ... showing up on the other hand ...
New year, they have the budget for fuel still?
-
@dcon said in In other news today...:
@luhmann said in In other news today...:
@cursorkeys said in In other news today...:
The was at it too
But that is no surprise ... showing up on the other hand ...
New year, they have the budget for fuel still?
Not any more they don't--just burned it all up.
-