Random Science Facts Thread!

[YNM]

Just now, Green Baron said:

A magma chamber exposed to seawater like 1883 ? That in today's densely populated world ? Hopefully not.

I hope so... at least if it were to happen we'd all have moved up to the cliffs...

[Green Baron]

According to model calculations, in a binary system with a protoplanetary disk there are two stable configs for the disk: either in the sama plane as the binary stars, or rectangular to them.

An example for a rectangular disk has been found:

https://www.nature.com/articles/s41550-018-0667-x

Edited January 27, 2019 by Green Baron

[KSK]

It’s very easy to tell the sex of a turtle. Females have flat bellies, males have concave ones.

[kerbiloid]

1 hour ago, KSK said:

It’s very easy to tell the sex of a turtle. Females have flat bellies, males have concave ones.

Because the former do fitness, the latter drink beer.

_{Upd. My fault, confused "concave" with "convex".}

Edited January 27, 2019 by kerbiloid

[StrandedonEarth]

36 minutes ago, kerbiloid said:

Because the former do fitness, the latter drink beer.

Beer would make their bellies convex, not concave. Concave must be from sucking in their gut whenever a flat-bellied she-turtle cruises by.

[kerbiloid]

Btw, you can easily define the leader in your turtle pair with help of a remote control for the TV.
Just put it in front and see who occupies.

Edited January 27, 2019 by kerbiloid

[sardet]

On 1/26/2019 at 3:13 PM, YNM said:

Krakatau - Before & After

... going back to Danan & Perboewatan ?

Impressive

[KSK]

19 hours ago, StrandedonEarth said:

Beer would make their bellies convex, not concave. Concave must be from sucking in their gut whenever a flat-bellied she-turtle cruises by.

I gather that a concave belly helps with the , ahem ‘stacking problem’ when producing more turtles. Those big shells do have drawbacks.  

[YNM]

On 1/26/2019 at 10:39 PM, Green Baron said:

People near Etna have the same preoccupation, Etna's eastern flank was reported to be moving, but it is not clear whether it settles or may one day give in to gravity, if the cause is pressing magma or crustal subsidence.

It seems that around the same time as with Krakatau, Etna was erupting and moving too. Not to say it's linked but I suppose it shows how dangerous volcanoes can be.

But nowhere else in the world can they grow crops 2-3x in a year without extensive modification...

4 hours ago, sardet said:

Impressive

... well, more like terrifying TBH...

Spoiler

~500 people died.

 

Edited January 28, 2019 by YNM

[sardet]

I meant the power of nature in general

[James Kerman]

The Statospheric Observatory For Infrared Astronomy (SOFIA) , a modified 747SP carrying 2.7-meter (106-inch) reflecting telescope will attempt to image Titans atmosphere as it flies through the moons shadow, cast on earth this evening.

Image Credit: NASA

The Australian Broadcasting Corporation has posted an article explaining the mission here:
https://www.abc.net.au/news/2019-02-01/sofia-flying-telescope-occultation-chasing-shadow-titan/10635802
 

Update:  The mission was successful and the results will be compared with Cassini data to see if Titans atmosphere has changed over time.

Edited February 2, 2019 by James Kerman
Update and fixed links

[Cunjo Carl]

Recent news in AI advancement

After mastering tic-tac-toe (the only way to win is not to play) AI has steadily advanced to tackle Chess, Go and is now burgeoning into Starcraft 2. AI developers have considered StarCraft 2 to be the next great stepping stone for AI because it's substantially more difficult to play- it has imperfect information, real-time play, a vastly greater move space and much more complex and layered interactions. This makes it very entertaining for us viewers because it gives the game a rich strategy which allows for different players' personalities to shine through.

To date, many teams of AI developers have tried to tackle StarCraft 2 only to come up short. Just last week though, Google's DeepMind team showed some impressive results with their new AI called Alphastar. Pushing AI the next step forward in StarCraft 2 required significant advancements both in technology and in how they created and trained AI agents. In a broad picture, they trained their AI in the normal way by creating neural network AI 'agents', vying them against each other, promoting the victors and making subtle random changes to their offspring for the next round. However, the DeepMind team attributes a large part of their recent success to 3 key and interesting choices:

1. They were very hands-off in the training process. It would have been easy to directly program the AI in many of the precepts of good Starcraft 2 play in order to give the starting agents a solid base-line and head start, but they instead let their agents learn everything from scratch. This led to several interesting discoveries in new/unusual play styles!

2. They trained AI agents using a league system. Rather than just throwing out the less successful agents, they would instead be left to play and evolve for a while in lower leagues. Some times, especially in AI, good ideas take a while to develop, and the league system gives those ideas time to mature rather than (more typically) discarding them in favor of the quick successes. This closely mirrors how people develop strategies in StarCraft 2; many great play styles develop from non-professional players in lower tiers, before they gain traction and catch on with the pros.

3. They selected specifically for agents that were difficult to exploit, rather than simply those with a high win rate. The quintessential problem of AI is that it tends to be fantastic at solving specific problems it saw during its training, but then be hilariously terrible when it comes to extrapolating this experience into the real world. So DeepMind focused hard on making their AI as least 'brittle' as possible. They selected for AI agents that lost the least consistently against other agents' specific play styles. Agents that won and lost in a well-rounded way were promoted, while the obliquely talented ones were left behind.

The results: So far the AlphaStar AIs have beaten StarCraft 2 professional players 10:1*. The AI agents tended to fight bizarrely and fearlessly, focusing hard on early-mid game pushes with masses of a single kind of unit. ([Day9]'s Funday Monday monobattles return!!)

I'll spare too many details, but there's a lot of nuances that makes this 10:1 victory much less impressive than it might look at first glance. In each of its victories, the AI was granted perfect control, infinite speed, the impossible ability to view/command the entire map at once, the ability to change AI 'agents' each game so humans couldn't adapt to any one's play style, and they played exclusively on an old less-practiced patch, with only one race & match up (Protoss vs. Protoss) and on only one map. Also, the 2 professionals chosen were superbly talented (actually two of my favorites!), but still far from world's best (at maybe 200th, and 20th place). Needless to say, there's a lot of room for the Alphastar AI to still improve. However, it's an incredible feat that an AI has taken a game off a professional even _given_ these caveats! Starcraft 2 is hard.

Google's DeepMind team showed some amazing technology that promises to develop further in lots of interesting ways. I'm excited to see where they take it from here.

Figured I'd share!

(Embedded video starts in the middle of the presentation on a cool looking tech graphic)

 

[KSK]

Interesting article. I had mixed feelings about the way the single game was won against the AI. (Attack back of base, withdraw when AI responds, watch AI go back to what it was doing, rinse-wash-repeat.)

Granted, it was a lot harder to execute but that kind of exploit spamming seemed a bit ‘same old, same old’ to me. Much effort expended in training an advanced AI - and it still gets stuck in that kind of decision-making loop.

 

[Cunjo Carl]

8 hours ago, KSK said:

Interesting article. I had mixed feelings about the way the single game was won against the AI. (Attack back of base, withdraw when AI responds, watch AI go back to what it was doing, rinse-wash-repeat.)

Granted, it was a lot harder to execute but that kind of exploit spamming seemed a bit ‘same old, same old’ to me. Much effort expended in training an advanced AI - and it still gets stuck in that kind of decision-making loop.

 

To double down on the point that the AI was exploitable, Mana had already exploited the AI's playstyle to put himself into a commanding advantage even before that hit and run harassment started. In other words, Mana could have won the game even if the AI responded correctly to the harassment by splitting off handfuls of units to defend each base.

The real key to Mana's success was that he (and TLO) realized each AlphaStar agent only had one tech path (play style) in mind, and wouldn't deviate from it. So, once he guessed and confirmed the techpath (mid game stalker push), he built an army that countered it hard. He won with better unit composition rather than the AI's better unit control, and the harassment was just icing on the cake. Not that it wasn't super funny to watch, mind you!

AlphaStar still has a long way to go, but even being able to play toe-to-toe for a little bit with professionals is something brand new for AI.

 

Conveniently, Mana talks about it a bit in his stream, I'll embed the video in here.

Spoiler

 

 

[wumpus]

On 2/1/2019 at 10:49 PM, Cunjo Carl said:

To date, many teams of AI developers have tried to tackle StarCraft 2 only to come up short. Just last week though, Google's DeepMind team showed some impressive results with their new AI called Alphastar. Pushing AI the next step forward in StarCraft 2 required significant advancements both in technology and in how they created and trained AI agents. In a broad picture, they trained their AI in the normal way by creating neural network AI 'agents', vying them against each other, promoting the victors and making subtle random changes to their offspring for the next round. However, the DeepMind team attributes a large part of their recent success to 3 key and interesting choices:

From what I've heard, Computer players have a huge built-in advantage in StarCraft 2 in being able to click with mouse/keyboard far faster than even pro players.  Are they just considering AIs that have been limited to pro (or even amature) production rates?  This sounds like a real challenge, as a computer has a sufficient advantage to being able to produce a zerg rush (and force a human or other AI to deal with that) far more effectively than a human, with or without an advanced AI.

[StrandedonEarth]

1 hour ago, wumpus said:

Are they just considering AIs that have been limited to pro (or even amature) production rates?

Nope, they allowed the AI's god-like control

On 2/1/2019 at 7:49 PM, Cunjo Carl said:

I'll spare too many details, but there's a lot of nuances that makes this 10:1 victory much less impressive than it might look at first glance. In each of its victories, the AI was granted perfect control, infinite speed, the impossible ability to view/command the entire map at once,

 

[Cunjo Carl]

7 hours ago, wumpus said:

From what I've heard, Computer players have a huge built-in advantage in StarCraft 2 in being able to click with mouse/keyboard far faster than even pro players.  Are they just considering AIs that have been limited to pro (or even amature) production rates?  This sounds like a real challenge, as a computer has a sufficient advantage to being able to produce a zerg rush (and force a human or other AI to deal with that) far more effectively than a human, with or without an advanced AI.

What you're talking about is totally a thing. The most recent bots vs people tournament I'm aware of had a case just like that, where the machine built and controlled a rush of zerglings perfectly to muscle out some poor human protoss player. I don't speak a word of Chinese, but you can hear the commentator's reaction to seeing the sudden flood of computer units "Aaaah....  GG?" (in other words, is that game?)

Human players can eke out victories even with these disadvantages though, often making it look easy like Stork does later in the same tournament. In Starcraft (and SC2), you can lose every battle but still win the war with advantages in economy, technology, positional control, etc. It's never enough to just have an advantage though, you need to know how and when to use it, and that's where humanity has always shined.

What's made AlphaStar interesting is seeing that it actually tries to get advantages in these other ways, and watching how it tries to leverage them.

In the final show match it lost against Mana, you can see it trying to win using an economic choke hold rather than brute force. It builds an additional resource-mining base ahead of Mana and then does its best to stunt Mana's economy by killing his workers. After that Alphastar maintains its very slight army and positional advantage by keeping back and defensive, meanwhile rapidly growing its economic advantage. As the game progresses, it constantly keeps tabs on Mana's economy to make sure he's stuck, and (this is the cool part) it forgoes army conflict to best leverage its bigger economy for an even better advantage in the bigger later battles. However, it wasn't counting on Mana happily building up an unstoppable high tech army on just 2 little bases and then marching his way to victory. Being able to feel out these complex and constantly evolving interactions of advantages is the strong point of the human mind versus programs and AI. So, humans can find a way to win (for now) even given a wild handicap in terms of speed and control.

Saying this there's definitely some moments when Alphastar is in its element and its decision making looks almost... human! That's what everyone wants to see developed ultimately, the point isn't just winning, so for this to be fair and worth-while in the long run the DeepMind team will need to find a way to impose human control abilities onto their AI agents.

[Green Baron]

Oldest rock from Earth may have been found by Apollo 14 on the Moon, as a Zirkon enclosed in Lunar breccia. The Zirkon may have formed under conditions that do not exist on the moon, but could have existed on earth.

Solidification of Lunar surface is dated to 4.533Gy, oldest rock on Earth up to now to 4.404Gy. So this may be the oldest rock from Earth ... found outside of the front door.

https://www.sciencedirect.com/science/article/pii/S0012821X19300202

 

[kerbiloid]

The moon where you can find the stones from the earth.
(Uncompressed magnet link in comments, 291 Mbytes)

Spoiler

 

 

About AI.

https://translate.google.com.tr/translate?sl=ru&tl=en&js=y&prev=_t&hl=ru&ie=UTF-8&u=https://meduza.io/feature/2019/02/16/laboratoriyu-openai-sozdali-dlya-razvitiya-iskusstvennogo-intellekta-na-blago-lyudey-u-uchenyh-poluchilsya-otlichnyy-generator-feykovyh-novostey

Edited February 20, 2019 by kerbiloid

[Barzon]

I saw that picture! I never could find the uncompressed version. If anyone wants, could they find the link for me?

[kerbiloid]

7 minutes ago, Barzon Kerman said:

I saw that picture! I never could find the uncompressed version. If anyone wants, could they find the link for me?

magnet:?xt=urn:btih:j6ixrzpitvpmkx5tgroy6utcpqr5hhhj&dn=updated%20mosaic.png&xl=304863698&fc=1

Edited February 20, 2019 by kerbiloid

[Green Baron]

The GTC (10.4m mirror, largest optical single mirror telescope today) rests on a very thin oil film. Though it weighs 385 tons it can easily be pushed by hand.

I checked during a guided tour.

Edited March 2, 2019 by Green Baron

[CastleKSide]

There are types of hawks in Australia that have allegedly learned to control fire.

https://news.nationalgeographic.com/2018/01/wildfires-birds-animals-australia/

[kerbiloid]

On 3/2/2019 at 1:37 PM, Green Baron said:

The GTC (10.4m mirror, largest optical single mirror telescope today) rests on a very thin oil film. Though it weighs 385 tons it can easily be pushed by hand.

I checked during a guided tour.

Then we read: "GTC obtains the deepest image of a galaxy from Earth" and see this:

Spoiler

It's not a galaxy. It's a finger.

 

[Green Baron]

One day i'll tell you the difference between a finger and a galaxy ;-)

------------

A short excerpt on the state of black hole science (should be open access, not behind the pay wall's event horizon):

https://www.nature.com/articles/s41550-018-0682-y

 

Edited March 7, 2019 by Green Baron