As the game enters the confrontation stage, Luo Sheng is looking forward to watching Nafu's first-person perspective to see how it performs.

This confrontation is [Terran VS Zerg], the human opponent is a Zerg player, and Naf chooses the Terran.

At the beginning of the game, Luo Sheng, who watched Nave's first-person perspective, felt extremely embarrassed, because "she" controlled the four farmers to wander around the map for a while, and then turned back to mine, which lasted about half a minute.

Yes, the start is very bad.

This is also peat, which is really a personal retardation.

This scene directly made Luo Sheng drunk.

I don't understand, I think "she" is full of curiosity about the world, and it's really an artificial retardation.

Subsequently, Nafu built two houses in a row, and Luo Sheng thought that "she" would put the barracks building down, but "she" did not build the barracks.

Instead, he saved a sum of money to open a second mine......

Luo Sheng, who watched the first-person perspective, suddenly felt that Nafu's current behavior and operations were the same as a pure newcomer's first contact with the game StarCraft, without any difference.

As the minutes ticked by, the Zerg players who were fighting against them came to scout and saw the opponent's opening.

Zerg players are also stupid.

What the heck?

What kind of play is this?

So a wave of puppies came over, and at this time Nave had just built the first machine gunner, and it hadn't come out yet.

As a wave of puppies from the opponent's reconnaissance force entered the mine, Nav resisted for a while and ended the match with a GG shot.

The first confrontation was killed by a small reconnaissance force of the opposing side.

In fact, the strength of the human player on the opposite side is not high, just an ordinary player, if you want to meet the real top professional players, Nav needs to constantly fight in the ladder, defeat one human opponent after another, and each victory will match stronger opponents, until you reach the top of the pyramid and fight against the real top professional players.

Of course, with Luo Sheng's technology, it is completely possible to forcibly modify the algorithm of the matchmaking system, so that Nafu can directly match with top players, but he did not do that.

At present, StarCraft is a popular e-sports game, and there are a lot of people playing it, especially in South Korea.

There is even an exaggeration that at the moment in South Korea, a man aged 10-40 will be madly ridiculed if he says he can't play interstellar.

……

Soon, the second game confrontation began, this time it was a TVT confrontation game, that is, [Terran VS Terran].

By the second game, Nave had a noticeable improvement, and although she lost the game as well, she at least pushed her pawn to her opponent's home.

After a wave of main forces were all sent away, they were pushed back by the opponent's mechanized tactics.

Hit GG again.

The next one is quickly reopened.

……

When it comes to the fifth game of confrontation, it is TVZ again, that is, [Terran VS Zerg].

Naf, who lost consecutive matches, finally showed the extraordinary side of "her", and her strong learning ability allowed "her" to adapt to this level of "TVZ" confrontation, and easily defeated the opponent with operation and crushing.

Although Nav defeated her opponent with a level of operation comparable to that of a top pro, one of the details was that "she" would never make a second mistake in the same place.

In other words, Nave's learning progress is terrifying, and every confrontation is learning for "her".

Luo Sheng also saw a flaw exposed by Nafu through these five matches against human players, that is, in addition to specific division tactics, "she" could not form a flexible distribution of troops.

In the losing fourth game, a high-level human player harasses Nave's base on multiple fronts, and "she" runs away as soon as she gets back to defend the opponent, and then turns back to harass her when she goes out to attack.

In this confrontation, in the face of the flexible dispatch strategy of human players, Nav could only respond passively and could not form an effective attack on human players, which also led to the final snowball effect.

In the end, he was defeated.

After the end of the fifth game, Luo Sheng asked Nafu to stop matching opponents, and then re-upgraded and optimized it again.

He felt that this kind of learning efficiency still seemed too slow.

Nafu's agent uses deep learning algorithms, which Luo Sheng has carefully upgraded, including neural networks, self-play through reinforcement learning, multi-agent learning, imitation learning and other algorithms or technologies.

The so-called self-play, simply put, is self-learning, through continuous trial and error to find the way to play the game with the greatest chance of winning.

Through self-play and reinforcement learning, it forms the basic paradigm of the open-end learning algorithm, learning human strategies and ensuring that Naf's agent continues to explore these strategies throughout the game, so that "she" can achieve greater upgrades and progress.

In addition, because StarCraft is a game with a huge amount of room for action, many of the previous reinforcement learning methods would be rendered ineffective.

Therefore, Luo Sheng also developed a new algorithm for Nafu to perform non-policy reinforcement learning, so that "she" can effectively update decisions from the old historical decision-making game.

And Luo Sheng's ultimate goal is not to make Nafu an "IMBA" StarCraft AI player, but to use this as a starting point to make "her" competent for other complex and changeable application scenarios.

If "she" only becomes a super master in the game of StarCraft, and not in another scenario, such as Go, language, or other fields, it is still a pure artificial retardation, and there is a huge gap between it and real artificial intelligence.

……

Time passed minute by minute, and Luo Shengmao unknowingly spent more than three hours in his office to complete the second optimization and upgrade for Nafu before leaving work, and the time was not early.

Before Luo Sheng left the office, he put Nav into the ladder of Blizzard's interstellar battle network, and let "her" constantly fight against human players.

I don't care about all this, I have to go to the Cte d'Azur company in the afternoon to deal with some work, and there is a meeting or something.

And the free-range Nav is doing one thing, constantly fighting against human players.

Nave's support for "her" is almost the entire Sioux City data center, which consumes computing resources and electricity bills.

In fact, if you don't have money, you can't afford an AI, let alone the development stage.

If it is an artificial intelligence R&D institution like IBM, it may be the developers themselves who restrict its development, but Luo Sheng, who has witnessed artificial intelligence, is the easiest for development.

But there are still problems that hold him back.

That is the hardware, the growth of Nafu is limited by the current hardware level, if there is a better hardware foundation, it will usher in a qualitative leap.

Deep networks require high-end GPUs to be trained in a reasonable amount of data in a reasonable amount of time, and these hardware are very expensive, of course, there is no such thing as expensive here in Luo Sheng.

But the problem is that there is no hardware that meets his requirements now, and he has no money to buy, so he can only start his own hardware in the future.

Without an advanced hardware foundation to train deep networks for high performance, it is not feasible.

It's not just about using high-end GPUs, it's about fast CPUs, storage, and fast and ultra-large amounts of RAM.

……

As time goes by, Nave's win rate has gradually reversed from the miserable start of the first game of "her" tournament career, and the win rate is rapidly improving.

On an operational level, Nave is undoubtedly the number one in the world, and no human player can compete with "her" at this level, even if it is limited to APM and EPM.

But that's not the point, the focus is on the "awareness", "behavior" and "decision-making" that Nav shows in the game, which is the embodiment of intelligence.

……

Before you know it, Nav has fought 27 games on the ladder, and her win rate has climbed to about 70%, which is a very high win rate.

In order to improve the efficiency of Nave's "evolution", Luo Sheng let "her" enter the massive database of Bluestar Technology Group and some websites of StarCraft to access the "battle report" videos of some top players at the same time as the battle, and then train against other opponents through parallel training through the game playback videos of human players.

A script running in the background will have her iteratively upgraded every once in a while, updated with reinforcement learning, and then a new confrontation will take place.

Each iteration matches a new opponent and freezes the opponent that was previously defeated, and the probability of matching the opponent determines the learning objective function adopted by the agent, so as to retain diversity and increase the difficulty.

As the game progresses, Nav gradually develops more and more mature tactics, or relatively optimal solutions, similar to the process of human players discovering tactics, with new tactical systems constantly defeating old tactical systems and then replacing old ones.

……

In the 32nd game, like climbing the ladder, Nav continued to defeat her opponents, constantly iterated herself, became stronger and stronger, and the human players she encountered became stronger and stronger, but 90% of them had become "her" subordinates.

At this moment, another round of confrontation began, and the players that Nav had matched had reached the level of confrontation of professional players.

All the human players who were defeated by "her" didn't know that their opponent had only started playing the game today, and that their opponent was not a human, but an AI.

This game is [Terran VS Zerg], and the IDs of both sides are [Nov] and [=Y.J=F91] respectively.

At the beginning of the battle, the two sides were operating and scouting normally, and the human player with the ID [=Y.J=F91] was located in a room in Hangzhou, and he had already left the "Artistic Conception" team.

But now he still retains this ID, and he is now a real interstellar professional player, and he is a first-class top master in the domestic interstellar world, and the Jianghu people give him the nickname "Jiangnan Dog King", as a Zerg player, his puppy operation is known as world-class in the first five minutes of the game.

...... in five minutes

Well, if F91 goes to South Korea, his strength is a dishwasher over there, and anyone who squats on a water dispenser can hang him, and any Korean interstellar professional player may hang him up and beat him.

“Nov……？？ Whose vest is this ID? The operation is good, and it has a bit of strength. At this time, this confrontation has entered the middle stage, and F91 did not win the game as lightly as the previous one, but played quite hard, and even clearly felt the pressure brought to him by the opponent.

The key is that he is very puzzled now, he has never seen this ID, and the players who can have this strength in the domestic interstellar circle are basically the cards, and they basically know each other.

But the ID of Nov is the first time I have seen it, and the style of play is also the first time I have seen it, which directly allows F91 to exclude those domestic interstellar masters he is familiar with one by one.

"No matter what, which way the gods come, if you meet your brother, you have to die." F91 thought silently in his heart, and then his main force began to go out, ready to attack.

[Chicken ground is attacked!] ]

But not long after his main force went out, the sound of the headphones frantically indicated that the base was under attack, and F91 subconsciously cut the screen to his main mine, and then saw a picture that shocked him extremely much.

"Lying roasting, lying roasting, what is it? What is this? Spicy buttocks this operation...... Lying on the roast!! ”

It turned out that at the same time that F91 was preparing to attack, Nafu had already sent five ships to attack, and also deliberately circled the position at six o'clock on the map to avoid the opponent's vision in the middle of the map, and used radar to do counter-reconnaissance in advance many times to ensure that his airdrop troops were not discovered by the other party.

If Luo Sheng sees it, he will be very satisfied, this seems to be a very ordinary decision for human players, but it is different if it is an AI execution.

This progress is not insignificant.

It is worth mentioning that although Nav is an AI, "she" does not "open the map", there is also a fog of war, and the operation also needs to cut the screen like a human player, and ensure that it is the same as a human player in these parameters, otherwise it will be a bully.

F91's troops had just left the house, and Nave's five ships had successfully reached the opponent's main mine heights.

Then, the F91 saw the operation screen that shocked him incomparably.

The opponent's five transport planes actually airdropped troops while flying, so that the combat troops were spread flat on the map like a carpet, so as to achieve the maximum firepower output in the shortest possible time.

This wave of airdrops directly destroyed the main mine of F91, but in a moment, another wave of carpet airdrops of the three ships destroyed the sub-mine base of F91 and the small number of troops left behind.

Seeing that the two families are blooming, and the main force has advanced more than halfway, it is meaningless to return to help at this time, and it will definitely lose.

So the F91 decided to fight for a shot.

Swap places with your opponents!

If the main force does not come back, only in this way may there be a chance of victory.

As a result, the Zerg troops rushed to the mining area where the Terrans were located, regardless of the situation of flowering at home.

"Architecture is doing a bit well......"

The troops of F91 pushed to the opponent's mine, and were a little uncomfortable being stuck by some buildings halfway, but after a while, they exclaimed again: "What is this thing with spicy buttocks?" This ...... How could he still have so many troops at home? Rioters like urine insipidus? ”

The shocked F91 has not been able to replace the opponent's piece of mine with great effort, and the main force has lost two-thirds, and the family has already blossomed.

There is no way to return to the sky, so I sent all the last remaining troops A to the light, giving myself a reason to play GG.

After exiting, F91 did not qualify, he immediately clicked on the video to watch the game, from the beginning to the end, watched the opponent's entire first-person perspective replay, the opponent's operation, rioters, climbing technology and offense including defense, etc.

This ID that he has never seen before, and the operation and operation it shows are extremely shocking to him, which is definitely the level of the top Korean masters.

Especially the carpet airdrop troops when the opponent attacked, the moment he saw that scene, he felt that this ID was operated by the interstellar "human emperor" Boxer, and the carpet airdrop was the extreme operation he created.

In fact, F91 knows how to operate a carpet airdrop, but he can't operate it at all, which is a hard injury.

Just now, the opponent's 5 ship soldiers carpet airdrop, its operation process needs to number each transport plane one by one, and then when landing the troops, you need to switch formations with your left hand in turn, and at the same time use the mouse with your right hand to click on the soldiers in the transport plane, and cycle the operation many times until all the airdrops are completed.

It's easy to say, but it's actually extremely difficult for the average player to do, and to do it means that you need to achieve 40 precise and coherent left-handed and left-handed operations with the mouse and keyboard in about three seconds, without sequential errors.

If that's all there is to it, StarCraft wouldn't be so hard to play, and you'll have to run your home while doing that.

At this moment, F91's mind is full of question marks: When did there be such a bad interstellar player in China?

This level of operation has definitely reached the world's top level, including smooth operation and terrifying thug ability, if it weren't for playing GG in person, fighting in person, and watching the video in person, F91 would never believe that this is a domestic player, and will definitely believe that this is the vest of a top Korean master.

After the shock, he continued to rank.

I hope to be able to meet again, even if it is abused, I am very happy, for professional players, I am very happy to fight against a master who is stronger than myself, even if I am beaten by the other party, because I can also improve.

However, he will not be able to make it to the next rank, and any human player who is defeated by Naf will not be ranked second.

After Nafu reaches the top in China, "she" will enter the battle network in South Korea and compete against the real world's top interstellar professional players.

In this short period of one day, Luo Sheng didn't know that this ID suddenly became popular in the domestic interstellar circle.

The god players who became "victims" from all walks of life began to post on the community forum to ask which kind of god this Nov suddenly popped up, and actually blasted the masters one by one.

I didn't even return my hand to GG.

And the "victims" also gathered together to "complain" to each other about how they were beaten, and at the same time, some videos of the battle with Nafu have also been uploaded to Blue Star Video, and more and more interstellar players who come to watch the video.

The ID called "Nov" became popular in the domestic interstellar circle almost overnight.

The conversation in the circle revolves around this mysterious and powerful Terran player, and they are all trying to find the real person with this ID.

……

()

Chinese Net