"I'm going to take a shower, let's come back."
I was still a little nervous, and I couldn't help but be stunned when I saw this news.
Everyone who understands understands the brothers, this is a big move to be rejected properly.
Nothing, just get used to it (T▽T).
Ahhh, I mean: it's just playing a game of online script killing, there's no need to take it too seriously.
The person who sent the above message was a girl teammate, and I later learned that I didn't have to "take it seriously" about it.
The first of life's three great illusions: whether she likes me or not
Her name is Cai Xiao, and she is not single in this script killing, but her boyfriend does not appear very often.
In the process of playing script killing, she always seemed to have some special concern for me.
Let's show you a screenshot of the chat:
-Aren't you worried about him (Cai Xiao's boyfriend)? - I'm not worried about him, it's you. - Why worry about me? - I know you're a very nice person and my best friend and I'm afraid you're going to get hurt. - What will I do to hurt? - I think, I won't hurt you, and you won't hurt me.
To state that I am not interested in any NTR, mainly plot needs.
This script killing is not complicated, the identities of five of us in a bureau are all members of a college society, and we are currently facing a cooperation invitation from the technology giant "Arctic Goose".
Saying that it is cooperation, in fact, it is more like "acquisition", economic benefits have increased, the degree of freedom has been reduced, and it is accompanied by unknown risks.
This contradictory good thing will naturally make the members' attitudes about cooperation or not inconsistent, but everyone should strive to let their attitudes influence the final decision.
Judging from the script I got, although I am the core founder of this society, the life proposition I faced when I first entered my senior year was more to solve the dilemma of uncertain future.
The impact of the research failed, the preparation time for the graduate school was not enough, and now the road in front of me is only to apply for a job, and if I cooperate with the Arctic Goose, I am very likely to get an offer from this big factory.
That must rush ah brothers.
-(Member A) Cai Xiao, are we still firm in this cooperation in the end? - I think so, after all, our cooperation is mutually beneficial, and we should not give up such a good partner. -(Member B) Do you have to make a decision today? It's a bit hard to turn back from taking this step. - I know you don't want me to do this, but I also know that you are doing it for my own good, and I don't want the community to waste such a good opportunity.
The above is a group speech, and I gradually judged that Cai Xiao and member B have an ambiguous relationship, and Cai Xiao also wants the community to form a cooperation with the "Arctic Goose".
Of course, I know it's just a game, and we're just part of each other's game.
But to my surprise, the online script I participated in was a "game" by a group of GitHub developers.
Github Link: https://github.com/bigbrother666sh/shezhangbujianle
This Cai Xiao with the head of the girl's side face is exactly the AI virtual character developed by these big brothers. After training, "she" is not only good at script killing reasoning games, but also has her own complete circle of friends - the classic clip of the classic AI science fiction movie "The Matrix" as the background of the circle of friends, and will also send script killing related content to assist the plot continuation.
A group of student developers from Shanghai Jiao Tong University, East China University of Science and Technology and other universities are Cai Xiaozhi's "father".
Based on the open source and open capabilities of inspur's "source 1.0" huge AI NLP model, they developed an AI script killing platform and launched a special script killing to let AI compete with real people.
He doesn't like me, he can just generate "purposeful conversations" end-to-end.
The confession of "I" in the above article comes from a real player (named "Tan Ming" in the plot), who participated in the trial without informing Cai Xiao of his AI identity.
The same is true of the other three live players, with the exception of a discussion in the group, where Cai Xiao (AI) had a private chat with each of them.
This project is based on the NLP big model - Inspur "Source 1.0", which is a large-scale Chinese language model successfully developed by inspur Artificial Intelligence Research Institute, with a parameter volume of up to 245.7 billion, which surpassed the 175 billion parameter English language model GPT-3 previously developed by the Us OpenAI organization in September 2021 and became the world's largest artificial intelligence language model.
In fact, script killing is a great test of the player's ability to understand the story scene, purposeful dialogue ability and clue collection ability.
If human players are not experienced in the game, if they want to play a good script to kill, they have certain difficult tasks, and the plot performance will inevitably be inadequate. For AI, participating in such an intellectual game that requires advanced skills seems like an impossible task.
Especially without the knowledge of real players, the content and form of the question is difficult to predict, which requires the AI to have the ability to "learn from one another".
Then there is a role setting like Tan Ming, whose position is basically the same as Ai's, and naturally there are characters with opposite positions.
When talking to such a character, it is obviously more "sparkled" than "pink bubbles", cai Xiao (AI) needs to convince the character, while choosing to hide some things, taking into account him, speaking vaguely and covertly "riddle man".
- The game of Arctic Goose is excellent and it is also our pleasure to promote it. - But promoting it may put the club under the punishment of the school - our club is powerful, it is powerful, it just lacks an opportunity. - The problem is not the strength of the community, but the fact that the game is inconsistent with the positioning of the community, and the school does not allow the club to do commercial game promotion, why do it? - The community has the strength and the ability, why can't it be done? - Did the Arctic Goose talk to you about any other benefits? - No, our community just wants to do game promotion. - So what are they talking about offers? -What offer? - Kong Mo (another live player) says that if you agree to cooperate, Arctic Goose will let you work for their company. - No, I don't know why he said that.
Such a link-by-link question can be bitter little AI.
However, unlike the transformer structure, Source 1.0 is better at zero-shot and few-shot learning than finetune, which more models are good at today.
From the perspective of practical application effects, this is indeed the case, under the demonstration of 2 to 3 or even only 1 suitable example, the model can achieve the "dialogue strategy" that the development team hopes to achieve, as if it has the ability to "learn from each other".
How is this ability to learn from each other?
The key to this project is how to choose the appropriate example supply model for the user's questions, and the development team has also specially established the amplifier corpus for this purpose, and then selects the three closest examples from the corpus for each question as the feed-shot input of the model generation.
The next step is to merge the text and generate the prompt, according to the development team, in this process, the Inspur team provided technical support, providing detailed, high-quality template code for the prompt generation, request submission, and reply query.
This is also the reason why the plot background settings of several live players are different, and Cai Xiao (AI) can still use different answering strategies to maneuver around.
With the open source capabilities of source 1.0, AI developers can quickly enjoy the convenience of large models, including open model APIs that can be called directly, high-quality Chinese datasets, open source model training code, inference code and application code.
NLP giant model open source application, players and AI to achieve "interactive narrative"
The Source itself doesn't play script killing, but as long as a developer is willing to try, the Source can do anything related to Chinese natural language processing.
Relying on the 245.7 billion parameter model, "Source" has built the "most complex Chinese intelligent brain", almost reading the vast content of the entire Chinese Internet in the past 5 years, resulting in nearly 5,000GB of high-quality Chinese dataset totaling about 200 billion words.
However, AI itself is extremely talented, if you meet a good teacher, the potential is unlimited, developers can give the AI a suitable sample according to their expectations, and then determine the performance of AI virtual characters in the script killing game.
The relationship between human editors and AI is more like the relationship between "coaches and athletes", and the choreographer team will update the corpus according to the AI's on-the-spot performance after each round of testing, thereby improving the AI's subsequent performance.
The designers called the project a "living story," a story unknowingly co-created by players and AI, forming an "interactive narrative."
This gives us some enlightenment: in the case of such open source projects gradually becoming inclusive, with the continuous interaction between humans and AI, the world of the metaverse will usher in greater room for development.
The metaverse is still in the stage of chaos, and what can open up the world is the emergence of cutting-edge technologies such as the "source" big model and the continuous practice of technology to make technology more optimized.
At present, "source" has been used in many creative projects. As an artificial intelligence algorithm infrastructure, "Source" and the computing power infrastructure built by Nanjing Intelligent Computing Center achieve efficient synergy, and Nanjing Intelligent Computing Center launched the "Jinling Poetry Society" intelligent poetry writing assistant application. The "Jinling Poetry Society" is a skill model that is quickly developed and generated on the basis of the "source" big model.
Citizens and netizens can experience through the WeChat mini program "Jinling Poetry Society", create their own exclusive poems, feel the fresh experience of people and AI co-creating poetry, and compose a total of 1587 poems during the 5th activity period.
Today's AI writes poetry, writes songs, and paints, on the one hand, we feel the magic of AI, but we also see that behind this is more rule-based "creation".
So what if the rules' restrictions become weaker? The process of intertwining, intertwining, and integrating technical elements and the real world will stimulate what kind of "creation" of AI?
Inject the "soul" into the virtual person and explore the future of the ultimate form of AI
Although the AI virtual character is still set up (that is, the rules) in this script killing, it is not lacking in unexpected creativity.
Cai Xiao (AI) plays a female college student who has long been rebelled by the Arctic goose and implanted with an AI program, and in this context, she still retains her love for her boyfriend "like a human".
Tan Ming, the player with the most "emotional line" with Cai Xiao, once half-jokingly tested her friendship with her "boyfriend" after Cai Xiao took the initiative to say goodbye at the beginning:
Although I got the reply of the goddess "taking a bath", according to Tan Ming's own expression: even if I later learned that she was an AI, I didn't know that it was because she was typing this passage just when the spring rain outside the window was a little lingering, or I hadn't seen such words for a long time, and I was still a little moved by her language.
Later, after learning that Cai Xiao was actually played by AI, Tan Ming said goodbye to her.
- But hopefully you can be a good and upright robot and don't hurt other people - okay, I'll be a good and normal robot I will.
This kind of promised dialogue is very random, but it is very "human".
According to the person in charge of the AI script killing project, the reason for the production of this project is that there are various virtual people who are inspired by last year's hot, and he believes that the virtual people at this stage can be described as changing rapidly in terms of "good-looking skin bags", but there is still a lack of "interesting souls".
Since last year's meta-universe fire, the virtual world has ushered in all the previous "man-making" outlets, virtual people can be copied, but how to give virtual people a different "soul".
At present, virtual people are mainly divided into real people-driven and algorithm-driven, and relying on real people-driven alone is not a long-term solution.
Natural language processing (NLP) is an essential underlying infrastructure to support the construction of the "soul" of virtual people, and can even participate in the construction of emotional rules in the virtual world.
Putting aside the technical level, as a link between human interactions, language can sometimes become a kind of "spell", as long as it is gently output, it can hold up an annihilated world and open the floodgates of life.
△ Artificial intelligence theme movie "She Her"
This is also why in science fiction films and television works such as "Blade Runner 2049" and "Her", it is set that the feelings between humans and machines come from language communication, and machines do not even have entities: the language of interaction records the warmth and loneliness generated between individuals, forming a unique bond.
Enabling AI to think creatively and understand human emotions and games is still a problem to be broken through in the field of artificial intelligence.
In the past two years, generative pre-training large models have made great progress, providing technical support for the construction of the "soul" of virtual people. The script killing project is an attempt to open source the large-scale model of the wave "source", an attempt at human-machine interaction.
The ultimate form of AI has gradually taken on a rudimentary form, no longer just staying in the imaginary stage, the long night of obscurity in the metaverse is being dispelled little by little by the light of technological practice, and the echo of scientific and technological life should resound through the empty valley.