If there aren’t enough people, AI will join in? "Who is the undercover" can still be played like this

AIGC is soaring all the way and is making waves in the pan-entertainment and social field. The introduction of AI, on the one hand, promotes the change of social connection methods, extending from the connection between people to the connection between people and virtual people, creating more imagination space for social gameplay; on the other hand, AI can further enhance social stickiness, as a A strong empathic and selfless social partner, providing the most important emotional value in social interactions and reducing interpersonal friction.

For a long time, SoundNet has continued to invest in the field of AI. For example, capabilities such as AI noise reduction and AI echo cancellation are based on SoundNet’s self-developed AI inference engine. Recently, Shengwang has also combined some applications in social gameplay to help developers use large models to achieve scene innovation, such as the training and polishing of customized models based on specific scenarios, the multi-modality of large models, especially the large-scale voice modality. Model exploration.

GPT X, the exploration of new gameplay

AIGC (AI Generated Content) can also be called generative AI, such as AI text continuation, text-to-image AI diagrams, AI hosts, etc., all fall under the category of AIGC. GPT (Generative Pre-Trained Transformer) is an Internet-based deep learning model for text generation that can be trained with data. GPT is a type of AIGC technology. Recently, digitalization, enterprise services, finance and other industries are exploring integration with GPT. It will also be an industry trend to integrate pan-entertainment social products into GPT gameplay. For example, virtual anchors, board games, and some proxy chat scenarios in social products are all suitable for combination with large language models.

Today we will take a look at the latest results of Shengwang’s exploration in the above fields, the combination of GPT and “Who is the Undercover” gameplay.

Who is the undercover is a classic multiplayer social reasoning game. At the beginning of the game, n (n≥3) people get the same word, and the remaining people (undercover) get another word related to it. . In each round, each person can only use one sentence to describe the word he got. He cannot say it directly, not only to avoid being discovered by the undercover, but also to give indirect hints to his companions. After each round of description, n people vote for the person suspected of being an undercover. The person with the most votes is eliminated. If there is a tie vote, the next round of description is entered. If there are only three people left (including the undercover), the undercover wins.

Because there are certain requirements for the number of participants, and this type of game has a more interactive atmosphere when there are more people. Based on this pain point, Shengwang launched a new gameplay of GPT: Who is the undercover? GPT acts as one of the players, learns and refines the speech formats of other players in the game, describes their own keywords, and can analyze them through Other players speak, make a judgment about "who is undercover" and vote.

Game scene display

In the game, we have encountered AI simulated as an undercover agent for testing. The results show that the AI ​​performs well in the game and even has the ability to win the undercover status.

The successful exploration of this mode can be applied to more game scenarios such as werewolf killing, script killing, etc. in the future. At this stage, the application of AIGC in the social industry is in full swing, and various "AIGC social" gameplays have also appeared on the market. For example, AI intelligent question and answer customer service can guide novice players in how to play the game; AI companion chat can interact with players one-on-one 1. Chat, which plays the role of listening and accompanying; AI host, can learn the rules of games and other topics and guide players to play games or chat.

In the audio and video social scene, Shengwang is also actively exploring more forms of AIGC X, giving AI different roles suitable for different gameplay and scenarios. For example, in e-commerce live broadcasts and show live broadcasts, AI can be used as a question and answer assistant to help anchors respond to some fans' questions in real time; in social applications, AI can be used as a chat assistant to bring more interaction and companionship to real users; and in social applications, AI can be used as a chat assistant to bring more interaction and companionship to real users; In game scenarios, AI can chat, interact and participate in games with real players, bringing a richer gaming experience.

4 major technical advantages bring a new social experience

1. The real-time interactive capabilities of Shengwang are combined with the Yuanverse and GPT gameplay

· Extremely low latency: Adopting the real-time interaction capability of Acoustic Network, the signaling delay is as low as 30ms, and the end-to-end voice delay is as low as 65ms, bringing a "face-to-face" immersive experience and reducing the delay of the overall AIGC solution. Let AI "react" faster

·Leading voice technology: Through Acoustic Network AI echo cancellation, noise reduction and other technologies, GPT can "hear" clear speech even in a noisy background environment and respond accordingly

· TTS and STT recognition is more accurate: Even in an environment with up to 80% packet loss, accurate human voice recognition and sentence segmentation can be used to make speech-to-text (STT) more accurate and text-to-speech (TTS) more accurate. Clear and smooth

·Currently, many developers are using IM to allow players to interact with AI characters, and are connected to Agora RTC. Various voice interaction methods make the gameplay more interesting and immersive

2. Scenario-based professional training capabilities

· Rich AI personality and style: Shengwang can meet the needs of users for personalized styles of AI characters in different pan-entertainment scenarios

· Fast model training: According to needs, Shengwang can improve and optimize different scene models in as fast as 1 week

· Professional scene construction capabilities: Based on Shengwang’s years of experience in the pan-entertainment social industry, combined with the knowledge and capabilities of GPT algorithm experts, it can output more interesting AI audio and video social scene gameplay

3. Have the advantages of GPT large model, STT, TTS and other partners

· Shengwang cooperates with well-known domestic manufacturers to provide relevant capabilities and can be freely replaced as needed in the plan, avoiding the pitfalls of enterprises themselves and improving the overall online speed

· In terms of using GPT large models, Shengwang will select appropriate large models according to customer scenarios, and create specific model databases and prompts according to different scenarios. At the same time, it will combine the low-latency characteristics of Shengwang RTC to allow users to interact The delay is optimized and the model output is more in line with the real scene

4. Combined with rich metaverse gameplay

Shengwang can provide voice-driven AI images in various models such as realistic and cartoon, so that AI players and assistants are no longer monotonous, and real players and AI can entertain together.

You can currently experience the DEMO of who is an undercover agent in AIGC, and other interesting AIGC social scenarios are also being rapidly developed!

The above is the detailed content of If there aren’t enough people, AI will join in? "Who is the undercover" can still be played like this. For more information, please follow other related articles on the PHP Chinese website!