AI Weekly News: Ma Huateng said that AI is a once-in-a-century opportunity, OpenAI uses iOS to lock in user stickiness, and embodied intelligence allows AI to perceive the real world | Insight Research

Starting next week, AI Weekly Insights will be updated daily - the Daily AI Insights column. Everyone is welcome to continue to follow Wall Street Insights and Wisdom Research.

New additions to AI news this week—New perspective of news

Weekly News

Summary of this week’s highlights:

1. Ma Huateng said that AI is comparable to the electric power industry revolution; Meituan is expanding algorithmic recruitment and quietly developing large models.

2. OpenAI releases the iOS version of chatGPT, opening 70 plug-ins to Plus users

3. Meta releases the AI ​​chip - MTIA, which will take 25 years to come out. It will still use NVIDIA GPU.

4. A new milestone in AI drawing - DragGAN enables an elephant to turn around and the car to "convert" with one click.

5. Embodied intelligence creates AI active perception, the next wave of artificial intelligence.

6. Yuncong Technology releases large-scale models. The commercialization path in vertical fields is the opportunity for domestic large-scale models.

7. AI black technology - you can experience Disney's "Beyond the Horizon" at home; the semi-mechanical "Spider-Man" subverts the perception of human-computer interaction.

New Perspectives of Seeing News

At Tencent’s 2023 shareholders’ meeting, Ma Huateng said: “At first, everyone thought that AI was a once-in-a-decade opportunity for the Internet, but now the understanding of AI has risen to a century-old development opportunity, which can be compared to the electric power industrial revolution.” Tencent Currently, we are also immersed in the research and development of AI technology, but we are not eager for short-term success. In the future, we will create more value in the application and content ecology. We will not only focus on the to-C side, but also attach importance to the to-B side opportunities.

In addition, Meituan is secretly developing large models and has been laying out the field in early March. Recently, the algorithm team is also expanding, and it is also planning to establish a separate "platform department" to help Meituan's large models pass specific business implementation.

Jianzhi Research believes: The current competition among large models is very intense, and the emergence of many open source large models has accelerated the involution. However, the problem with open source large models is that they are difficult to commercialize and are mostly used for academic research. However, if overseas closed advanced large models are used in some key fields, there will be security risks.

Therefore, the trend of developing domestic large models lies in the richness of the Chinese prediction library, strong localization advantages, and high security and confidentiality. In the future, the market demand for Chinese-specific large models will very high.

What deserves special attention is the commercial value of combining large models with applications. Whether it is openAI, Microsoft or Google, they have successively begun to expand their ecological territory. This is also the inevitable path for the development of domestic AI. R&D results must eventually be realized and generate greater commercial value.

Breaking Release

1. OpenAI releases the iOS version of chatGPT, opening 70 plug-ins to Plus users

OpenA officially launched the iOS version of chatGPT this week. Users need to use iOS 16.1 or higher operating system version. And promises that an Android version will be released soon.

ChatGPT on the mobile phone supports synchronizing user history records across devices, and also integrates OpenAI’s open source speech recognition system Whisper. Users can input content using voice; it can perform question and answer, language translation, educational coaching, and automatically generate text.

In addition, ChatGPT opens the networking function to PLUS users, allowing the use of 70 third-party plug-ins.

Jianzhi Research believes: Whether it is the promotion of mobile applications or the use of open third-party plug-ins, these are OpenAI's efforts to improve user stickiness and further achieve user sinking.

Opening the mobile terminal will greatly increase the frequency of user use, because it is more convenient and easier to use than the PC terminal. Since the launch of ChatGPT, users have been wanting to use ChatGPT on mobile devices. The commercial value and daily active volume of ChatGPT will reach new heights again with the opening of the mobile terminal. In addition, as the number of visits increases, the demand for computing power will further expand.

In addition, although third-party plug-ins are currently only open to PLUS paying users, judging from the current degree of AI involution, it will be just around the corner to be fully free.

2. Meta releases AI dedicated chip-MTIA

MTIA is a programmable chip designed for training and inference. Its launch has greatly enhanced Meta’s hardware strength in the field of artificial intelligence. In the end, the competition between technology giants cannot escape the core hardware. Especially in the era of developing AI, computing power level is the cornerstone of development. If computing power cannot be mastered, the development process will inevitably be controlled by "others".

But MTIA still has a lot of room for optimization, and it is expected to wait until 25 years before it comes out. In terms of NNP and GPU performance tests, MTIA has better performance on low and medium complexity models, but it is still far behind GPU on high complexity.

Jianzhi Research believes that Meta develops AI chips for the long term. After all, chips are the core hard power in our hands. However, the road to high-performance chip development is very long. The design of this chip It also started as early as 2020. At present, Meta will still use NVIDIA GPUs. After all, in 2022, Meta just carried out a disruptive design for its data center to introduce NVIDIA GPUs. In the future, it will mainly rely on the RSC supercomputing center to develop AI.

3. A new milestone in AI drawing-DragGAN realizes all imagination

DragGAN completely breaks the exclusive position of the Diffusion model in the field of AI drawing. The paper titled "Drag Your GAN" has detonated the AI ​​drawing circle. The paper was jointly published by scholars from MPII, MIT, Penn State, Google and other institutions, and has been accepted by SIGGRAPH2023.

This model can meet almost all people's needs for photo editing. It can change the object shape, details, and even the direction and layout. It can be called a nuclear bomb-level Photoshop.

Users only need to set a few operation points (red points) and target points (blue points) on the photo, and then drag and drop to generate a new image.

Jianzhi Research believes that: The emergence of DragGAN shows that machine training in image learning has reached a new level. It is worth noting that DragGAN has more powerful generalization capabilities and can create images that exceed the training data. For example, the shape of the lion's mouth has been completely changed. This is basically newly generated content, rather than the modification that people originally thought. graph function.

Compared with previous methods, DragGAN does not rely on modeling or auxiliary networks in specific fields. Instead, it uses a general framework, uses GAN to identify image quality, and uses point tracking to complete image deformation. Function. With this powerful function, videographers and photo retouchers will have a lot of fun.

4. Embodied intelligence creates AI active perception, the next wave of artificial intelligence.

At the ITF World 2023 Semiconductor Conference, NVIDIA CEO Jensen Huang made another bold statement that the next wave of artificial intelligence will be embodied intelligence.

Jianzhi Research believes that:The value of AI brought by embodied intelligence is far greater than that of humanoid robots. The greatest characteristic of embodied intelligence is the ability to autonomously perceive the physical world from the perspective of the protagonist, learn using an anthropomorphic thinking path, and thus provide behavioral feedback expected by humans, rather than passively waiting for data to be fed. Among the five major human senses, vision accounts for more than 80% of the information acquired, and it is also very important for machines to understand human language. Therefore, machine vision and multi-modal large models are the two keys to unlocking machine self-perception learning. For details, see What is NVIDIA’s popular “embodied intelligence”? The value of AI is far greater than that of robots.

5. Yuncong Technology releases the large model of Congrong

Yuncong Technology, an artificial intelligence platform company, released the Congrong model in Guangzhou and demonstrated its basic abilities such as dialogue, programming, reading, and answering real questions in the high school entrance examination. The large model is currently in the internal beta stage. This model is a large Vincentian model and cannot yet complete the functions of multi-modal large models such as Vincentian diagrams.

Performance in the open test: The response speed is fast, but the content accuracy needs to be improved. Moreover, the timeliness of the database is relatively low, still 21 years old. In addition, the model's performance in mathematics and reasoning capabilities has not yet reached expectations.

Jianzhi Research believes that:The advantage of domestic large models is that the richness of the Chinese corpus is much higher than that of foreign advanced large models. Although it is difficult to catch up with ChatGPT in terms of leadership, the Congrong Big Model will take the lead in the application development of vertical industries in the future, especially in the development of exclusive industry models in the fields of finance, government affairs, and manufacturing, and is committed to the commercialization of models. Realize.

AI Black Technology

1. You can experience Disney’s “Beyond the Horizon” at home

Foreign developer Nils Bakker successfully created a "virtual space transmission" system using ChatGPT, using Unreal Engine 5.1 ChatGPT Google Maps 3D Tiles API. Users only need to enter the location, and the system will take you from a first-person perspective. Overlooking the beautiful scenery around the world, this is the time to experience the joy of flying over the horizon at home.

Combine the APIs of Google 3D Tiles and ChatGPT, and then use the capabilities of Unreal Engine to allow users to experience space travel immersively. Now you can feel the charm of flying over the horizon while lying at home.

Jianzhi Research believes that: AI is still in the early stages of industry development, imagination and creativity are very important, and industry tracks and business opportunities will spring up like mushrooms after a rain.

2. The semi-mechanical "Spider-Man" is here

The Japanese robotics company Jizai Arms has designed a spider-like robot limb system that allows humans to have freely controllable robotic arms. The system consists of six arms that can be controlled by the user wearing them. Up to four robotic arms can be installed. What is noteworthy is that this system changes the way human-machine interaction is done.

The prosthesis is very flexible and can perform a variety of tasks. Its applications range from warehouses to hospital operating rooms. In the future, it can help improve the quality of life of disabled people.

Jianzhi Research believes: The "fusion" of robotic arms and real people opens up the imagination space of human-machine integration and refreshes the upper limit of people's understanding of robot development. There will be more impossibilities in the future be realized.

What to watch next week

Looking forward to OpenAI’s first open source large model, can it rewrite Meta’s status as the open source king?

The above is the detailed content of AI Weekly News: Ma Huateng said that AI is a once-in-a-century opportunity, OpenAI uses iOS to lock in user stickiness, and embodied intelligence allows AI to perceive the real world | Insight Research. For more information, please follow other related articles on the PHP Chinese website!