Reporting by XinZhiyuan
EDIT: LRS
In the era of rapid development of artificial intelligence, due to the great improvement of the level of technology, how to use AI to do some innovative landing applications has become more and more important. No, someone summed up a few interesting ideas for 2021 and let's see how AI can change lives!
There is still a long way to go to truly understand the future direction of artificial intelligence.
——Ray Kurzwei, 2009
In 2018, the United States Phoenix launched a driverless car, opening the curtain on the smart car track.
In 2019, MIT's AI system successfully discovered an entirely new antibiotic, Halicin, in just a few days, while it usually takes years to decades for humans to develop an antibiotic.
In 2020, Google built an open chatbot Meena with its Transformer, and the openAI next door also released GPT-3, taking human-machine dialogue to a whole new landscape: AI finally stopped talking nonsense.
The researchers later tested the language model on intelligence, and the results were on par with humans.
By 2021, the development of AI technology is exploding, and dystopian movies such as "The Matrix" that describe the proliferation of technology and human-machine contradictions now seem to be gradually obsolete.
In the past year, we have gradually been filled with a variety of AI black technology, tailor-made music, short videos, movie recommendations, more and more smart home devices, and search engines that gradually read your heart, and every technological development will be reflected in life.
Recently, Dr. Alan D. Thompson, Chairman of Former Mensa International, AI expert and advisor, summarized some of the innovative ideas from the AI community in 2021 to see how AI can change the world!
Microchip design
Google Brain published a paper in Nature in June demonstrating AI's ability to design new chips that can be laid out in a grid to plan where individual components in a computer are placed.
The main difficulty in chip design is that the state space is too large, if there are 1000 nodes in a grid, the state space may be as high as 10 to the 2500 power, let people design, it is really to develop to the ground.
The theoretical findings in the paper have been used in Google's current version of TPU design.
Software programming
In June, GitHub announced the introduction of AI technology into programming with the release of Copilot.
Trained on a large number of open source repository codes, the model has the ability to automatically complete and generate multiple lines of code, and supports a very large number of languages, including Java, C/C++, C#, Python, JavaScript, Ruby, Go, and more.
A GitHub survey later confirmed that for some programming languages, about 30% of the new code uploaded is automatically generated by the GPT-3-powered Copilot.
Of course, this is not all good, and some have compared the invasion of GPT-3 to the first nuclear bomb dropped by humans in July 1945. The air in the atmosphere at that time was almost filled with radiation, as if the code was filled with code automatically generated by AI.
Although the air has gradually improved since then, it is still uncertain whether AI will be a programming helper or atmospheric radiation.
Before March 2020, the data on the Internet was basically created by humans, and the emergence of GPT-3 has changed the pattern of data generation.
Voice sentiment analysis
Another important ability of AI is to infer the hidden real state behind it according to the surface phenomenon of the data, and an interesting application scenario is to infer the real psychological state of the speaker at this moment based on the person's body language and the way of talking.
For example, when CEO is unscripted, open board meetings, and other impromptu conversation scenarios, they may say something completely different from what they thought, and they can't disguise it at all.
When the AI model was backtested against Enron CEO Ken Lay, it was found that the model was really useful!
On August 16, 2001, the CEO still delivered an impassioned speech at the staff meeting: "I am very happy to be back, although we are facing many challenges now, but the hardest time is over, and the business is developing well."
Three months later, Enron declared bankruptcy.
AI models can easily find loopholes by characterizing the tone, rhythm, emphasis, and phrases of the spoken words he speaks, and it is expected to introduce AI models to analyze emotional states in speeches and interview transcripts in the future.
It's like a real chatbot
Since GPT-3's language generation performance is still very amazing, if you combine chatbots and avatars, isn't it as if you are chatting with real people?
In August 2021, Leta, a combination of GPT-3 and Synthesia.io avatar technology, was unveiled at the World Gifted Conference.
The two IIs are chatting and look like human video calls. Give a theme of "how does it feel to do artificial intelligence", if you don't turn off the video, they can keep nagging.
https://www.youtube.com/watch?v=JcAY-ae2Drw
Stylish design
Compared with foreign artificial intelligence technology, domestic production is also high.
The Alibaba M6 model (Multi-Modality to Multi-Modality Multitask Mega-transformer) can be trained on a clothing database to creatively design some clothing.
Want Hepburn? Arrangement.
Even men's dresses can be designed for you!
As well as the T-beam of the two-dimensional wind, it can also be easily done. Well? It's not the "beware of that two-dimensional" wind.)
All elements of each piece of clothing, from design, pattern, color, are created or selected by the AI model, and even the product text description next to each piece is generated by the AI model.
And the model design of the clothes can already be bought on the mainstream online shopping platform.
Virtual characters
Since AI is called artificial intelligence, then as technology becomes more and more advanced, can humans really create a human being artificially?
The concept of virtual characters came into being.
On June 1st, Hua Zhibing, driven by Wudao 2.0, was officially unveiled in Beijing and entered the Knowledge Engineering Laboratory of the Department of Computer Science and Technology of Tsinghua University, joining Tsinghua University under the tutelage of Tang Jie.
Hua Zhibing can compose poetry, paint, create music, and also has a certain ability to reason and interact emotionally. Unlike ordinary virtual digital people, Hua Zhibing has continuous learning ability and can gradually "grow up", including continuous learning from texts, visions, images, videos, etc., just like humans can constantly learn behavior patterns from the things they experience around them.
Over time, huazhibing's new abilities learned for new scenarios will be organically integrated into its own models, thus becoming more and more intelligent.
And never age.
Resources:
https://lifearchitect.ai/the-sky-is-on-fire/