An overview of artificial intelligence in today's tech companies
In movies, artificial intelligence gains self-awareness one by one, movies about artificial intelligence: Space Odyssey, Terminator, Matrix, Agent Smith, Prometheus, Westworld and many other movies, let us see to the power of artificial intelligence. With the power of artificial intelligence, we pay more and more attention to the question of when artificial intelligence will awaken, whether artificial intelligence will be like the movie, whether artificial intelligence will really awaken, this has been accompanied by the development of artificial intelligence technology.
The Washington Post reported on June 12 that Google engineers believe that Google's AI (LaMDA) has come alive and has a certain degree of consciousness. He showed evidence that LaMDA is sentient. Attempts were also made to make Google executives aware of AI's sense of "personality," but a Google VP investigated his claims and dismissed them.
Overnight, news of Google's artificial intelligence awakening became a trending topic on Twitter. Some people are excited, cheering for the great leap in human technology; some people are afraid, worrying that if the "Skynet" wakes up, the Judgment Day of the Terminator will definitely come.
LaMDA2 at Google I/O '22 Keynote
Whether LaMDA is conscious, we still need more evidence and information. As developers in the machine learning and AI industries, it is necessary for us to sort out the current development of various technology giants in the field of artificial intelligence. That house's AI will wake up first.
For the sake of comparison, we will analyze the AI level of major technology companies from the following aspects.
Talent: Who leads the company's AI team.
Research capabilities: the number of excellent papers published by the company and the excellent models established.
System Capabilities: AI Systems and Software
Computing capabilities: supercomputers, AI chips, hardware, etc.
Datasets: Large Datasets
Engineering Capabilities: The ability to translate research into products
Google Brain (Google Artificial Intelligence)
Talent:
Jeff Dean and Geoffrey Hinton lead the Google Brain team, and many of the company's AI leaders have worked at Google.
Jeff Dean - head of artificial intelligence at Google; famous for MapReduce, Spanner, big table, tensorflow, etc.
Geoffrey Hinton - Known as the "Godfather of Artificial Intelligence" and "Godfather of Deep Learning"; Turing Award (2018) with Yoshua Bengio and Yann LeCun; known for Backpropagation, Boltzmann Machines, Deep Learning and Capsule Neural Networks Famous; famous students: Yann LeCun, IIya Sutskever, Ruslan Salakhudinov…
Research capacity:
Many well-known models come from Google, such as Transformer, BERT, LaMDA/LaMDA 2, the protagonist model of the AI awakening event, and PaLM, Google's 540 billion parameter model.
Transformer is a deep learning model from the Google paper (Attention Is All You Need). Transforms were originally used in natural language processing and are now widely used in various tasks. Many models such as BERT, GPT, LaMDA, and PaLM are based on Transformer.
更多Transformer模型VIT 模型SWIN Transformer模型参考头条号：人工智能研究所
Here, the data of ICML, NeurIPS and ICLR (top3 machine learning conference) in 2021 are selected as examples.
ICML 2021 Data Analysis by Sergey Ivanov
NeurIPS 2021 Data Analysis by Sergey Ivanov
sharonzhou ICLR2021 stats
In 2021, Google has 109, 177, and 197 papers accepted by ICML, NeurIPS, and ICLR, ranking first.
Calculate ability:
TPU: Tensor Processing Unit is an artificial intelligence accelerator application-specific integrated circuit (ASIC) developed by Google specifically for neural network machine learning.
System Capabilities:
TensorFlow is a free and open source software library for machine learning and artificial intelligence. It mainly focuses on the training and inference of deep neural networks. TPU is tailor-made for TensorFlow.
Google Cloud Platform (GCP) - Google Cloud Computing Services
data set:
YouTube-8M segments、open images、google landmarks、natural question等。
Engineering Capabilities:
From Goole I/O 2022, we can see that almost all Google products, whether it is Google Translate, Google Maps, YouTube, or the unreleased Google AR glasses, are based on Google AI. Waymo-Google's self-driving car project is also Google's most important project at present.
It can be seen from the above that Google AI has comprehensive strength. The Google Brain team focuses on machine learning, computer vision, robotics, natural language, health, and more.
DeepMind
DeepMind is a subsidiary of Google parent company Alphabet Inc. Headquartered in London, DeepMind is not a "big" tech company, but it has more influence in the field of artificial intelligence than any other big tech company. For example, AlphaGO defeated human professional Go player Lee Sedol in 2016 with its star product, and the AlphaFold large model that made significant progress in protein folding in 2020. These impressive scores bring AI to market.
Talent:
Demis Hassabis (CEO), David Silver (Principal Research)
Ian Goodfellow: Known for Generating Adversarial Networks (GANs)
Research capacity:
38 (ranked 3) papers were accepted by ICML 2021, 81 (ranked 3) papers were accepted by NeurIPS 2021, and 42 (ranked 4) papers were accepted by ICLR 2021.
Engineering Capabilities:
AlphaFold, AlphaGo / Alpha Go Zero / AlphaZero / MuZero Go, AlphaStar
Meta AI (Facebook AI)
Talent:
Yann LeCun - Known for Deep Learning; Turing Award (2018) with Yoshua Bengio and Geoffrey Hinton; Called "Godfather of Artificial Intelligence" and "Godfather of Deep Learning".
Research capacity:
36 (rank 4) papers were accepted by ICML 2021, 78 (rank 4) papers were accepted by NeurIPS 2021, and 55 (rank 3) papers were accepted by ICLR 2021.
In recent years, many practical models, such as the new design paradigm RegNet, the feature pyramid network FPN for target detection, the new target detection method YOLO, etc., are all developed by Facebook AI Research (FAIR)
System Capabilities:
PyTorch - An open source machine learning framework based on the Torch library for computer vision and natural language processing, mainly developed by Meta AI.
Tesla Autopilot, Uber's Pyro, and HuggingFace's Transformers are all built on top of PyTorch.
Calculate ability:
The AI Research SuperCluster (RSC) is one of the fastest AI supercomputers running today and will be the fastest AI supercomputer in the world when fully completed in 2022.
data set:
Casual Conversations Dataset、Deepfake Detection Challenge 等。
Engineering Capabilities:
FAIR (Meta AI) meta AI is currently mostly developing the Metaverse.
The Meta AI team led by Yann LeCun is very good, especially in long-term AI research. Its research covers computer vision, conversational artificial intelligence, natural language processing, human-computer intelligence, reinforcement learning and other fields.
As meta shifted the company's strategic focus to the metaverse this year, Facebook AI Research underwent a major reorganization, and Facebook AI Research (FAIR) became the core business unit of Reality Labs.
OpenAI
OpenAI is an artificial intelligence research and deployment company that is very similar to DeepMind and is considered a competitor of DeepMind. Its AI talent team is led by Ilya Sutskever, chief scientist at OpenAI and co-inventor of AlexNet, and John Schulman.
OpenAI is also a company with a relatively large impact on AI (DeepMind is another). GPT-3 released by OpenAI in 2020 is one of its star products. Many popular applications are built with GPT-3, such as GitHub Copilot, Duglingo, and Keeper Tax.
Another product released by OpenAI, DALL·E 2, is also on fire. It can create realistic images based on descriptions in natural language.
Received a US$1 billion investment from Microsoft in 2019. Perhaps Microsoft will acquire OpenAI in the future. Sincerely hope that OpenAI can continue its strong upward trend in AI research
Microsoft AI Research
Microsoft Research was established in the early 1990s. In the early days, Microsoft Research did a lot of excellent work and trained a lot of talents, many of whom went to Google and Facebook later. Microsoft Research was surpassed by other giants in the years when big data, machine learning, and deep learning exploded. This is partly due to the increasing age of the researchers. In 2014, Satya Nadella succeeded Steve Ballmer as Microsoft's new CEO. He led Microsoft Azure to great success. Since then, Microsoft has increased its investment in artificial intelligence. Microsoft Research has made significant progress in artificial intelligence and has infused related results into its products, including Bing, Cortana, Microsoft Translator, and more. In 2021, Microsoft ranks second in ICML, NeurIPS, and ICLR.
Research capacity:
There are 53 papers (ranked 2) accepted by ICML 2021, 116 papers (ranked 2) accepted by NeurIPS 2021, and 63 papers (ranked 2) accepted by ICLR 2021.
Apple AI
In today's increasingly open technology environment, Apple is a great company. Unfortunately, like their iOS, we don't know much about Apple's actual level of AI development. But we can see the application of artificial intelligence technology in every product, such as Siri, photo editing, AI camera, AI chip, etc.
Apple is more willing to acquire cutting-edge artificial intelligence startups or technologies and use them in future products. They are committed to building Apple's system, SDK and Apple Silicon into an AI-friendly ecosystem.
Tesla AI
The Tesla AI team led by young artificial intelligence scientist Andrej Karpathy became the focus of last year's (2021) Tesla AI Day. Tesla showed us advanced technologies like Neural Networks, Evaluation Infrastructure, Dojo Chip, Dojo System, and Tesla Bot.
Although Tesla AI is more focused on self-driving, its vision is not limited to making self-driving cars, but also humanoid robots. Tesla's engineering prowess is excellent, using its vast amount of driving data to train the model and deploy it on Tesla vehicles with the latest version of FSD.
HydraNet - a powerful artificial intelligence neural network model for Tesla's autonomous driving
Tesla is expanding its AI team in various ways, including the Tesla AI Day.
Amazon AI
Although there are not so many star products and scientists, Amazon also has its place in the field of artificial intelligence. It provides a powerful infrastructure for machine learning, deep learning, and other AI tasks (AWS, Amazon Web Services). According to Statista, AWS has a 33% (No. 1) cloud infrastructure market share as of Q4 2021, while the next two competitors, Microsoft Azure and Google Cloud, have 21% and 10%, respectively.
IBM AI
Many people may forget that in 1997, about 19 years after AlphaGO defeated the human Go world champion, Deep Blue, an IBM-built supercomputer, defeated the world chess champion Garry Kasparov 4-2 (three wins and one tie). However, in areas such as autonomous driving, deep learning, machine translation, and computer vision that represent cutting-edge AI research, we hardly see IBM. Of course, IBM still has many respected scientists who publish many papers every year.
DB2 and WebSphere, compared to their competitors MySQL and Tomcat, are bloated, unwieldy and difficult to use. But relying on IBM's powerful system, the sales and market share of these software are good. Under the overall IBM structure, many technologies will become bulky and inefficient. The same goes for artificial intelligence. Both IBM and Google are two big companies, hoping to avoid some bad habits of big companies.
NVIDIA AI
Nvidia is another AI infrastructure provider, and many machine learning and deep learning models are trained using Nvidia's GPUs.
In addition to hardware, NVIDIA also developed CUDA, a parallel computing platform, and APIs that allow software to use certain types of GPUs.
It is also at the forefront of papers published by companies each year. 8 papers were accepted by ICML 2021, and 20 papers were accepted by NeurIPS 2021.
In addition to the above companies, there are many other excellent companies that have participated in this artificial intelligence revolution, such as Tencent, Baidu Huawei, HKUST Xunfei and so on. Although the development of China's AI industry is relatively late and the relative technology is not perfect, I believe that China's AI technology will flourish one day sooner or later. come on!
Although there is no evidence that AI has awakened, the AI products released by these technology companies are enough to amaze us, such as Tesla's FSD, Google Translate, Siri, GitHub Copilot, etc. These products have directly or indirectly changed our way of life, study and work. We have long been used to artificial intelligence products, so one day artificial intelligence will have self-awareness, we don't have to be afraid. I believe that humans will solve the problem of machine awakening.
So in the future, when artificial intelligence is truly awakened, what should human beings do? ?
VX search applet: AI artificial intelligence tools, experience different AI tools
