After the Lunar New Year, the giants announced ChatGPT-like progress
In February, Microsoft announced the latest versions of its artificial intelligence search engine, Bing and Edge, powered by ChatGPT.
Microsoft's stock jumped 4.2 percent, sending its market value soaring $80 billion.
Almost at the same time, Google will also integrate new artificial intelligence into the search engine.
It was announced that Bard, an artificial intelligence conversational robot, is in closed beta and will be available to the public in the coming weeks.
Baidu will launch an ChatGPT-like AI chatbot in China in March.
Baidu's Hong Kong stock gains expanded rapidly, rising 15.33% by the close.
Tencent, Alibaba, and Huawei have also announced that they are applying for artificial intelligence human-machine dialogue patents.
360 also recently said on the interactive platform that it plans to launch a demo version of ChatGPT-like products as soon as possible.
The emergence of ChatGPT has raised the upper limit of natural language processing capabilities, opened the door to the commercialization of artificial intelligence technology and become a popular model on the consumer side, so it has epoch-making significance.
Differentiation strategy, stock price double heaven
At the end of December 2022, Google felt the pressure from ChatGPT and quickly dispatched employees to conduct research on ChatGPT's competing products.
Google has invested nearly $400 million in AI startup Anthropic, which is testing competing products from ChatGPT, a company founded in 2021 by OpenAI, the developer of ChatGPT.
Google announced that it will launch an AI chatbot called Bard, an experimental AI program based on the Language Model for Conversational Applications (LaMDA), which is currently being tested and made available to the public in a few weeks.
Microsoft responded quickly and accelerated the development of related applications.
Recently, Microsoft announced the integration of ChatGPT into its entire product line and expanded its deep cooperation with OpenAI, the owner of ChatGPT.
Launched the latest versions of artificial intelligence search engines Bing (Bing) and Edge browser, supported by ChatGPT and incorporating GPT-3.5 language technology.
Microsoft launched a new version of the search engine New Bing and browser Edge that integrate ChatGPT technology, which has also become an important weapon for Microsoft to fight Google in the search engine field.
Big companies like Google and Meta will definitely have an advantage because they have the human, technical, and financial resources.
Therefore, these companies have enough space in the market and it will be a new competition, especially in terms of depth and correctness.
In contrast, Microsoft and Amazon have chosen to form an alliance with OpenAI in an attempt to capitalize on it.
Bard, as a key benchmark against ChatGPT, has attracted much attention.
According to Google, Bard is an experimental conversational AI service, powered by LaMDA and built using Google's large language models and network information.
But I never expected that Google, which had a late episode, overturned at its own conference.
After the wrong answer shown in the press conference, Google's stock price fell 7.68%, and its market value shrank by about 720.2 billion yuan overnight, which is the most expensive mistake AI has made so far.
There is the credit of AIGC behind the out-of-circle
The core change of AIGC occurs at the content layer, which sits above the data layer.
The importance of AIGC's breakthroughs in painting and conversation, in the field of artificial intelligence, is equivalent to the value of the 2016 Alpha Dog victory over the human Go champion.
The biggest significance of ChatGPT is to let people see the outline of a more intelligent and efficient era.
ChatGP itself is also the product of AI technology iteration to a certain stage.
Server + network opens the era of everyone on the Internet, and computing power + algorithm will open the era of everyone computing.
AI will create new needs for users based on scenarios, applications, and content.
AIGC technology mainly involves two aspects, natural language processing (NLP) and AIGC generation algorithms, AIGC investment direction mainly includes software and hardware and data sets.
The large-scale language model (LLM) represented by ChatGPT will become the focus of competition for giants in the future.
ChatGPT shows great application potential, not just for creating text or images.
There is still a long way to go before AIGC is truly implemented and applied.
Right now ChatGPT is more of a [toy] than a productivity tool.
Western Securities predicts that AIGC may usher in a big year of development in 2023, and the application of autonomous driving will break through the inflection point two years ago.
The biggest highlight of implementing ChatGPT is also the biggest difficulty
ChatGPT needs to turn large models into small models or sparse models as much as possible, and the training and use costs of large models are high.
In addition to the model, another difficulty is how to realize the second half of ChatGPT manual process into automatic training.
ChatGPT is based on the big model and adds a lot of human feedback. This requires a lot of labor, and this part of the work is currently crucial.
Up to now, OpenAI has received a total of more than $11 billion in financing, Microsoft, Google Ventures, Musk and others have all invested, and it is not currently profitable.
With such a huge investment, I am afraid that today's Internet giants can only be deterred.
But on the other hand, it also gives entrepreneurial teams an opportunity. Targeting a specific niche can cut a piece of the cake by digging deeper into a certain technology.
High computing power with fire chip industry
ChatGPT has more than 100 million registered users, and massive user visits have repeatedly caused ChatGPT's official website to be down due to too many visits.
This means that if OpenAI wants to ensure user experience, it must invest more in computing power.
And this requires the support of a large number of chips, which undoubtedly directly injects a strong medicine into the chip market, which is already about to fall into silence.
The total computing power consumption of ChatGPT is about 3640PF-days, and ChatGPT has introduced at least 10,000 NVIDIA high-end GPUs.
Citigroup estimates that ChatGPT may lead to sales of Nvidia-related products reaching $3 billion to $11 billion in 12 months.
The fire of ChatGPT will greatly increase the demand for computing power, 350-watt inference chips will become the norm, and technically promote large model compression to a new climax, while many enterprise and personal services can be built on similar technologies.
High-performance computing chips are the underlying foundation of all AI, and at present, AI chips mainly have two types: general-purpose GPU computing power chips and ASIC-specific chips.
Perhaps with the help of ChatGPT, the chip industry will usher in a new round of market dividends.
ChatGPT brings a paradigm revolution in AI technology
Compared with other large models, ChatGPT's changes in AI technical routes and training methods have brought unexpected breakthroughs in the industry.
ChatGPT has a very powerful technical base, which is the InstructGPT model.
Previously, GPT and BERT model routes have been competing, and in 2018, the BERT model won first, but the GPT model parameters and data scale are getting larger and larger.
Finally, combined with artificial feedback reinforcement learning (RLHF), a major breakthrough has been made in ChatGPT, and the effect has surpassed the BERT model route.
ChatGPT introduces a reinforcement learning mechanism, introducing human-labeled data on the new version, which is targeted and optimized through human feedback.
The difficulty lies in what kind of reinforcement learning mechanism to establish. In the past, the industry has also tried reinforcement learning mechanisms, but many of them have not been very successful, and ChatGPT has made a breakthrough in this matter.
ChatGPT is very particular about data quality and diversity.
The amount of data in ChatGPT reinforcement learning is not large, but the data diversity and labeling system are cleverly designed, so that the data plays a powerful role.
Costs lead to different domestic and foreign routes
The focus of ChatGPT is to create, through a large number of corpus training and strong error correction ability, the material is digested, sorted out and re-exported, which is a completely different way of thinking from the cloud computing or visual technology that was concerned in China in the past.
The reason for the above difference is also straightforward, that is, cost.
ChatGPT is supported by a massive database and huge computing power, but the only way to monetize is to provide services to enterprises and replace some manual operations.
For productization, even if the cost is reduced by another 10 times, this business model cannot be established, because this cost has no advantage over humans.
The support behind it is the artificial intelligence big model
Most of the current artificial intelligence is trained for specific scenario applications, and the generated models are difficult to migrate to other applications, which belongs to the category of [small models].
The whole process not only requires a large number of manual parameter adjustment, but also needs to feed the machine a large amount of labeled data, which reduces the efficiency of artificial intelligence research and development and has a high cost.
[Large models] are usually trained on large datasets without labels, using self-supervised learning.
In applications in other scenarios, developers only need to fine-tune the model to meet the needs of new application scenarios.
This means that the improvement of large models can benefit all downstream small models, greatly improving the applicable scenarios and R&D efficiency of artificial intelligence.
Therefore, large models have become the key investment direction of the industry, and Open AI, Google, Facebook, Microsoft, domestic Baidu, Alibaba, Tencent, Huawei and KLCII have launched super-large models.
In particular, the good performance of the OpenAI GPT 3 model in the fields of translation, question answering, and content generation has made the industry see the hope of achieving general artificial intelligence.
The current version of ChatGPT is GPT 3.5, which is tuned on top of GPT3 and further enhanced.
The industry generally predicts that GPT4 will be launched this year with more powerful general-purpose capabilities.
ChatGPT's future and investment opportunities
The emergence of ChatGPT model is of great significance for the AIGC application of text/speech modalities, which will have a significant impact on the upstream and downstream of the AI industry.
From the perspective of downstream related beneficiary applications, including but not limited to no-code programming, novel generation, dialogue search engines, voice accompaniment, voice work assistants, dialogue virtual humans, artificial intelligence customer service, machine translation, chip design, etc.
From the perspective of increasing demand in the upstream, including computing power chips, data annotation, natural language processing (NLP), etc.
With the continuous advancement of algorithm technology and computing power technology, ChatGPT will further move towards more advanced and powerful versions, and be applied in more and more fields to generate more and better conversations and content for human beings.
One of the core technologies is Transformer
Transformer technology is one of the biggest highlights of artificial intelligence technology in recent years.
A deep learning model proposed by Google in 2017 that uses an attention mechanism can assign different weights according to the importance of each part of the input data.
The accuracy and performance of Transformer are better than the previous popular CNN (convolutional neural network), RNN (recurrent neural network) and other models.
The effect of model training has been greatly improved, allowing artificial intelligence to further enhance its capabilities on the basis of larger models, more data, and stronger computing power.
At present, artificial intelligence has found a solution to the problem, and it is only a small part of the computable problem.
If traditional machine learning or artificial intelligence is to explore at a depth of one meter under water, then the emergence of deep learning has brought people to a hundred meters deep water;
After Transformer, GPT and other architectures, practitioners can directly reach the Mariana Trench with a depth of more than 10,000 meters.
In the past 5 years, large-scale pre-trained language models have received more and more attention in the industry.
Today, artificial intelligence research around the world has many directions and has produced a large number of results.
But strictly speaking, more than 90 percent of all these most amazing results are based on these big model technologies that Transformer iterations have made.
At present, almost all AI tasks and projects will be related to large language model technology, which has represented the most mainstream scientific research and technology iteration direction in the AI industry.
In the future, the integrated innovation of ChatGPT and more AI, cloud computing and other information technologies will create tools to change the productivity curve and become a new driving force for economic development.
End:
ChatGPT evokes people's associations in the era of strong artificial intelligence, that is, artificial intelligence has its own intentionality to set goals.
Although there are still many debates about strong artificial intelligence itself, analysts generally agree that the development of artificial intelligence and a new stage of the information age are approaching.
Part of the data reference: Beijing News: "Artificial intelligence big change: technology giants ushered in ChatGPT "shockwave"", Lieyun selection: "Giants chaotic ChatGPT, will the new era be enlightened or over-deified?" "ChatGPT "Crazy", Interface News: "The Faces of Technology Companies Under the Crazy Hype", Daily Economic News: "ChatGPT Makes Everyone Crazy, Depth, Accuracy or Counterattack "Killing Skills", China Newsweek: "Why is the ChatGPT that continues to evolve so powerful?" Tencent Research Institute: "Ten Questions About ChatGPT"
The manuscripts and pictures published in this public account come from the Internet, only for communication use, if there is infringement, please contact to reply, we will deal with it within 24 hours after receiving the information.