With the in-depth development of artificial intelligence models, quantum computing, brain-inspired intelligence, cloud native, digital engines, audio and video and other technologies, new technologies, new models and new formats continue to emerge. Aiming at the above fields, we have planned a series of popular science columns T-chat Frontier Hot Questions and Answers, in the form of ten questions and ten answers, in the form of popular language, to communicate with you the latest and hottest cutting-edge technology topics. You are also welcome to guide and ask questions of interest to study together, continue to iterate cognition, and make progress together.
How popular is ChatGPT now?
On November 30, 2022, ChatGPT was born out of nowhere, forming a heated discussion around the world. According to data from Similarweb, in January this year, an average of about 13 million unique visitors per day used ChatGPT, more than double that of December last year. The cumulative number of users exceeded 100 million, setting a record for the fastest application on the Internet to exceed 100 million, surpassing the previous TikTok The speed of breaking 100 million in 9 months.
In addition to the strange problems of the vast number of users, it has triggered a large amount of social media dissemination, and ChatGPT's muscles in various fields have further strengthened its popularity. The University of Pennsylvania found that ChatGPT was able to pass the final exam of the school's MBA program. Nature magazine announced on January 24 that papers that list artificial intelligence tools as authors cannot be published in the journal. According to Nature, authorship implies responsibility for the work, which cannot be assumed by artificial intelligence tools. It is rumored that ChatGPT also successfully passed Google's programming interview and got an offer from an L3 engineer with an annual salary of $183,000.
The core technology behind ChatGPT?
ChatGPT is a form of generative AI that Gartner named as the number one strategic technology trend for 2022. Gartner predicts that by 2025, generative AI will account for 10% of all generated data, but currently it accounts for less than 1%.
The support behind ChatGPT is the large model of artificial intelligence. This technical point was elaborated in last year's Tencent Research Institute's 2022 Top Ten Digital Technology Frontier Application Trends report: most of the current artificial intelligence is for specific scenario applications After training, the generated model is difficult to migrate to other applications and belongs to the category of small model. The whole process not only requires a lot of manual parameter adjustment, but also needs to feed a large amount of labeled data to the machine, which reduces the efficiency of artificial intelligence research and development, and the cost is high. Large models are usually trained using self-supervised learning methods on large unlabeled datasets. Later, in the application of other scenarios, developers only need to fine-tune the model or use a small amount of data for secondary training to meet the needs of new application scenarios. This means that the improvement of the large model can benefit all downstream small models, greatly improving the application scenarios and R&D efficiency of artificial intelligence, so the large model has become the direction of the industry's key investment, Open AI, Google, Facebook, Microsoft, domestic Baidu, Ali, Tencent, Huawei and Zhiyuan Research Institute have launched super-large models. In particular, the outstanding performance of the OpenAI GPT 3 large model in the fields of translation, question answering, and content generation has given the industry the hope of achieving general artificial intelligence. The current version of ChatGPT is GPT 3.5, which is an optimization based on GPT3, and its capabilities are further enhanced.
One of the core technologies used by ChatGPT is Transformer. This can also be seen from its full name, Chat Generative Pre-trained Transformer. Transformer technology is one of the biggest highlights of artificial intelligence technology in recent years. It is a deep learning model using attention mechanism proposed by Google in 2017. It can assign different weights according to the importance of each part of the input data. . The accuracy and performance of Transformer are due to the previously popular models such as CNN (Convolutional Neural Network) and RNN (Recurrent Neural Network), which have greatly improved the effect of model training, allowing artificial intelligence to work on larger models, more data, Further enhance capabilities based on stronger computing power. In addition, he also has a strong cross-modal ability, not only in the field of NLP (Natural Language Understanding), but also in speech and images.
Why ChatGPT can succeed?
ChatGPT is a very successful classic case from technology research and development to engineering, and then to igniting the enthusiasm of the public.
technological innovation
In the continuous research of Open AI, we continue to explore and introduce new technical routes. The reinforcement learning method introduced this time has greatly improved the effect of the model. In the research, Open AI introduced human experts in model training. On the one hand, experts help ChatGPT to write answers that are more in line with human habits. On the other hand, they also rank the generated results. Through such a reward mechanism, the fine-tuning and optimization of the model is realized.
group Culture
Since its inception, Open AI has been committed to building the capabilities of general AI, and has continued to invest firmly, which is a key factor for the success of ChatGPT. Under such a vision, a large number of high-level talents were attracted to carry out research and development work without any commercial KPIs, and finally achieved a major breakthrough.
feedback mechanism
Through the use of users, Open AI can obtain user feedback on the pros and cons of the model and the user experience, thus forming a closed loop from model use to experience feedback, thereby further optimizing the model. In addition, Open AI also organizes a special feedback contest, encouraging users to give feedback on the risks and hazards in the answers, novel answer suggestions, etc. Participants have the opportunity to win $500 API points, which can be exchanged for corresponding prizes.
project planning
A technology with influence out of the circle is inseparable from its project planning. ChatGPT's question-and-answer and multi-round dialogue form have well stimulated the enthusiasm and creativity of the public. Everyone is free to play based on the topics they are interested in, whether it is humorous or serious, and frequently swipes the screen in the circle of friends and the media. Including the launch of GPT3 before, his ability to write news, compose poems, translate, code, etc. has also generated heated discussions and discussions in the society. This kind of project design strategy that interacts well with the public is also worth learning. Including the previous AlphaGo Go game and Deepmind's cracking of the protein folding structure problem, they are all good topic and task settings.
Is ChatGPT open source?
It has not yet been open sourced, and currently serves in the form of API (application programming interface) calls, and there is currently no open source plan.
Open AI is not Open, which is criticized by many people in the industry. Open source is one of the core driving forces behind the vigorous development of the software and Internet industry over the years. The open source method can mobilize the enthusiasm of developers around the world. Everyone can download the source code for use, optimize it and share it in the community. This mechanism of using the power of the whole society to innovate has greatly accelerated the process of technological research and industrial application.
Regarding why ChatGPT is not open source, some experts in the industry also agree, because artificial intelligence technology is still a black box so far, and its internal mechanism is unknown. If the code is open source, it is difficult to avoid the use of this technology for some things that are not conducive to society and human beings. aspect. In addition, since Open AI gave up its position as a non-profit organization and accepted investment from Microsoft and others, from the perspective of commercialization, it will also adopt the closed source of the overall model and open application interface to promote it. At the same time, it will also open source a small number of models. Enrich the developer ecosystem.
Is ChatGPT free? Can you make money in the future?
Currently, ChatGPT is free for ordinary users.
According to Fortune, in 2022, OpenAI's revenue is expected to be less than $30 million, with a net loss of $545 million. With the popularity of ChatGPT, its losses may be further increased, because every time a user calls, Open AI will pay more computing resources and bandwidth costs.
Of course, Open AI has also begun commercialization attempts. In February 2023, OpenAI announced the launch of a paid pilot subscription plan ChatGPT Plus, priced at $20 per month. Premium features include skip-the-line during peak hours, fast response times, priority access to new features and improvements, and more.
Open AI predicts that as ChatGPT becomes an important tool to attract customers, its revenue will grow rapidly, with an estimated revenue of US$200 million in 2023 and more than US$1 billion in 2024.
Additionally, Microsoft is reportedly in talks to invest about $10 billion in OpenAI, bringing its valuation to $29 billion. Previously, in 2019, Microsoft invested $1 billion in OpenAI, providing it with a large amount of financial support and Azure cloud computing power support.
Will ChatGPT replace search engines?
In the short term, no. In the long run, it is possible to form a new type of search engine through integration.
The answers given by the current ChatGPT are uneven. It can give brief answers to many questions and even write an industrial planning report that meets the level of a junior planner, but ChatGPT also inevitably often appears serious nonsense The situation, which makes its confidence need to be further improved. Moreover, ChatGPT's current knowledge base is mainly data before September 2021, and the lack of new data greatly limits its ability to provide new information. At the same time, Open AI also launched WebGPT, which can find information on the Internet and provide information sources, which will largely complement ChatGPT's real-time information sources, thereby further optimizing answers.
Under the logic of search engines, the right to choose lies with the user, and the search engine will generally give many relevant results, and continuously optimize the search results according to the user's click feedback. In recent years, search engines have also been making attempts to uniqueize results, such as providing unique answer choices directly in the search box, but currently only limited results can be provided, and multiple rounds of dialogue cannot be carried out.
The launch of ChatGPT has brought great shock to search engine manufacturers, and also inspired them to create a new search experience direction. Google's two founders held several discussions to deal with ChatGPT's potential to disrupt the search engine business. Google CEO Pichai said that a large-scale artificial intelligence-based language model similar to ChatGPT will be launched in the near future, and Bard, a conversational AI service powered by LaMDA, will allow users to use it as a search companion. Microsoft has added the ChatGPT function to the Bing search engine and carried out a small traffic test. Baidu also plans to launch a new feature in March, adding a ChatGPT-like conversational robot Wen Xin Yi Yan to the search engine.
We can also look at ChatGPT's own answer:
What other models and applications are similar to ChatGPT?
ChatGPT is a typical AIGC large model. At present, corresponding generative models have emerged in text, code, image, video, 3D and other fields at home and abroad.
Currently, a strong competitor of ChatGPT is Claude, who is a generative Al model recently launched by Anthropic. Anthropic was founded in 2021 by several former OpenAl researchers, including Dario Amodei, former vice president of research at OpenAl, Tom Brown, first author of the GPT-3 paper, and others. By the end of 2022, Google invested about $300 million in the startup, according to the Financial Times. On its website, Anthropic places more emphasis on its work on building reliable, explainable and manipulable artificial intelligence systems than ChatGPT.
Tencent's mixed-element large model integrates CV (computer vision), NLP (natural language understanding), and multi-modal understanding capabilities. It has successively reached the top of the list of five authoritative data sets such as MSR-VTT and MSVD, achieving cross-modal A grand slam in the field of ecology. In May 2022, Tencent's Hunyuan AI large model reached the top of the CLUE (Chinese Language Understanding Evaluation Collection) ranking list, reading comprehension, and large-scale knowledge graph at the same time, breaking three records in one fell swoop. In December, Hunyuan launched China's first low-cost, practical NLP trillion-dollar model, and once again topped the list of natural language understanding tasks CLUE. Hunyuan uses 100 billion models for hot start, and can complete the training of the trillion-parameter large model HunYuan-NLP 1T in one day at the fastest with only 256 cards. The overall training cost is only 1/8 of the direct cold start training trillion model.
Will big models like ChatGPT get us to general artificial intelligence?
Since the launch of GPT3, which contains 175 billion parameters, in June 2020, the industry's discussion on achieving general artificial intelligence has further heated up. As Microsoft CEO Nadella said in an interview with the Wall Street Journal recently, the development of GPT is not linear, but exponential. Therefore, compared with GPT3, the current GPT3.5 has demonstrated stronger capabilities. The industry generally predicts that GPT4 will be launched this year and will have stronger general-purpose capabilities.
But at the same time, we also need to have a clear identification, AI is not a panacea. That is to say, no matter how powerful AI is, the problems it solves are only a small part of all the problems faced by human beings. Because there are a large number of problems in the real world that are not mathematical problems, it is impossible to solve them by calculation. The problems for which artificial intelligence has found solutions are only a small part of the computable problems.
What social or ethical issues might arise from ChatGPT?
ChatGPT also faces problems that artificial intelligence has faced for a long time, such as ethical issues such as discrimination, security issues such as being abused for fraud, fraud, political rumors, forged identities, and intellectual property issues. Recently, OpenAI stated that it is considering adding model watermarks to ChatGPT, that is, special marks that can be recognized by the monitoring model to reduce the problem of model abuse.
In addition, in response to the general public concern about unemployment, ChatGPT shared a post on LinkedIn and wrote: AI will not replace you. A person using AI will replace you.
What changes and new opportunities will the future of ChatGPT bring?
ChatGPT's own answer has shown a typical path of its development, that is, as a human companion robot.
But more importantly, ChatGPT will become one of the future human-computer interaction the new entrance is likely to change the interaction mode of the existing APP, allowing users to use the software and call skills in a more natural dialogue. For example, in document editing software in the future, people can describe the requirements and let the software directly generate text content, or generate a graphic, or even directly perform work such as retouching. In terms of programming, ChatGPT, a real-time programming method, will change traditional working methods and application interaction rules, and push the industry into a new stage of software 3.0.
In addition, ChatGPT can also have diverse application prospects in the future:
One is that the popular digital people, under the good-looking skin, need a model like ChatGPT to provide dialogue capabilities, so that digital people can have an interesting soul and better accompany and serve people. At the same time, this ability can also be embedded in the robot body, making future humanoid robots smarter and more human-like.
The second is that a large number of developers can use the underlying platform such as ChatGPT to tune the model based on the large model according to different industries and scenarios, so as to create various rich applications that meet the needs of users, thereby forming an ecology of conversational AI. For example, Jasper, which is popular abroad, can help users write marketing copy through fine-tuning on the GPT3 model. Some people even completed a 120,000-word novel on it. Big companies such as IBM, Airbnb, and Autodesk are its corporate users. Regular users can try the 10,000-word version for free, after which they can order 20,000 words for $29, or the ability to generate 50,000 words for $59.
The third is to become a more professional human assistant in the fields of professional services such as education, medical care, advertising marketing, e-commerce, market and strategic consulting, enterprise services, and coding. Substitute for some entry-level professional work. In terms of enterprise services, on February 8, Microsoft launched an advanced Teams product powered by ChatGPT. ChatGPT can automatically help participants generate meeting minutes. Even if they do not attend the meeting, the smart review function can help users generate meeting minutes and highlights. . If a colleague mentions it in a meeting, ChatGPT will also provide users with personalized timeline markers to quickly view shared and discussed content. The service will cost $7 per month in June, and then increase to 10 USD. In terms of code writing, Tesla’s former AI director Andrej Karpathy recently revealed that 80% of his code is now generated by AI (GitHub Copilot), and has an accuracy rate of 80%.
The fourth is the combined innovation with other modal AI tools. The integration of ChatGPT with text-generated images, text-generated videos, and even tools that directly generate 3D models in the future can bring great enrichment of UGC content and become the core engine of content industrialization.
In the future, the integrated innovation of ChatGPT and more AI, cloud computing and other information technologies will create tools that change the productivity curve and become a new driving force for economic development.
