73 years ago, the concept of "machine thinking" was first proposed by Alan Turing, the father of computer science. Since then, passing the Turing test has become a milestone goal for human beings in the field of AI.
Today, 73 years later, after decades of continuous evolution and iteration of AI, there is still no AI that has passed the Turing test in a true sense.
The birth of ChatGPT has become the closest human beings have come to this milestone.
01
Sort out the past and present of ChatGPT
The era of OpenAI has finally arrived
When it comes to the birth of ChatGPT, one has to mention OpenAI, the development team behind it.
OpenAI was founded as a non-profit research organization dedicated to advancing the development of artificial intelligence to ensure that the technology can safely benefit all human beings. But over time, organizations gradually realized the potential impact of artificial intelligence on society and business, and decided to start expanding into the commercial field.
In 2019, OpenAI received investment from Microsoft, restructured into a for-profit company, and opened it to private investors to provide greater financial support for its long-term development. After the reorganization, the company focused on the development and marketing of commercial artificial intelligence technology to achieve faster growth and more business opportunities, but still retained a part of the non-profit nature.
Since then, OpenAI, which has a stable source of funds, has started the full development of the GPT model. After several years of technical precipitation, it finally ushered in the era of OpenAI.
★
The past and present of ChatGPT
June 2018
OpenAI released the GPT-1 model with 110 million parameters.
November 2018
OpenAI released the GPT-2 model with 1.5 billion parameters, but due to concerns about abuse, it does not open all the code and data of the model to the public.
February 2019
OpenAI opened up some of the code and data for the GPT-2 model, but still restricted access.
June 10, 2019
OpenAI released the GPT-3 model, with 175 billion parameters, and provided access to some partners.
September 2019
OpenAI has released all the code and data of GPT-2 and released a larger version.
May 2020
OpenAI announced the beta release of the GPT-3 model, which at 175 billion parameters is the largest natural language processing model to date.
March 2022
OpenAI released InstructGPT, using Instruction Tuning
November 30, 2022
OpenAI has fine-tuned the GPT-3.5 series of large-scale speech models, and the new conversational AI model ChatGPT is officially released.
December 15, 2022
The first update of ChatGPT improves the overall performance and adds new features to save and view historical conversation records.
January 9, 2023
The second update of ChatGPT improves the authenticity of answers and adds a new function of "stop generation".
January 21, 2023
OpenAI released a paid version of ChatGPT Professional limited to some users.
January 30, 2023
The third update of ChatGPT not only improves the authenticity of the answers, but also improves the mathematical ability.
February 2, 2023
OpenAI officially launched the subscription service of the paid version of ChatGPT. Compared with the free version, the new version responds faster and runs more stably.
March 15, 2023
OpenAI launched a large-scale multi-modal model GPT-4, which can not only read text, but also recognize images and generate text results. It is now connected to ChatGPT and is open to Plus users.
ChatGPT, which is based on GPT-3.5 technology, has demonstrated excellent language generation capabilities. After a new round of training iterations, GPT-4 has surpassed the limitations of single-domain and multi-model, and can achieve multi-modal tasks. Execution has made the generation efficiency of AI leapfrog again.
And for AI, what will be the next step of "multimodality"?
02
Large-scale outbreak of generative AI in foreign countries
The AI war presents a situation of "one super and many strong"
The AI war in Silicon Valley presents a situation of "one superpower and many powerhouses", and the competition vision mainly focuses on OpenAI and Google.
In fact, contrary to the current situation where GPT seems to have won a big victory, although OpenAI has been launching new technical products since its establishment a few years ago, the results presented are far from those in the AI field at that time. Strong compared to Google.
At that time, Google acquired DeepMind and was preparing to develop AGI. If DeepMind was successfully developed, Google would be in a monopoly position. One of the reasons why Sam Altman founded OpenAI was to avoid Google's monopoly in the AI field.
In 2016, after the establishment of OpenAI, it continued to launch technical products such as Gym and Universe, but all these were overshadowed by AlphaGo, which has defeated the international Go champion many times. At this stage, Google acquired more than a dozen AI companies while the iron was hot, and invested huge resources in the field of AI.
In the stage where Google was leading the way, OpenAI launched the GPT series of models, opened up business roads in rapid iterations, and successfully checked and balanced Google.
★
Large-scale outbreak of generative AI in foreign countries
November 30, 2022
OpenAI has fine-tuned the GPT-3.5 series of large-scale speech models, and the new conversational AI model ChatGPT is officially released.
December 6, 2022
DeepMind, a UK-based cutting-edge AI research institution under Google's Alphabet, released Dramatron, an AI script collaboration tool.
December 24, 2022
The emerging search engine http://You.com in the United States announced the official launch of the beta version of the conversational AI model youChat. You can try the new chat function on the homepage of http://You.com .
January 7, 2023
Foreign media broke the news that Microsoft conceived of introducing OpenAI's AI model into Word, PowerPoint, Excel, Outlook and other software.
January 17, 2023
Anthropic, a programmer from the GPT-3 R&D team, founded the AI company Anthropic, and its chat robot Claude has entered the testing phase.
February 1, 2023
Microsoft officially announced Microsoft Teams Premium, an advanced paid version of Teams, powered by ChatGPT, which can automatically generate meeting minutes, generate meeting templates and other functions.
February 1, 2023
OpenAI launched an AI text classifier to test whether the text is generated by AI, but the accuracy rate is currently low.
February 3, 2023
Microsoft briefly launched a new version of Bing powered by ChatGPT
February 8, 2023
Microsoft officially launched a new version of the Bing search engine and Edge browser powered by ChatGPT to improve user search efficiency.
February 9, 2023
At the Google Bard press conference, there was a rollover trial, and the answer was wrong, and the market value plummeted by 100 billion U.S. dollars that day.
February 24, 2023
Meta officially announced the SOTA large language model LLaMA, which is open source for non-commercial research use cases.
March 2, 2023
ChatGPT officially announced the open API, through which third parties can integrate the chat model ChatGPT and speech-to-text model Whisper into their own applications and services.
March 6, 2023
Google launched the Whisper-standard speech model USM, which supports more than 100 languages and 2 billion parameters.
March 8, 2023
Microsoft has announced the launch of AI computer vision model Florence in public preview.
March 8, 2023
Google released the largest multimodal embodied visual language model in history - PaLM-E, which can understand images and generate language.
March 14, 2023
Stanford released a model Alpaca fine-tuned by LLaMA 7B, trained for 3 hours, and its performance is comparable to GPT-3.5.
March 16, 2023
Google announces integration of AI tools into Workspace
March 17, 2023
Microsoft announced Microsoff 365 Copilot, fully integrating GPT-4 into the Microsoft Office family bucket.
Judging from the results of the competition at this stage, OpenAI and Google have different advantages in terms of language model technology. Google's 1.6 trillion Transfotmer model won the competition in terms of scale, but from the perspective of commercialization, efficiency determines the iteration speed and cost, and the efficiency of OpenAI's GPT iteration makes it a bigger victory in the commercial field.
Model scale: Google > OpenAI
Model efficiency: OpenAI > Google
Model application field: OpenAI>Google
03
Major domestic AI manufacturers have layouts
AI Big Boss Entrepreneurship Boom Comes
According to iResearch, the scale of my country's AI industry will reach 195.8 billion yuan in 2022, and with the outbreak of generative AI in 2023, AI product forms and application boundaries will continue to expand.
AI basic technology layout of domestic Internet giants
Comparison of AIGC technology research and development of domestic Internet giants
AI commercialization progress of domestic Internet giants
And the Q1 quarter of 2023 is not yet over, the AI large-scale model track has already been crowded with "players" from all walks of life at home and abroad. Zhou Bowen, the former head of JD.com AI, Li Yan, the former core figure of Kuaishou AI...corporate tycoons, bosses, overseas returnees, executives of big factories, and professors, can be described as the Eight Immortals crossing the sea and showing their magical powers.
On the whole, most of the high-profile announcements of rushing into the track are professionals with high reputation and professional ability in the field of AI.
Looking at the commercial development of AI technology in the past five years, with the support of big bosses, my country's AI product and service providers will continue to work hard around technology development, scene innovation, business value creation, and refined services; demand-side companies are also starting from Single-point experiments, data accumulation, and the development route of strategic reform are gradually deeply bound with AI technology.
04
Academia is wary
The era of deep integration of education and AI is coming
Since the birth of ChatGPT, the academic community has always maintained a cautious attitude towards it.
At present, many "conservative" schools have banned the use of ChatGPT, such as Brown University in the United States, the University of Wisconsin, the University of Madison, and the University of Toronto in Canada.
However, there are still some schools that choose to adapt to the changes of the times and have publicly supported the use of ChatGPT. For example, the Hong Kong University of Science and Technology in China, the University of California, San Diego, and the University of Los Angeles in the United States, and Oxford University and Cambridge University in the United Kingdom also expressed that they are considering using ChatGPT for teaching.
★
Response to ChatGPT in the scientific field
December 6, 2022
On December 6, 2022, Stack Overflow, a well-known developer question-and-answer website, issued a temporary policy: disable ChatGPT to generate content to answer questions on Stack Overflow.
December 9, 2022
DeepMind's AI programming model AlphaCode is on the cover of Science.
December 20, 2022
A paper on the medical preprint platform medRxiv listed ChatGPT as an author.
December 21, 2022
ChatGPT co-authored a paper discussing the anti-aging application of rapamycin with Dr. Alex Zhavoronkov, the founder and CEO of Insilicon Intelligence, and published it in Oncoscience Oncology Science Journal.
January 3, 2023
The International Conference on Machine Learning (ICML) has announced a ban on papers containing text generated from large language models such as ChatGPT, unless these generated texts are presented as part of the paper's experimental analysis.
January 17, 2023
An article titled "Will ChatGPT Get a Wharton MBA?" ” paper, University of Pennsylvania professor Christian Terwiesch gave ChatGPT a final exam for Wharton’s MBA program and found that it scored between B- and B on the test.
January 24, 2023
Nature announced two principles: (1) No large-scale language model tools are accepted as authors of papers; (2) If such tools are used in the creation of papers, researchers should clearly state them in the methods or acknowledgments section.
January 28, 2023
According to a survey conducted by the online learning platform http://Study.com among 1,000 students over the age of 18, more than 89% of students admit to using ChatGPT to complete homework.
January 31, 2023
The world's largest preprint publishing platform arXiv announced a new policy: it does not allow generative AI tools such as ChatGPT to be listed as authors.
February 4, 2023
A paper published by Stanford University scholars proves that the theory of mind (ToM, Theory of Mind), which was originally considered to be unique to humans, has been demonstrated on the AI model behind ChatGPT. According to the research results, the empathy ability of the AI model after GPT-3 Probably about the size of a 9 year old.
February 9, 2023
A research paper published in PLOS Digital Health showed that ChatGPT achieved passing or near-qualifying scores on the USMLE without human assistance. The study authors see this as a notable milestone in the maturation of AI in the clinical arena.
February 10, 2023
"Jinan Journal (Philosophy and Social Sciences Edition)" issued a note on the use of AI writing tools, mentioning: temporarily do not accept any large-scale language model tools (such as ChatGPT) individually or jointly signed articles; if relevant tools have been used in the paper creation , it is necessary to explain in detail how to use it and demonstrate the author's own creativity; for example, if an article citing AI writing tools is used as a reference, a detailed citation demonstration is required.
February 11, 2023
"Journal of Tianjin Normal University (Basic Education Edition)" issued relevant instructions, suggesting that authors explain the use of AI writing tools in references, acknowledgments, and other texts.
February 11, 2023
Shen Yang, a professor of journalism and communication at Tsinghua University and a doctoral supervisor, posted a Weibo request to form a team to resist the pressure brought by ChatGPT to China.
February 19, 2023
The University of Hong Kong expressly prohibits teachers and students from using ChatGPT for teaching purposes, and illegal use will be punished as plagiarism.
March 13, 2023
Some courses at the Hong Kong University of Science and Technology have taken the lead in encouraging students to use ChatGPT, stating that in the mid-term report to be submitted next week, students will get extra points if they use ChatGPT.
The popularity of generative AI has conflicted with the existing educational models and concepts. The macro-education system and specific college teachers and students are not yet ready for new changes. It takes time to discuss and evaluate the limits of the use of this tool. , and how to use it correctly can improve the teaching efficiency and quality.
How to take advantage of the new opportunities AI brings to education, make ChatGPT a teaching aid, relieve teachers’ pressure, enhance students’ creativity, and start to get used to the teaching concept of coexistence with AI, which is the most concerned and ongoing discussion in the entire education community. question.
The birth of any new technology will go through a period of adaptation from resistance to acceptance by people, and finally it will be popularized. Blind avoidance cannot stop the wave of technology from erupting. Perhaps the era of deep integration of education and AI has arrived.
05
AI large model track carnival,
The undercurrent of capital competition is surging
★
The Carnival of AI Capital Circuit
December 16, 2022
According to foreign media citing sources familiar with the matter, OpenAI expects revenue to reach $200 million in 2023 and $1 billion in 2024.
January 6, 2023
Foreign media quoted people familiar with the matter as saying that OpenAI is negotiating to sell existing shares through a tender offer. Its valuation is about 29 billion U.S. dollars, making it one of the most valuable American start-up companies.
January 9, 2023
The first stage: Foreign media broke the news that Microsoft is negotiating to invest 10 billion US dollars in OpenAI, and the transaction target will be completed in 2022. Microsoft will receive 75 percent of OpenAI's profits until it recoups its initial investment.
January 2023
The second stage: When OpenAI's profit reached 92 billion US dollars, Microsoft's share of OpenAI's profit dropped to 49%. Meanwhile, other venture investors and OpenAI employees will be entitled to 49 percent of the company's profits.
January 23, 2023
The third stage: Microsoft announced that it will invest $1 billion in OpenAI. After OpenAI's profit reaches $150 billion, the shares of Microsoft and other venture investors in OpenAI will be returned to OpenAI's non-profit foundation.
January 30, 2023
"ChatGPT Concept" was added to the Flush concept section. ChatGPT immediately became a hot word in the secondary market, and the popularity of related concept stocks soared, setting off a wave of daily limit.
February 8, 2023
The Shanghai Stock Exchange and the Shenzhen Stock Exchange respectively issued letters in response to abnormal fluctuations in stock trading. Several ChatGPTs issued announcements saying that the company did not cooperate with OpenAI, and that ChatGPT products and services did not bring business income to the company. ChatGPT concept stocks, which have been rising for several days, began to fall back.
At the beginning of 2023, the primary market is already slowly heating up. The explosion of ChatGPT not only ignited the AIGC track, but also made the AI track come alive again. Investment institutions are optimistic about the business prospects of ChatGPT and AIGC tracks, which are expected to generate trillions of dollars in economic value in the future.
The secondary market is also optimistic about the development prospects of AI. The excellent performance of ChatGPT has not only attracted investors to actively participate, but also brought new opportunities to the entire technology field.