Understanding ChatGPT and OpenAI in one article
Titanium media note:This article is from WeChat WeChat official account Universe All Metaverse (ID: All Metaverse 001), written by Yuan Allen, and published by Titanium Media with authorization.
Are you afraid that your job will be replaced by AI? ChatGPT, a chat bot, has exploded social networks in recent days, and global technology giant Microsoft also plans to add 10 billion dollars to it. To what extent is his chat ability, can he really subvert the current artificial intelligence ecology? The following 49 ChatGPT usages may help you to open your cognition.
Yes, creativity, research and code work that we find difficult to replace in traditional cognition may be challenged by this upstart AI with 1.3 billion high-quality parameters. After the measured experience, the author thinks that ChatGPT has the ability to easily pass the Turing test (that is, people mistakenly think it is a conversation with a real person), and its answer is more detailed, convincing and continuous logical reasoning ability than siri, Xiao Bing and other mainstream conversation robots.
At the end of November 2022, the American head artificial intelligence research institute OpenAI released the latest generation of ChatGPT, which broke through one million users in just one week. According to UBS Group AG, ChatGPT has accumulated 100 million users in just two months, which is the fastest consumer Internet product in human history, breaking Tiktok9’s record of breaking 100 million users in nine months.
How does ChatGPT break the stereotype of "artificial mental retardation" of chat robots and create Wow moment for over 100 million users? What is the background and origin of OpenAI, and why it can overtake a number of information technology manufacturers and take the lead? This article will take you to draw a clear line and read ChatGPT and OpenAI in one article.
· ChatGPT：
It is an AI chat robot program developed by OpenAI, which is also often called artificial intelligence language model. This program is mainly based on the large-scale language model of GPT-3.5 architecture and reinforcement learning from human feedback (RLHF).
· GPT-3：
The full name is Generative Pre-trained Transformer-3, which was released in May 2020. It is the third generation natural regression language model developed by OpenAI and adopted by Microsoft in September 2020. GPT— 3 includes 175 billion parameters, which is the largest artificial intelligence model in history.
· GPT-3.5：
The natural regression language model completed in early 2022 includes the supervised learning version of the model text— davinci— 002 and reinforcement learning version text— davinci— 003。 ChatGPT is formed by fine-tuning a model in GPT-3.5 series.
· AIGC：
That is, artificial intelligence generates content, and artificial intelligence independently creates new content forms such as text, image, music, video and code based on training data and generation algorithm model. Acumen research predicts that the AIGC market will reach $110 billion in 2030.
Generative AI:
That is, Generative AI, a general term for AIGC overseas, allows artificial intelligence models to learn components or elements of artifacts from data through various machine learning, and then generates original and reassembled real artifacts (content, products, tasks). These artifacts refer to training data and are similar, but not copied.
· NLP：
Natural language processing technology is one of the core topics of artificial intelligence, which forms feedback by matching combined answers in the answer base through semantic document analysis.
Human feedback reinforcement learning:
Compensation learning with human feedback (rlhf) is used to optimize and adjust the output results of the model by manually scoring the results generated by the model.
Transformer model:
Proposed in 2017, it is a neural network model based on self-attention mechanism, which was originally used to complete translation tasks between different languages, including encoder and decoder, which are responsible for encoding the source language text and converting the encoded information into the target language text information respectively.
Large language model:
Large Language model(LLM) is an AI model that integrates various requirements and adapts to differentiated business scenarios. Through pre-training, I can gain the ability of generation, world knowledge and context learning. After fine-tuning the instructions, I have the ability of emergence. In some texts, I even have a level beyond human beings, which requires great computational support.
Small language model:
The dominant research direction before 2015 is generally used to deal with single or similar tasks, which requires much less data and training than the AI big model, but the overall expression ability is also far from that of human beings.
In essence, ChatGPT is an advanced AI model, which has a very rich range of application answers. This product belongs to the AIGC track, and the answer is generated in the form of text-text. It can also be used as a prompt word to be extended to generate more content, such as text-image, that is, AI painting.
Development history of OpenAI company
OpenAI is an American artificial intelligence research laboratory, which consists of a profit-making organization OpenAI LP and a parent company’s non-profit organization OpenAI Inc Its purpose is to deal with the potential risks brought by the development of artificial intelligence, open its patents and research results, promote and develop friendly artificial intelligence, and benefit mankind as a whole. OpenAI is the unicorn behind the AI model ChatGPT, GPT-3, DALL-E 2 and the AI dialogue product ChatGPT, with a recent valuation of 29 billion US dollars.
December 2015:
Elon Musk and Greg Brockman (former Stripe CTO, now Chairman & President), Sam Altman (former President of YC, now CEO) and Ilya Sutskever (Geoffery Hinton, father of neural network, now chief scientist) and others founded OpenAI in San Francisco. The start-up capital is as high as $1 billion, including the fundraising they have completed from friends and relatives, including Reid Hoffman, co-founder of Linkdin, Peter Thiel, co-founder of Paypal, Jessica Livingston (co-founder of YC), and even AWS, YC and Infosys.
In 2016:
OpenAI developed the first intelligent system, which defeated the professional players in the Dota2 game.
In 2017:
OpenAI released GPT-1, which is a language processing model with 117 million parameters based on the encoder Transformer.
February 2018:
Elon Musk resigned from OpenAI.
In 2018:
OpenAI has received a new batch of donations from individual investors, including game developer Gabe Newell, Skype founder Jaan Tallinn, and former American and Canadian Olympians Ashton Eaton and Brianne Theisen-Eaton.
In 2018:
OpenAI cooperated with Microsoft to launch AzureAIPlatform, which is a new project to make it easier for developers to build and deploy AI applications.
March 1, 2019:
OpenAI LP subsidiary was established for profit. In the same period, OpenAI officials announced the pre-seed round investment from ReidHoffman Charity Fund and Khosla Ventures at that time, and the amount was not disclosed.
July 2019:
Microsoft injected $1 billion into OpenAI. This strategic investment makes Microsoft Azure the "exclusive" cloud computing service provider of OpenAI.
In 2019:
OpenAI released GPT-2, which is a more powerful language processing model of GPT-1, with 1.5 billion parameters after the park. More tasks are added, and its data set and model parameters are expanded.
In 2020:
OpenAI released DALL-E, a creative AI system that can be described as an original image according to this book.
June 2020:
OpenAI released GPT-3, which is the strongest language processing model in the past, with parameters as high as 175 billion. Massive training has created a strong performance of GPT-3 in article writing and representative writing.
September 2020:
Microsoft obtained the exclusive license of GPT-3.
In 2021:
OpenAI launched an AI venture capital fund with a total amount of 100 million US dollars. The LP behind it includes Microsoft and other partners. OpenAI hopes to invest in industries where AI technology can have a transformative impact, especially in medical insurance, education, climate change and tools to improve productivity.
January 2022:
OpenAI received $250 million in Series A financing.
November 30, 2022:
OpenAI released ChatGPT, which attracted millions of users in a week. It can answer informative and open questions, and can also interact continuously according to the context, and can create fluent poems, novels and even code content.
In 2022:
OpenAI’s annual revenue is about 36 million US dollars, its expenditure exceeds 544 million US dollars, and its annual loss exceeds 500 million US dollars.
January 2023:
Microsoft announced that it will invest $10 billion in OpenAI with a valuation of $29 billion, and all its products will integrate ChatGPT across the board. In addition to the previously announced search engines Bing and Office, Microsoft will also integrate ChatGPT in the cloud computing platform Azure, and Azure’s OpenAI service will allow developers to access the AI model. Microsoft CEO Nadella said that every product of Microsoft will have the same AI capability and be completely transformed.
At the end of January 2023:
Just two months after its launch, ChatGPT has exceeded 100 million monthly users at the end of January 2023, making it the fastest growing consumer application in history.
At the end of January 2023:
ChatGPT announced an important update: improved "authenticity" and "mathematical ability".
February 2023:
OpenAI launched ChatGPT Plus subscription service, which costs $20 per month, giving paying users priority in using the service during peak hours.
Why OpenAI has become the top three AI laboratories in the world today can be seen from the above development history.
1. Super founder:
OpenAI is an organization established by the global super technology giants, born with a golden spoon. Elon Musk’s aura needs no further explanation. Samuel Altamn, CEO of Elon Musk, is a genius who dropped out of school majoring in computer science at Stanford University. He is also the former president and current chairman of YC, a well-known American venture capital institution, and can be said to be an absolute man of the hour in Silicon Valley.
2. Banknote capacity:
Before Microsoft’s huge investment of 10 billion yuan came out in early 2003, OpenAI had completed several rounds of personal and institutional investment, and its accumulated financing amount exceeded 4 billion US dollars, and it only reached the A round. This is inseparable from the founder’s rich and profound investor relationship and the cutting-edge artificial intelligence model that OpenAI constantly outputs.
3. Top talents:
At present, OpenAI has 375 resident employees in California, most of whom are top talents in the fields of artificial intelligence and machine learning. These technicians earn as much as $200 million a year.
It took more than two months, and ChatGPT has been fully circled around the world. What is the value of such an artificial intelligence chat robot that is responsive but not completely accurate?
1. The efficiency of the writing profession has doubled:
ChatGPT’s core competence lies in text output, including high-quality logic, examples and discussions, which has an impact on all office white-collar worker that need text output. It is foreseeable that the creators of AIGC that can be flexibly used in the future will have stronger productivity and workplace competitiveness, and it is unknown how many text jobs will be replaced by artificial intelligence like ChatGPT. Take Buzzfeed, an American news aggregation website, as an example. Recently, it announced that it would lay off 12% of its staff and introduce ChatGPT to create, and its share price soared by 300%.
2. The efficiency of code development doubled:
At present, ChatGPT has shown strong programming ability, even can output complete app programming, and has made some achievements in debugging. Programmers can use ChatGPT to build infrastructure, and on this basis, ChatGPT can optimize the code or manually optimize the code by constantly prompting new requirements.
3. Customer service, NPC, virtual employees’ soul upgrade:
At present, a large number of online customer service actually call fixed answers from a fixed question-and-answer library, which lacks flexibility and extensibility. chatgpt’s emergent ability can greatly improve the service ability of artificial intelligence customer service and optimize customer experience.
4. Provide cost reduction paradigm for AIGC research and development:
ChatGPT has accumulated massive pre-training model based on OpenAI, and achieved better product output effect with tens of billions of selected data far lower than that of the third generation model, which also highlights the importance of data quality and accurate RLHF, leading machine learning to shift from stacking data to optimizing data quality.
5. Open a corner of the next generation search engine:
Google CEO has sounded a red alert inside Google in the face of the phenomenal performance of ChatGPT. After all, ChatGPT has demonstrated a more efficient ability to answer questions in some fields than traditional search engines, and it is not excluded that ChatGPT will unlock the Internet search ability to supplement its answers in the future. This ability can be said to be a fundamental threat to Google’s foothold.
In 2022, "AIGC" took over the baton of "Metauniverse" and "Web3" and became a hot keyword in science and technology circles and venture capital circles, and ChatGPT can be said to be the first truly "AI killer application".
Venture capital giant Sequoia predicts that in the future, there will be different killer applications in the fields of text, code, images, speeches, videos and even 3D content. By then, the production and lifestyle of human beings will be reshaped:
Therefore, our best countermeasure should be to actively learn and embrace these new AI tools and do a good job of "man-machine collaboration" in order to have a foothold in the next era.