Click the card below to follow us↓↓↓
Recently, ChatGPT has become a phenomenal hot topic. Everyone is very concerned about its technology, impact, and other related content. This article provides professional answers to ten questions about ChatGPT.
1. How Popular Is ChatGPT Right Now?
On November 30, 2022, ChatGPT was launched, sparking heated discussions worldwide. According to Similarweb data, in January this year, there were about 13 million unique daily visitors using ChatGPT, more than double that of December last year, with a cumulative user base exceeding 100 million, setting the record for the fastest application to reach 100 million users on the internet, surpassing TikTok’s 9-month record.
In addition to the myriad of unusual questions from users, which have led to extensive dissemination on social media, ChatGPT’s impressive performance across various fields has further strengthened its visibility. The University of Pennsylvania found that ChatGPT was able to pass the final exam for its MBA program. The journal Nature announced on January 24 that it would not publish papers authored by AI tools. Nature stated that authorship implies responsibility for the work, which AI tools cannot assume. It is rumored that ChatGPT also successfully passed Google’s programming interview, receiving an offer for an L3 engineer position with an annual salary of $183,000.
2. What Are the Core Technologies Behind ChatGPT?
ChatGPT is a form of generative AI, and Gartner (a technology research and analysis company) ranked it as the top strategic technology trend for 2022. Gartner predicts that by 2025, generative AI will account for 10% of all generated data, although this proportion is currently less than 1%.
The support behind ChatGPT is the large AI model, which was detailed in Tencent Research Institute’s report on “The Top 10 Digital Technology Trends of 2022”. Current AI is mostly trained for specific application scenarios, making it difficult for the generated models to transfer to other applications, falling under the category of “small models”. This entire process requires not only a large amount of manual tuning but also feeding the machine vast amounts of labeled data, which lowers the efficiency of AI research and development and incurs high costs.
Large models are typically trained on unlabeled large datasets using self-supervised learning methods. Subsequently, in other application 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 improvements to large models can benefit all downstream small models, significantly enhancing the applicability and research efficiency of AI, thus large models have become a key investment direction in the industry, with OpenAI, Google, Facebook, Microsoft, and domestic companies like Baidu, Alibaba, Tencent, Huawei, and Zhiyuan Research Institute launching super-large models. Notably, OpenAI’s GPT-3 large model has demonstrated impressive performance in translation, Q&A, content generation, etc., giving the industry hope for achieving general artificial intelligence. The current version of ChatGPT is GPT-3.5, which is an enhancement over GPT-3, further improving its capabilities.
One of the core technologies used by ChatGPT is the Transformer, which has been one of the biggest highlights of AI technology in recent years. It is a deep learning model proposed by Google in 2017 that employs an attention mechanism to assign different weights based on the importance of different parts of the input data. The accuracy and performance of the Transformer surpass those of previously popular models like CNN (Convolutional Neural Networks) and RNN (Recurrent Neural Networks), greatly improving the effectiveness of model training and enabling AI to further enhance its capabilities based on larger models, more data, and stronger computational power. Additionally, it possesses strong cross-modal capabilities, demonstrating excellent performance not only in NLP (Natural Language Processing) but also in voice and image processing.
3. Why Has ChatGPT Succeeded?
ChatGPT is a classic case of successful technology research and development, engineering, and igniting public enthusiasm.
In its ongoing research, OpenAI continuously explores new technological routes. The introduction of reinforcement learning has significantly improved the model’s effectiveness. In this research, OpenAI involved human experts in model training. On one hand, experts help ChatGPT generate responses that are more aligned with human habits, and on the other hand, they rank the generated results, implementing a reward mechanism for fine-tuning optimization.
Since its inception, OpenAI has been committed to building the capability for general AI and has consistently invested in this vision, which is a key factor in ChatGPT’s success. This vision has attracted a large number of high-level talents to engage in research and development work without any commercial KPIs, ultimately leading to significant breakthroughs.
Through user usage, OpenAI can obtain feedback on the model’s strengths and weaknesses and user experience, forming a closed loop from model usage to experience feedback, further optimizing the model. In addition, OpenAI organizes special feedback competitions, encouraging users to provide feedback on risks, harms, and novel response suggestions, with participants having the chance to win $500 in API credits and redeem corresponding prizes.
A technology with significant influence is closely related to its project planning. The Q&A and multi-turn dialogue format of ChatGPT has effectively stimulated public enthusiasm and creativity, allowing users to express freely based on topics of interest, whether humorously or seriously, frequently trending on social media and in the news. The previous launch of GPT-3, which could write news articles, compose poetry, translate, and code, also sparked considerable discussion in society. This project design strategy that interacts well with the public is worth learning from. Previous projects like AlphaGo’s Go showdown and DeepMind’s protein folding problem-solving have also set good topic tasks.
4. Is ChatGPT Open Source?
It is not yet open source and currently serves via API (Application Programming Interface) calls, with no plans for open sourcing at this time.
OpenAI’s lack of openness is a point of criticism from many in the industry. Open source has been one of the core drivers of the software and internet industry’s robust development over the years. The open-source approach can mobilize the enthusiasm of global developers, allowing anyone to download, use, optimize, and share the source code within the community. This mechanism of innovating with the power of society greatly accelerates the progress of technological research and industrial applications.
Regarding why ChatGPT is not open source, some experts in the industry also agree, as AI technology is still a black box, and its internal mechanisms are not yet known. If the code were open-sourced, it would be difficult to prevent the technology from being used for purposes detrimental to society and humanity. Furthermore, since OpenAI abandoned its non-profit organization status and accepted investments from companies like Microsoft, it is likely to adopt a closed-source model for the entire model while promoting open application interfaces, and will also open source a small portion of models to enrich the developer ecosystem.
5. Is ChatGPT Free? Can It Make Money in the Future?
Currently, it is free for regular users to use ChatGPT.
According to Fortune, in 2022, OpenAI’s revenue was expected to be less than $30 million, with a net loss of $545 million. However, with the explosive popularity of ChatGPT, its losses may further increase, as each user call incurs more computational resources and bandwidth costs for OpenAI.
Of course, OpenAI has also begun commercial attempts. In February 2023, OpenAI announced the launch of a paid pilot subscription plan, ChatGPT Plus, priced at $20 per month. The paid version features include queue-free access during peak times, faster response, and priority access to new features and improvements.
OpenAI predicts that as ChatGPT becomes an important tool for attracting customers, its revenue will grow rapidly, with projected revenue of $200 million in 2023 and expected revenue exceeding $1 billion in 2024.
Additionally, reports indicate that Microsoft is negotiating to invest about $10 billion in OpenAI, raising its valuation to $29 billion. Previously, in 2019, Microsoft invested $1 billion in OpenAI, providing substantial funding support and Azure cloud computing resources.
6. Will ChatGPT Replace Search Engines?
In the short term, it will not. In the long term, it may form a new type of search engine through integration.
Currently, the answers provided by ChatGPT vary in quality; it can give brief answers to many questions and even write an industry planning report at a junior planner level, but it also frequently produces nonsensical responses, which necessitates further improvement in its reliability. Moreover, ChatGPT’s knowledge base mainly consists of data prior to September 2021, lacking new data, which greatly limits its ability to provide new information. At the same time, OpenAI has also launched WebGPT, which can search for information on the internet and provide sources, significantly supplementing ChatGPT’s real-time information sources, thus further optimizing its answers.
In the logic of search engines, the choice lies with the user, and search engines typically provide numerous relevant results, continuously optimizing search results based on user click feedback. In recent years, search engines have also been attempting to provide unique results, such as offering a single answer directly in the search box, but the currently available unique results are quite limited, and multi-turn conversations cannot be conducted.
The launch of ChatGPT has greatly shaken search engine manufacturers and inspired them to create new search experiences. Google’s two founders have repeatedly held discussions to address the disruptive potential of ChatGPT on search engine business. Google CEO Pichai stated that a large language model based on AI similar to ChatGPT will be launched soon, a conversational AI service supported by LaMDA, allowing users to use it as a “search companion”. Microsoft has already integrated ChatGPT features into its Bing search engine and has conducted small-scale testing. Baidu also plans to launch a conversational robot “Wenxin Yiyan” similar to ChatGPT in March.
We can also look at ChatGPT’s own answers:
7. What Other Models and Applications Are Similar to ChatGPT?
ChatGPT is a typical AIGC large model, and corresponding generative models have emerged in various fields such as text, code, images, video, and 3D both domestically and internationally.
Currently, a strong competitor to ChatGPT is Claude, a generative AI model recently launched by Anthropic. Anthropic was founded in 2021 by several former OpenAI researchers, including former OpenAI research vice president Dario Amodei and lead author of the GPT-3 paper Tom Brown. According to the Financial Times, at the end of 2022, Google invested approximately $300 million in this startup. Compared to ChatGPT, Anthropic emphasizes on its website its work to build “reliable, interpretable, and controllable AI systems”.
Tencent’s Hunyuan large model integrates CV (Computer Vision), NLP (Natural Language Processing), and multi-modal understanding capabilities, having topped the rankings of five major authoritative datasets like MSR-VTT and MSVD, achieving a grand slam in cross-modal fields. In May 2022, Tencent’s “Hunyuan” AI large model topped the overall leaderboard, reading comprehension, and large-scale knowledge graph rankings in CLUE (Chinese Language Understanding Evaluation), breaking three records. In December, Hunyuan launched the first low-cost, practical NLP trillion-parameter large model in China, again topping the natural language understanding task leaderboard CLUE. Hunyuan uses a hundred billion model for hot-start training, completing the training of the trillion-parameter large model HunYuan-NLP 1T in just one day with only 256 cards, with overall training costs only 1/8 of directly cold-starting the trillion model.
Image Source: Sequoia Capital “Generative AI: A Creative New World”
8. Will Large Models Like ChatGPT Help Us Achieve General AI?
Since the launch of GPT-3, which contains 175 billion parameters, in June 2020, discussions in the industry about achieving general AI have intensified. As Microsoft CEO Nadella recently stated in an interview with The Wall Street Journal, the development of GPT is not linear but exponential, so compared to GPT-3, the current GPT-3.5 has already demonstrated stronger capabilities. The industry widely predicts that GPT-4 will be launched this year and will possess even more powerful general capabilities.
However, at the same time, we need to have a clear distinction that AI is not omnipotent; no matter how powerful AI becomes, the problems it solves are just a tiny fraction of all the issues faced by humanity. The problems that artificial intelligence has found solutions for so far are only a small part of computable problems.
9. What Social or Ethical Issues Might ChatGPT Raise?
ChatGPT also faces long-standing ethical issues that artificial intelligence has encountered, such as discrimination, security issues related to misuse for fraud, political misinformation, identity forgery, and intellectual property issues. Recently, OpenAI stated that it is considering adding model watermarks to ChatGPT, which would allow monitoring of special markers recognizable by the model to reduce the risk of misuse.
Additionally, regarding the widespread concern about job loss, ChatGPT shared a post on LinkedIn stating: “AI will not replace you; a person using AI will replace you.”
10. What Changes and New Opportunities Will ChatGPT Bring in the Future?
ChatGPT’s own answers have already demonstrated a typical path for its development, namely as a companion robot for humans.
However, more importantly, ChatGPT will become a new entry point for future human-computer interaction, likely changing the interaction methods of existing apps to a more natural dialogue style, allowing users to use software and invoke skills more intuitively. For example, in future document editing software, people could directly generate text content or create graphics simply by describing their needs, or even directly perform image editing tasks. In programming, the instant programming method of ChatGPT will change traditional work methods and application interaction rules, pushing industries into a new phase of software 3.0.
In addition, ChatGPT can have a diverse range of application prospects in the future:
First, the currently popular digital humans, beneath an attractive exterior, require models like ChatGPT to provide conversational abilities to give them interesting personalities, thereby better accompanying and serving humans. This capability can also be embedded in robotic bodies, making future humanoid robots smarter and more human-like.
Second, many developers can utilize platforms like ChatGPT to fine-tune models based on different industries and scenarios, thus creating a variety of rich applications that meet user needs, forming an ecosystem of conversational AI. For example, the popular Jasper abroad fine-tunes the GPT-3 model to help users write marketing copy, and some have even completed a 120,000-word novel on it, with major companies like IBM, Airbnb, and Autodesk as its enterprise users. Ordinary users can try a version that allows 10,000 words for free, after which they can subscribe for $29 for 20,000 words or $59 for the ability to generate 50,000 words.
Third, in fields like education, healthcare, advertising marketing, e-commerce, market and strategic consulting, enterprise services, and coding, ChatGPT can become a more professional human assistant that not only generates content but also invokes various professional capabilities, potentially replacing some entry-level professional jobs. In enterprise services, on February 8, Microsoft launched an advanced Teams product powered by ChatGPT, which can automatically help participants generate meeting minutes. Even if not attending the meeting, the intelligent review feature can help users generate minutes and key points. If mentioned by colleagues during the meeting, ChatGPT can also provide personalized timeline markers for users to quickly view shared and discussed content. This service will charge $7 per month starting in June, increasing to $10 in July. In coding, former Tesla AI director Andrej Karpathy recently revealed that 80% of his code is generated by AI (GitHub Copilot), with an accuracy rate of 80%.
Fourth, through combinatorial innovations with other modal AI tools, ChatGPT can integrate text-to-image, text-to-video, and even future tools that directly generate 3D models, greatly enriching UGC content and becoming a core engine of content industrialization.
In the future, the integration of ChatGPT with more AI, cloud computing, and other information technologies will create tools that change productivity curves and become new drivers of economic development.
This article is authorized to be reproduced from the WeChat public account “Tencent Research Institute”
The author Wang Qiang is a senior expert at Tencent Research Institute and the director of the Frontier Technology Research Center
Original title:Read This Article to Understand:Ten Questions About ChatGPT
Some people will inevitably be eliminated! But can ChatGPT really disrupt the news industry?
The annual newspaper verification begins! The construction of integrated media is the focus of verification!
The 33rd China News Award self-nomination and nomination for entries has begun!
