2023 AI Industry Highlights: From GPT-4 to the Battle of a Hundred Models
-
Recently, the GPT Store officially launched, and the AI community is buzzing again, with players delving into how to create and monetize GPTs.
Since ChatGPT sparked the large model frenzy, 2023 has seen continuous breakthroughs and applications in generative AI technology globally, constantly reshaping our understanding. From the iterative upgrades of large language models to the rise of multimodal AI and Agents, and the deep integration of AI across various industries, competition in the AI field has intensified.
Domestically, major tech companies have rolled out their AI products, igniting the 'Battle of a Hundred Models.' Numerous AI startups are now 'competing' in AI innovation applications, while policy frameworks are gradually improving to regulate and guide AI technology.
Illustration: NUPD "Top AI Player" has chronologically organized the major events in the AI industry in 2023 and selected the top ten most notable ones for review below. Overseas AI players like OpenAI are leading the frontier of AI technology. Although the AI features of domestic content platforms may not match the application scope and discussion heat of top products, considering their user base and influence in China, these attempts are still noteworthy.
The AI industry is evolving rapidly. As AI technology continues to mature, we will see more AI-native applications and use cases. AI will become more deeply integrated into our daily lives—from work assistants to life companions, from content creation to decision-making support. AI will gradually become a significant force driving social progress and innovation.
The Debut of GPT-4: The Most Powerful Model in History On March 15, 2023, OpenAI released the large multimodal model GPT-4, which can accept both image and text inputs and output text. Although its capabilities in many real-world scenarios are not yet on par with humans, it performs at human levels on various professional and academic benchmarks.
GPT-4 blog address: https://openai.com/research/gpt-4
GPT-4 technical report address: https://cdn.openai.com/papers/gpt-4.pdf Compared to the previous generation GPT-3.5, GPT-4 possesses a broader knowledge base and stronger problem-solving abilities, with improved performance in creativity, visual input processing, and handling longer content.
Attached is OpenAI's significant release timeline:
- November 2022: OpenAI quietly released GPT-3.5, followed by the launch of ChatGPT
- February 2023: ChatGPT Plus subscription plan was introduced, starting at $20 per month In March 2023, ChatGPT API, GPT-4, and ChatGPT plugins were successively released
In May 2023, the free ChatGPT iOS app launched with voice input support
In July 2023, ChatGPT added "Custom Instructions" feature and released Android version
In August 2023, ChatGPT Enterprise was introduced
In September 2023, OpenAI released DALL-E3 and integrated it into ChatGPT In November 2023, OpenAI announced updates including GPT Builder, GPT-4 Turbo, and Assistants API at DevDay.
Baidu launched Wenxin Yiyan, igniting the 'Hundred Models War'
Following ChatGPT's remarkable debut, major internet companies, research institutions, and AI startups in China have entered the arena, initiating the 'Hundred Models War'. On February 7, 2023, Baidu announced its preparation for a new large model project—ERNIE Bot (English name ERNIE Bot). By March 16, ERNIE Bot was open for testing, demonstrating its performance in five scenarios: literary creation, business copywriting, mathematical reasoning, Chinese language understanding, and multimodal generation, marking the first shot in China's large model competition.
Its foundational model, Wenxin Large Model, was upgraded to version 4.0 in October. Simultaneously, Baidu released over ten AI-native applications, including a fully revamped new search and new maps. By the end of December 2023, ERNIE Bot's user base had surpassed 100 million.
Baidu's founder, chairman, and CEO Robin Li was named a Global AI Leader by Time magazine. He predicted, "We are about to enter an AI-native era." At the GeekPark Innovation Conference 2024 on December 16, he reiterated: "Only by focusing on AI-native applications can we create value; the progress of large models is not an opportunity for most people." AI Stefanie Sun Sweeps Across the Chinese Music Scene
In May 2023, 'AI Stefanie Sun' made a stunning debut, with netizens using the AI voice conversion technology Sovits4.0 (full name So-vits-svc) to replicate her voice, covering songs like Jay Chou's 'Hair Like Snow' and the Yunnan folk song 'Give Your Big Butt a Pinch,' accumulating over 10 million views on Bilibili.
However, such synthesized AI singers pose copyright infringement risks, as unauthorized use of others' voice materials for AI audio training and release may constitute copyright infringement. Stefanie Sun's personal response escalated the event to a new level. She said, 'What's there to compete with someone who releases a new album every few minutes?'
She believes that the emergence of AI will threaten thousands of jobs created by humans, including singing. 'Ironically, it's only a matter of time before humans can no longer surpass it,' she added. 'You're not special; you're already predictable, and unfortunately, you're also customizable.'
From a positive perspective, the development of generative AI has further lowered the barrier to AI music production and covers. Platforms are beginning to embrace AI with creators, standardizing AI music. For example, YouTube has collaborated with singers like Charlie Puth and Troye Sivan to test the Dream Track feature, which can generate music clips imitating the artists' styles.
The rise of AI portrait products like Miaoyan Camera On July 17, 2023, the Miao Duck Camera mini-program was officially launched and quickly went viral on social media.
Users only need to upload a certain number of photos and select templates to generate AI-styled portraits with one click, including ID photos and artistic portraits. Although the first generation requires a 9.9 yuan fee for digital avatar creation, thousands of people still queued up late at night during its peak popularity.
The Miao Duck Camera was developed by Alibaba's Digital Media and Entertainment team, which internally praised it as "the first C-end product to go viral in China's AIGC field". Subsequently, many similar AI portrait products emerged in China, with apps like B612 Kaji, Meitu, and Lightleap also offering similar AI portrait/avatar features. The main challenge for such tool-based products is how to sustain their popularity amidst competition and continue to attract more users. In the following months, Miaoya Camera continuously updated its photo templates, some of which were collaborations with brands. Additionally, it introduced new features such as hairstyle design, AI face retouching, and group photo-taking, and partnered with other Alibaba-affiliated businesses, such as sponsoring the variety show Street Dance of China 6.
Interim Measures for the Management of Generative Artificial Intelligence Services Announced
Over the past year, the rapidly developing AIGC industry has welcomed national-level regulatory policies. On April 11, 2023, the Cyberspace Administration of China (CAC) publicly solicited opinions on the "Interim Measures for the Management of Generative Artificial Intelligence Services (Draft for Comment)." We provided a timely and focused interpretation at the first opportunity. In July, the "Interim Measures for the Management of Generative Artificial Intelligence Services" (hereinafter referred to as the "Measures") were officially announced and came into effect on August 15.
The Measures propose that the state adheres to the principle of balancing development and security, promoting innovation while ensuring lawful governance. It introduces effective measures to encourage the innovative development of generative AI and implements an inclusive, prudent, and classified regulatory approach for generative AI services.
Regarding the regulation of generative AI, countries worldwide are still in the exploratory phase. The Measures represent a pioneering effort by Chinese legislators, laying the groundwork for more comprehensive and detailed future legislation. Currently, the Artificial Intelligence Law has been included in China's 2023 legislative agenda, signaling that unified AI regulation is now on the horizon. Smarter AI Drawing Tool: DALL·E3 + ChatGPT
DALL·E3 is an AI drawing model launched by OpenAI in September 2023, natively built on ChatGPT. Users can transform their ideas into accurate images through conversation, even generating correct text.
For example, when a user asks ChatGPT, "What should my 5-year-old's super sunflower hedgehog look like?" ChatGPT immediately writes four different styles of prompts and generates corresponding images. For casual users who struggle with crafting prompts, this improvement significantly lowers the threshold for AI-generated art.
Compared to Midjourney, DALL·E3 offers more intuitive and convenient interactive drawing capabilities, capable of directly interpreting lengthy text descriptions to automatically generate images. Moreover, Microsoft - which has a deep partnership with OpenAI - has integrated DALL·E3 into Bing, making it freely accessible.
From being ousted by Apple to his triumphant return, Jobs took 12 years. The "Jobs of the AI era" achieved this in just 4 days. On the afternoon of November 17, 2023 (US time), OpenAI suddenly announced a leadership change. CEO Sam Altman was fired because he "was not consistently candid in his communications with the board, hindering its ability to exercise its responsibilities."
The news shocked the entire AI community. The drama saw several reversals—one moment Sam Altman was back at the company negotiating, seemingly poised to return to OpenAI, and the next, Microsoft announced that Altman and his colleagues would join Microsoft to lead a new AI research team. Ultimately, with the support of numerous colleagues and Microsoft's backing, Altman returned. The board underwent a major reshuffle, adding a non-voting observer—Microsoft VP Dee Templeton.
The exact cause of this executive turmoil remains unclear, but OpenAI has since returned to normal operations, officially launching the GPT Store and continuously building its AI commercial ecosystem. Pika 1.0 Released: The Dawn of AI Video Era
On November 28, 2023, the AI video generation product Pika 1.0 was released, with its official announcement post surpassing 10 million views within just one day.
The beta version of Pika supports generating videos in various styles by inputting text or images. The newly released Pika 1.0 introduces features such as real-time local repainting and video size expansion, delivering results comparable to Runway Gen 2 and even surpassing it in terms of usability and 3D animation generation. We experimented with AI tools like ChatGPT and Pika to create a Christmas animation. For enthusiasts without professional animation training, AI can now help realize 70-80% of the storyboard.
Pika became an overnight sensation. Surprisingly, its team consists of only four people. Within six months of its founding, it attracted 500,000 users on Discord, and the company's valuation exceeded $200 million.
The field of AI video generation is advancing rapidly, with new players entering the market. Besides Pika, Meta released Emu Video, Stability AI launched its first open-source AI video model SVD, and Midjourney is also about to release a video generation model. The New York Times Sues Microsoft and OpenAI for Copyright Infringement
This could be the most challenging AI copyright lawsuit in OpenAI's history.
At the end of December 2023, The New York Times sued Microsoft and OpenAI for copyright infringement, attaching 220,000 pages of evidence that included up to 100 irrefutable examples demonstrating that ChatGPT's outputs were nearly identical to their news content. The New York Times has demanded that OpenAI and Microsoft destroy models and training data containing infringing materials, holding them accountable for "billions of dollars in statutory and actual damages" caused by the illegal copying and use of the Times' uniquely valuable works.
OpenAI responded in a statement, asserting that its use of publicly available internet materials to train AI models is justified, and that it has provided opt-out options. The company also pointed out that The New York Times did not present the full facts, suggesting that instances of AI "plagiarism" were the result of intentionally manipulated prompts to induce the model to "regurgitate" content.
The copyright dispute over AI models and training data has drawn significant attention, and this lawsuit could become a landmark case shaping the future of the AIGC (AI-generated content) field.
Through these selected key events, we have reviewed the vigorous development of the AI industry in 2023. We have every reason to believe that 2024 will bring even more exciting technological innovations and practical applications. "At least in the next five or ten years, this technology will be on a very steep growth curve. All existing models will become the dumbest models." As Sam Altman recently stated in an interview, multimodality, customization, and personalization are crucial. AI can handle more complex tasks and truly enhance productivity.
AI will continue to unlock its potential across various fields like content creation, healthcare, and education. Meanwhile, as AI technology advances, we will face new challenges including ethical dilemmas, privacy concerns, and security issues. We must adapt quickly to ensure this technological power serves all of humanity.