What is shared today is:AI seriesIn-depth Research Report:AI Topic: Content Quality Improvement and Technological Innovation, Looking Forward to the Implementation of AI Applications in 2024
Report Producer: Caixin**).
Report total:Page.
Featured Report**: The School of Artificial Intelligence
The overall regulatory posture has been relaxed at the margin, and the industrial application of new technologies and models represented by AI general artificial intelligence, AR VR MR, and data elements has been actively emphasized. On the one hand, since the beginning of this year, we have observed that positive results have been achieved in the rectification and governance of network security, Internet platform economy, pan-cultural entertainment and other fields, the operation of the industry has become more standardized and rationalized, and the overall supervision of the Internet economy has shown a marginal relaxation trend. On the other hand, under the stock economy of the Internet, the regulator attaches great importance to the application and development of new technologies such as AI and virtual reality, and timely introduces supporting policies to guide them.
Looking back on the media since 2014: First, from 2014 to 2015, the media index was significantly larger, mainly due to the rapid development of the mobile Internet industry, the relatively relaxed policy and regulatory environment, and the accelerated growth of mergers and acquisitions in the capital market. Second, from 2016 to 2018, the strengthening of supervision and the sequelae of mergers and acquisitions led to a decline in its own performance year after year, and the media index entered a long-term adjustment. As the profitability of M&A assets fell short of expectations, there was a goodwill thunderstorm, inventory accounts receivable and other large asset impairments, which seriously dragged down the performance of the sector; In addition, the supervision of the entertainment field was strengthened in 2018, such as the suspension of the distribution of game version numbers from April to November 2018, the sky-high remuneration, yin and yang contracts, and tax rectification, etc., and the media index was adjusted at a low level for a long time. Third, from 2019 to 2020, the release of supply and traffic dividends will drive the media index**. With the resumption of the issuance of game licenses in December 2018, the issuance of 5G licenses revolutionized the computing network infrastructure and gave birth to the concept of cloud gaming, and the home traffic dividend of the epidemic in early 2020 drove the media index to the bottom**. Fourth, from 2021 to 2022, the overall trend will be downward under the heavy pressure of regulation, but the concept of the metaverse driven by VR hardware generation will drive the media index to rise for a short time. In 2021, the regulator launched a heavy blow against multiple fields such as anti-monopoly of the platform economy, K12 education double reduction, online data security, minors' addiction to online games, and the rectification of the entertainment and fan circle, and the second round of distribution of game version numbers was suspended from August 2021 to March 2022. However, with the explosion of the metaverse concept and the development of 5G networks in the second half of 2021, VR hardware is expected to drive the Internet into the next generation of development, driving the media index to rise for a short time.
Since 2023, the media sector can be roughly divided into three waves***First, from the beginning of 2023 to the end of April, marked by the release of overseas ChatGPT in November 2022, many Internet and technology companies such as domestic, iFLYTEK, and SenseTime have successively announced self-developed large models, driving the media sub-sector to form a general rise**. In May, as AI sentiment in the market subsided, the media index peaked at a down** of nearly 20%. Second, in June 2023, the market gradually formed a consensus on the main line of science and technology throughout the year, games have become the most promising industry for AI, and the published corpus has a high value for AI large model training, so games and publishing have led the rise**. From July to October, some major shareholders of AI concept companies announced their divorce and **announcement, triggering **start**; In addition, some game companies reported less than expected, the third quarter report lacked bright spots, and the market catalyst was less, and the market sentiment of AI weakened. As of late October 2023, the media index is down 35%-40% from its peak in June. Third, from November to early December 2023, under the leadership of innovative business models such as live-action interactive games and short dramas, film and television, games, IP publishing and other sectors took the lead in opening ***At the end of November, the AI multimodal model represented by PIKA ** generation tool and Google's Gemini made more progress than expected, superimposed on the weak pro-cyclical sentiment of the market, and the media index started to rise**, compared with the low point level in October, the media index was the highest** by more than 20%.
Overseas AI giants are leading the way, and multimodal iteration is progressing rapidly.
1) Represented by OpenAI, the multimodal iteration speed is eye-catching. Since the beginning of this year, ChatGPT's multimodal function iteration has exceeded expectations, and GPT-4 was launched in March, with a higher level of image recognition ability and longer contextual input restrictions. In the second quarter, ChatGPT launched many plug-ins and networking functions, and targeted large language model APIs (including GPT-4 and GPT-3).5-turbo) has added major updates such as function calling function and reducing usage cost. After entering the third and fourth quarters, the iterative acceleration of AI multimodal functions: in September, ChatGPT added voice and image functions, and introduced the "consensus" plug-in to make **retrieval more efficient; At the same time, the Dall·E3 version of the Wensheng diagram model was announced; At the first Worldwide Developers Conference in November, OpenAI launched GPT-4 Turbo, GPTS and other model iterations, supporting longer text, more modalities (support** text input, integration of DALL-E 3 speech synthesis speech-to-text and other cross-modal fusion capabilities), support user-defined GPTS, developer-defined assistants API, etc. With CEO Altman back on the board, OpenAI expects the GPT store to go live in early 2024; We expect the smarter GPT5 to be launched next year as well.
2) Google released Gemini, a multimodal large model. On December 6, Google announced the multimodal large model gemini1Version 0 is officially launched (benchmarked against GPT-4), which can automatically generate text, summary content, etc., and can understand **, audio and **content. Gemini is available in three versions: Nano, Pro, and Ultra. Among them, Ultra is the most powerful version, which can be used for highly complex tasks, and the AI computing power consumption is very large; The parameters of Pro are only half of those of Ultra, which is very suitable for large-scale application deployment, and has been connected to Bard; Nano is a lightweight model for mobile devices and personal, small-scale deployments. In addition, Google has also launched the most advanced Wensheng diagram large model, Imagen 2, which can generate high-quality, realistic images based on user prompts, and adds the ability to render text and logo design, which can be benchmarked against head Wensheng diagram models such as Dall·3 and Midjourney.
3) Generate tool pika1Version 0 is generally available. Pika's founding team is Pika Labs, an American AI startup, with a team of just 4 people. In late April, the first version of PIKA was in open beta on Discord; In July, the beta version of Pika officially launched on Discord; November 29, pika 1Version 0 is officially launched, which can generate various styles of **, animation, 3D animation, movies, etc. according to the user's language description, and can also realize **style conversion, expand the canvas background with high consistency and other functions through simple language description. Pika's overseas counterparts are mainly Runway and Stability AI, with Runway launching Gen-1 and Gen-2 in February and November this year, respectively; In November this year, Stability AI launched Stable 3D, a new open-source Wensheng ** diffusion model, Stable Video Diffusion, and an open-source Wensheng graph model, SDXL Turbo.
The TOC paid subscription has driven OpenAI's revenue to soar, and commercialization has begun to take shape, and there is room for future development. According to foreign media The Information, OpenAI's CEO Altman revealed that since the launch of the paid version of ChatGPT in February this year, OpenAI's annual revenue has reached $1.3 billion (roughly as of October 2023), an increase of 30% from this summer and 0$2.8 billion in revenue, a more than 45-fold increase. The company's revenue growth rate has increased significantly, mainly from user subscriptions to its conversational chatbots. In terms of users, since the release of ChatGPT on November 30, 2022, it has accumulated 100 million users in just 2 months, and it has taken 9 months and 30 months for the world's top consumer apps TikTok and Instagram to reach 100 million users, respectively. According to the AI New Intelligence Circle, in May 2023, ChatGPT reached a peak of 1.9 billion global visits, dropped to about 1.4 billion in August, rose again to about 1.5 billion in September, and reached about 1.7 billion in October and November. At the same time, the use of the ChatGPT app is also rising month by month, from 300 million in August to about 1 billion in November.
Film and television theaters: improve the quality and efficiency of content, and pay attention to the distribution of short dramas and films
In the context of marginal relaxation of policy supervision, the demand for offline entertainment has recovered after the epidemic, and the content industry has shown a trend of improvement in both supply and demand.
The popularity of micro-short dramas continues to rise, and the number of filings and online dramas has grown rapidly. The State Administration of Radio, Film and Television defines a short drama as "a single episode lasting from a few tens of seconds to about 15 minutes". Compared with the long drama, the production cost of micro-short dramas is low, the production cycle is greatly shortened, the filing and online process is relatively simple, the return on investment is higher, and the plot setting can more intensively hit the audience's cool point. Therefore, in the context of the current diversification of user needs, micro-short dramas have ushered in rapid development. According to the data of the State Administration of Radio, Film and Television, the average number of micro-short dramas filmed and filed in the country in the past two years has exceeded 2,500, an increase of more than 5 times compared with the level in 2021. In addition, according to Detaven data, from October 2022 to the end of August 2023, a total of 1,125 micro-short dramas were launched on the long-short ** platform, compared with less than 500 in the past three years, achieving leapfrog growth; In terms of platforms, the short** and long** platforms have become the main broadcast platforms for micro-short dramas, with 455 new and 214 new ones on Douyin and Kuaishou respectively; Youku, Tencent, and Mango followed with 189, 176, and 66 new releases respectively; iQIYI has only launched 12 films; Long-tail platforms such as Station B, Blockbuster TV, and Sohu** have also gradually entered the micro-short drama, with less than 10 online works.
We estimate that the current market size of micro-short dramas is around 20 billion yuan, and it is expected to continue to maintain rapid growth in the future. According to ByteDance data, in September 2023, the average daily consumption scale of the short drama industry** exceeded 50 million yuan. According to the new wrist, the peak daily consumption of the short drama industry during holidays such as the National Day exceeded 100 million, so we estimate that the current micro-terminal drama market size is around 20 billion yuan. With the recent regulatory rectification of the short drama industry, Douyin has tightened the quality of investment on behalf of the platform, WeChat, Kuaishou, etc. have successively strengthened the review of short drama content, and there are strong producers represented by traditional film and television companies entering the market, and the industry is expected to usher in a more standardized and healthy development. However, we believe that the emergence of the short drama industry actually meets the needs of people at different levels, especially the fragmented entertainment needs of users in the sinking market, so the industry is expected to continue to maintain rapid growth in the future.
The film is released on a trial water line to meet the needs of more levels of movie-watching. Separate distribution, that is, different from the previous unified screening of a film in theaters across the country, but the film party can choose to provide better conditions for the cinema chain or cinema investment management company to trade, such as providing more screenings, a higher proportion of film scheduling, a longer screening cycle, and a higher proportion of film box office settlement and accounting, all of which belong to the category of merit-based consideration. In the "14th Five-Year Plan for the Development of Chinese Films", innovative businesses such as sub-line distribution, multi-round distribution, regional distribution, and focus distribution are clearly encouraged. At the 24th National Film Promotion Conference and the first National Film Trade Fair held recently, 24 films including "Reunion Again", "Porco Rosso" and "Asteroid Hunter" completed the sub-distribution contract, and "Silent Notes" released on December 1 became the first new film to be released in sub-production. The distribution model of Chinese films has been officially launched, and on the basis of unified distribution and screening, market-oriented distribution has been explored.
Report total:Page.
Featured Report**: The School of Artificial Intelligence