AIGC explosion? Will intelligence replace humans?

  In 1968, American sci-fi writer Philip Dick released his most famous work – “Do Androids Dream of Electric Sheep”, which is also the original novel of the classic sci-fi movie “Blade Runner”. In the novel, the author explores humanity and ethics by telling the game between natural people and bionic people: When artificial intelligence develops to a certain level, when robots can also think, sleep, and dream, how should we get along with artificial intelligence?
  Since Alan Turing proposed the concept of “Turing Test” in his paper “Computing Machines and Intelligence” in 1950, artificial intelligence technology has become the most common element in human society’s prediction of the future world, accompanying it What is more, is the discussion or even worry about whether artificial intelligence will surpass and replace human beings. Today, artificial intelligence is showing a level closer to human intelligence, especially since 2022, the application of AIGC (artificial intelligence production content) in the field of painting and writing has made people exclaim that artificial intelligence can be “like human beings” to create.
  In November 2022, OpenAI, an American artificial intelligence research institution, released a new chat robot model ChatGPT. It can not only “speak and write”, but also automatically fix bugs, triggering “crazy” chats between the whole people and artificial intelligence after its release. So, what are the business opportunities behind the wave of AIGC represented by ChatGPT?

A lively dialogue between field and field intelligence

  On November 30, 2022, OpenAI, an American artificial intelligence research laboratory founded by Silicon Valley technology tycoons such as Musk, Altman, president of the American business incubator YCombinator, and Peter Thiel, co-founder of the global online payment platform PayPal, released a free robot dialogue The model ChatGPT triggered a carnival in the AIGC community.
  GPT (pre-training model) is the artificial intelligence model behind the AIGC that is currently receiving much attention. In 2018, 2019, and 2020, OpenAI successively released three generations of models: GPT-1, GPT-2, and GPT-3. The parameters of each generation model have increased by more than 10 times, or even 100 times, compared with the previous generation.
  Among them, as the first Transformer-based pre-training model, GPT-1 adopts two stages of pre-training + FineTuning, uses Transformer’s decoder as a feature extractor, stacks 12 layers in total, and has 110 million parameters. Compared with GPT-1, GPT-2 uses more network parameters and a larger data set. Its largest model has a total of 48 layers and a parameter volume of 1.5 billion. GPT-3 was released in 2020. It has the same structure as GPT-2. Its model has reached 96 layers, and the model parameters are about 100 times that of GPT-2. It can be seen that the number of parameters used in the three generations of models has increased significantly.
  The ChatGPT released this time belongs to the GPT-3.5 stage. ChatGPT adopts the RLHF (reinforced learning from human feedback) method for the first time, and begins to have the ability of association and memory, and has a stronger understanding of human language. Users can interact with ChatGPT by chatting, consulting information, writing articles, modifying codes, etc. For a while, various “experience posts” using ChatGPT appeared on the Internet. From writing emails, writing copywriting to writing code, ChatGPT has been watched on a large scale. Musk even asked ChatGPT how to design Twitter, and posted the conversation on Twitter.

  Since its launch, ChatGPT has had more than 1 million users. As a conversational AI based on GPT-3.5, the biggest feature of ChatGPT is that it can “understand” the semantics of the interlocutor, enabling more effective feedback and continuous dialogue. Based on deep learning of ultra-large-scale data, ChatGPT can almost “disguise the real” in the field of text generation, making users feel that it really has consciousness.
  However, ChatGPT is not the current perfect model. According to the reminders given by the OpenAI R&D team when ChatGPT was launched and repeated trials by netizens around the world, it was found that ChatGPT has at least three problems: one is the large-scale language model behind ChatGPT. The most recent data of the training set ends at the end of 2021, which means that it cannot give an accurate answer for what happened in 2022; second, the quality and accuracy of ChatGPT’s answers are not stable; third, the questioner The entered question description will affect the accuracy of the ChatGPT answer.
AIGC? Extravagant

  In fact, AIGC represented by GPT has become popular all over the world this year. In the context of the fiery concepts such as Metaverse and Web3, AIGC also took advantage of the momentum to rise. The so-called AIGC corresponds to the concepts of PGC (Professional Produced Content), UGC (User Generated Content), and PUGC (Professional User Generated Content) on the Internet. AIGC is the automatic or assisted generation of content through artificial intelligence technology.
  Since 2022, artificial intelligence comics have become popular all over the Internet, and multiple application software have emerged at home and abroad, including DiscoDiffusion launched by Google in February 2022, DALL E2 by OpenAI in April 2022 and Make-AScene by Meta in July 2022 wait.
  Among them, DALL·E is a new model released by OpenAI on January 6, 2021, which can generate images based on text, and an upgraded version, DALL·E2, was released more than a year later. Compared with DALL·E, DALL·E2 has higher resolution and lower delay when generating images described by users, and the new version also adds new functions such as editing original images. DiscoDiffusion is an artificial intelligence image generation program that became popular in early February 2022. It can render corresponding images based on keywords describing the scene. It can be run directly on Google Drive or deployed locally.
  The current fiery artificial intelligence painting can be traced back to 2014 before the establishment of OpenAI. Artificial intelligence expert Ian Goodfellow proposed the algorithm model generation confrontation network in this year. Based on a deep convolutional neural network, a generative adversarial network confronts two neural networks, the generator and the discriminator. Among them, the generator is used to generate “fake data”, and the discriminator is used to judge the authenticity of the data. And the two have also gradually evolved a powerful “falsification ability” in the confrontation, which is used for image synthesis. Since 2015, the generative confrontation network has been put into application and has become the most common algorithm model in artificial intelligence generation and image processing tasks.
  In the past few years, more and more players have emerged in the AIGC field. Within two years after the launch of OpenAI’s GPT-3, similar large-scale language models such as GoogleBERT have emerged, but the status of GPT-3 is still unshakable. In 2021, GPT-3 will also be selected as one of the “Top Ten Breakthrough Technologies in the World” by MIT Technology Review. It is believed that this large-scale computer model for learning natural language is moving towards building a system that can understand humans and interact with the human world. A big step forward for AI.
  Whether it’s AI painting or AI chatting, writing, or modifying code, artificial intelligence is undoubtedly becoming more and more creative. In 2022, the international research organization Gartner will evaluate AIGC as “one of the five most influential technologies of the year”.
AIGC? Extravagant

  The emergence of such a phenomenon-level technological application naturally cannot escape the eyes of capitalists with a keen sense of smell. Objectively speaking, although all kinds of AIGC applications are popular at this stage, it will still take some time before large-scale application or commercialization. However, this does not prevent capital from investing heavily in the AIGC field. At present, many AIGC companies have obtained capital support.
  In October 2022, Jasper.AI, an artificial intelligence unicorn company that focuses on text generation, announced the completion of a US$125 million Series A round of financing, with a post-financing valuation of US$1.5 billion; StabilityAI also announced that it had received US$101 million in financing, becoming a valuation With more than 1 billion unicorns, the company said it will continue to develop AI generation models for generating pictures, languages, and videos. The influx of capital is also continuously pushing up the valuations of AIGC-related companies. For example, the valuation of big data plus artificial intelligence company Databricks has reached 38 billion U.S. dollars, and the valuation of OpenAI has exceeded 20 billion U.S. dollars in October this year.
  In my country, a number of AIGC applications have also sprung up, many of which are Internet giants. For example, Baidu launched an artificial intelligence art and creative assisted painting platform Wenxinyige; Tencent created a writing robot “Dream Writer” ; Lubanner, an artificial intelligence online design platform under Alibaba, helps marketers produce banners. Gartne predicts that by 2025, data generated by generative AI will account for 10% of all data.
  It can be seen that artificial intelligence technology is “evolving” rapidly, and its weight in the development of human society is becoming more and more important. Following this, the harmonious coexistence of human beings and artificial intelligence has become a topic of concern for scientific and technological elites. As Musk said when talking about the original intention of OpenAI: “What can we do to ensure that the future brought by artificial intelligence is friendly?” The answer to this question may lie in the efforts of every scientist.

error: Content is protected !!