In the past year of 2023,GPT large modelIt has opened the "pandemonium box" for ordinary people to enter the world of AI, and has also kicked off a large-scale model "arms race" among technology companies around the world.
As large models of text, images, and videos emerge, the application side is in a state of rapid innovation, which also leads to a shortage of AI chip resources. The "AI replacement theory" and "risk doubts" are constantly being played out. On one hand, AI is the human's fanatical pursuit of advanced productivity, and on the other hand, it is the human's concern about new technologies.
Looking back from 2024, AI is still far from replacing humans, and technical bottlenecks have also emerged: chip technology limitsAI(AI) Towards General Artificial IntelligenceAGI) speed; the data used to train large models are regarded as "precious materials" and are in the hands of Internet companies; and commercialized AI applications have not yet benefited everyone. Domestic ones are not easy to use and overseas ones are not cheap.
If we look at it from a longer time perspective, the current AI model is at most like humans just discovering the source of fire. How and where to use it will be the direction of AI in 2024. We also see that AI technology companies such as OpenAI and Google have begun to develop chips, and the "fire thieves" are turning up the heat.
Source Note: The image is generated by AI, and the image is authorized by Midjourney
Internet "New World"
In 2023, a lot of history has appeared in the field of AIfirst, many of which are brought by ChatGPT.
ChatGPT allows ordinary peopleFirstThe first time I saw computers understand natural language, artificial intelligence is no longer the magical "secret weapon" in movies; non-human ChatGPTfirstListed as one of the “Top Ten Scientists of 2023”.
From its release in November 2022 to now, the attention and impact of ChatGPT invented by OpenAI have surpassed almost all hot spots in the history of information technology: the number of users reached 1 million in 2 days and 100 million in 2 months, breaking TikTok's previous record. Half a year later, the ChatGPT iOS App quickly topped the Apple App Store's overall rankings as soon as it was released.
What is more epoch-making for ChatGPT is that it has successfully broken the monopoly of technology giants on AI technology and brought products that can understand human language to every ordinary person.
ChatGPT is a lifelongFirstThis is the first time I have come into contact with such a highly intelligent dialogue system. It can write copy, do math, know astronomy and geography, understand stories, and understand Internet memes. Although it will "talk nonsense seriously" and produce hallucinations at first, it will correct itself with human prompts.
The speed of updating of the general language model behind ChatGPT is also amazing. In just one year, it was upgraded from GPT3.5 to GPT4Turbo. More than a month ago, in early November 2023, OpenAI officially announced the GPTs plan at the Development Day. ChatGPT Plus users can use the data they own or find on the basis of GPT-4 to train an exclusive ChatGPT dialogue robot.
OpenAI once again surprised everyone, and the trademark application for GPT-5 has been registered. According to the information displayed by the US Patent and Trademark Office, the functions provided by GPT-5 include natural language processing, text generation, understanding, speech transcription, translation, prediction and analysis.
OpenAI files trademark application for GPT-5
In OpenAI's official blueprint, they will later launch a GPTs store to support users to list their trained robots and provide them to other users for paid use, thereby establishing a new business model. In less than two months, users have built more than hundreds of thousands of ChatGPT assistants in specific fields through custom GPTs, which shows how popular it is.
In 2024, if OpenAI opens the "GPT Store", another wave of application craze will sweep the Internet.
Application explosion
OpenAI's success is like Columbus' discovery of the New World. He told everyone: This road is feasible. In just one year, ChatGPT directly stimulated the artificial intelligence arms race among global technology companies.
Data shows that as of October this year, in less than a year, there are more than 250 Chinese manufacturers and universities with large models with more than 1 billion parameters, not including the number of large models from abroad. AI applications are gushing out. According to the Sensor Tower report, in the first half of 2023 alone, AI application downloads increased by 114% year-on-year, exceeding 300 million times, exceeding the full-year level of 2022. At the same time, AI application in-app purchase revenue surged by 175% year-on-year, approaching $400 million.
In the fierce competition, thousands of large models have evolved powerful multimodal capabilities such as text-to-image, image-to-image, text-to-video, and image-to-video. People were just amazed at ChatGPT's eloquence, and in the blink of an eye, models such as Bard and Clude that recognize Internet memes appeared.
In addition, some vertical fields have also produced "unicorns".
In terms of image generation, Midjourney takes the lead and becomesStrongestFrom its birth to now, in just half a year, Midjourney has been upgraded to version V6, and has developed from the initial text-generated images to image-generated images and AI-enlarged images.
What’s even more amazing is that behind Midjourney is a team of 11 people that was established only two years ago. As Midjourney became more popular, the team has expanded to 40 people, and achieved $200 million in revenue this year alone, achieving financial independence early on.
Unlike most startups that seek venture capital, Midjourney has not taken a penny from VCs. "To put it politely, he doesn't need VCs to enter his life," said Michael Stewart, a partner at M12, a venture capital fund under Microsoft.
The speed at which AI applications are developing can be described as “daily updates”.
Shortly after Midjourney became popular, Runway's Gen-2 took over in the video field, just one month after its first generation product Gen-1.up to dateThe latest version of Gen-2 can not only generate an 18-second video based on a single prompt, but also use movie camera language proficiently. Recently, Gen-2 has added the function of generating videos from images, which can make it move by simply "painting" a certain area in the image.
AI is entering the video field, and rising stars such as Pika1.0 and Stable Video Diffusion are catching up. Various AI video tools are "fighting" and evolving at an increasingly fast pace. Among them, the video capabilities of Pika1.0, which was just launched not long ago, are already comparable to those of Gen-2, and evenfirstBringing AI image enlargement capabilities to the video industry.
In 2024, these unicorns will continue to explore the multimodal potential of large models, and even create new unicorns. Text, pictures, audio, video and other products that represent human natural language will continue to be carefully cultivated by AI tools, and the tools will likely become more handy and cheaper at scale.
Chip bottleneck
Big tech companies are competing for big models, small companies are competing for applications, and in addition to the competition, Nvidia, the "water seller", is making a lot of money. Data shows that Nvidiaup to dateThe revenue in the third quarter reached 18.1 billion US dollars, a year-on-year increase of 206%; the net profit was 9.2 billion US dollars, a year-on-year increase of 1259%. The "leader" OpenAI, which triggered this competition, only had a revenue of 1.3 billion US dollars in 2023. Last year, OpenAI's revenue was only 28 million US dollars.
In 2023, NVIDIA has undoubtedly secured the top spot in the AI market. Chip prices have been rising steadily. Even with the high prices, chips are still hard to come by. It is reported that the delivery cycle of NVIDIA H100 ranges from 36 to 52 weeks. Such a long delivery cycle obviously cannot meet the needs of AI product development. This may also be the reason why GPT-5 has not arrived for a long time.
Computing power is one of the three driving forces for the development of AI. The shortage of chips has directly affected the evolution speed of large models. OpenAI and Google have started to develop their own chips to fill the gap in their own large model training.
Nvidia's competitors are also scrambling to get a piece of the "cake". Intel and AMD have launched Gaudi3 and Instinct MI300X high-performance AI chips respectively, Microsoft has announced the AI acceleration chip Azure Maia100, and Amazon has released Trainium2, an upgraded accelerator chip for AI systems.
At the same time, the US chip sanctions have further exacerbated China's chip shortage crisis. In order to maintain the market, Nvidia can only launch some "castrated" chips with reduced performance.
By the end of 2023, there is still no chip comparable to Nvidia H100, either in China or abroad. Although chip manufacturers are working at full capacity, the shortage of chip resources will continue in the short term. Perhaps only when the AI chip resource problem is solved can GPT-5 and more advanced and rich products come faster.
But from a security perspective, this is not entirely a bad thing. During this brief respite, humans can make a better choice for the direction of AI development.
You know, when GPT-4 was just released, a joint letter from thousands of technology elites pushed the AI safety issue into the spotlight. They jointly requested to "suspend the training of AI systems more powerful than GPT-4." In the words of Gary Marcus, an expert in the field of AI: "Is it worth it to exchange the 1% risk of human destruction for the pleasure of talking to a machine?"
Indeed, when ChatGPT was first launched, its disruptive creativity was truly shocking, and Microsoft founder Bill Gates also emphasized that the possibility of AI getting out of control exists.
At the beginning of this year, the Internet was full of concerns about "AI stealing jobs" and "AI replacing humans", and even the "carbon rapist" meme appeared. But a year has passed, and now everyone's attitude towards AI removal has returned to rationality.
We have not really seen large-scale unemployment of “working people” due to AI. At least in China, the widespread application of AI in work is still in the future. At the same time, representative companies such as OpenAI have to keep up with AI safety measures, and AI safety review agencies have been established in various countries.
The advancement of AI is inevitable, but how “fire” is used depends on human choice. Some people are making furnaces to control the flames, while others are focusing on what other deserts the flames can light up.
In 2024, security will still be one of the themes of AI development, and chips will be the key to further improving productivity.onlyWhat is certain is that AI, like the Internet, will become an indispensable human tool in the future.