Image source: Visual China
In the past week, the global AI field has ushered in a moment of confrontation.
On May 13, local time, OpenAI released the new GPT-4 series model GPT-4o, as well as the desktop version of ChatGPT, through live broadcast. OpenAI CEO Sam Altman, who was not physically present at the live broadcast, posted on X: "Our new model, GPT-4o, is our best model so far, it is smart and fast, and it is still a native multimodal model. ”
OpenAI CTO Mira Murati said GPT-4o will be available to all users, including free users, and it will be faster than the previous generation, with improved text, video, and voice capabilities, and improved service quality for ChatGPT in 50 different languages. GPT-4o will also be powered by OpenAI's API, where developers can build applications using the new model.
During the live broadcast, Murati and two technicians conducted a field test of GPT-4o. The new model reduces the sluggishness of human-machine conversations, and in the process, people can interrupt the output of GPT-4o at any time, and the communication of emotions has improved, and combined with video capabilities, the machine is more like a real person who can see and hear.
If the previous GPT was based on data and gave people the shock of technological leaps at the level of rational analysis, then this time GPT-4o presents a supplement to the perceptual level similar to that of humans, which makes people sigh that the plot of a science fiction movie has happened.
In this iteration, OpenAI did not meet the expectations of the outside world for GPT-4.5 or GPT-5, but the "o" is a more direct expression of an evolution than a number. According to OpenAI's official website, the "o" here stands for "omni", which means "all-powerful" and is derived from Latin. In English, it is often used as a root word that can be associated with "omniscient" and "everywhere".
It is worth noting that the GPT-4o release point was stuck the day before the Google I/O developer conference. In the coming month, Microsoft's Build annual developer conference and Apple's WWDC will also be held successively.
On May 14, local time, Google announced new AI progress at the I/O developer conference, and under the paving of OpenAI the day before, the market will naturally look at the new products of this old giant with a higher vision.
OpenAI did manage to put pressure on Google through this simple and short live broadcast of the set - the real-time dialogue display of the developer's site is reminiscent of the display video previously released by Google's large model Gemini, which was accused of editing and exaggerating publicity. In practical application, the model did not achieve the effect of its publicity. In response, Google admits that the demo video has been edited to reduce latency for brevity, as well as the Gemini output. The 2024 I/O Developer Conference is a key opportunity for Google to save the day.
In fact, Google CEO Sundar Pichai stated at the I/O developer conference seven or eight years ago that he would shift from "mobile first" to "AI first", but Google did not continue to take the lead under the condition of seeing the direction in advance. Google, which was accustomed to being at the top of the previous generation of the Internet era, seems to have become a catch-up after experiencing the challenges around AI in the past year. It needs to send a positive signal to the outside world about the latest technology.
At the I/O developer conference, Google showcased Project Astra, which can compete with GPT-4o, demonstrating multimodal understanding and real-time conversational capabilities. Through the demonstration, it can be seen that Astra's latency is longer than GPT-4o, and the emotional diversity needs to be improved.
At the same time, Google also launched Veo, a video generation model that benchmarks against Sora. In February of this year, the Sora was launched at about the same time as the Gemini 1.5 Pro.
In terms of search engines, Google has taken the initiative to revolutionize itself, introducing AI Overviews in the search function to summarize search results. Google says the search will have multi-step inference capabilities and support solution searches in the form of videos. At the same time, Google has expanded Gemini 1.5 Pro, which has a context window of 1 million tokens, to 2 million tokens, which means that the large model can process more information and have a more comprehensive understanding.
Since ChatGPT has been a direct threat to the search field since it entered the public eye at the end of 2022, the market has speculated that OpenAI will launch a ChatGPT-based search engine at this press conference. But the company has not announced such a step at this time.
OpenAI and Google's R&D race is at a stalemate, which threatens companies including Apple and Amazon. They need to keep up with the ever-increasing pace, at least not too far behind, or both Siri and Alexa will be embarrassed. Since Apple's AI R&D progress is uncertain, there are already rumors that it is negotiating with OpenAI to license the technology to apply the most advanced AI technology to Apple products in the short term.
AI competition has spread from infrastructure to application and ecosystem construction.
Google's accumulation of enterprise applications and consumer applications provides a natural advantage for it to build an AI ecosystem. Perhaps OpenAI has noticed this. In January of this year, OpenAI launched the GPT Store, a personalized AI app store through which users can discover and build GPT. Currently, the GPT Store is mainly open for paid versions. This feature is considered to be the next Apple Store. From the announcement at the end of last year to the beginning of this year, it lasted two months, during which it experienced the turmoil of Ultraman being unexpectedly removed and reinstated.
Up to now, GPT Store is still just a channel, and there is still a big gap between the establishment of the ecosystem envisioned by the public through the potential of OpenAI. However, with the technological iteration of GPT-4o, the imagination of AI applications has been further enriched, and the possibility that OpenAI is building an architecture for the ecosystem cannot be ruled out.
At the end of OpenAI's conference, Murati said: "Today we are focused on free users, new models, and new products, but we are also very concerned about the next frontier. Soon we will update you on the progress of the next big event. ”
This has to lead to speculation that OpenAI has reservations about actual progress. In the face of fierce competition, keeping hole cards may have become a habit for the major giants. The official announcement time of OpenAI's "next big event" may be found at another time node after the follow-up press conferences of Microsoft and Apple. After all, Ultraman has always been good at releasing the right message at the right time. OpenAI's previous technology exposure rhythm has achieved good results. (Fortune Chinese Network)
Fortune magazine launched the "Fortune China Technology 50" list for the first time this year, trying to find these technology companies that were born in China and are influencing the world. Their success lies not only in their technology and products, but also in the spirit of innovation and global vision they represent. We look forward to these Chinese companies making more outstanding achievements in technological innovation and global expansion, and making greater contributions to technological and commercial progress.
At Fortune Plus, netizens have made many in-depth and thoughtful comments on this article. Let's take a look. You are also welcome to join us and talk about your ideas. Other hot topics today:
Check out the actionable view of "EU says Chinese automakers don't provide enough information to raise tariffs".
Check out the great view of "Microsoft is preparing to launch its own AI model".