Not long after OpenAI released its new flagship model, GPT-4o, the limelight was quickly overshadowed by Google.
On May 14, local time, the Google I/O Developer Conference was officially held, and a series of AI applications including the "Family Bucket" were released, among which many products are considered to be positive benchmarks against OpenAI. Google CEO Pichai said that a total of 121 mentions of AI were mentioned at the press conference that day, which is enough to show how much Google attaches importance to AI.
On the other hand, OpenAI, which has been "targeted", is undergoing a personnel change. On May 14, local time, OpenAI co-founder and chief scientist Ilya Sutskever announced her decision to leave OpenAI. The scientist was seen as a key figure in the events surrounding the ouster of OpenAI's co-founder and CEO Sam Altman a few months ago, but with Sam Altman's return and a board reshuffle, OpenAI's power struggle has come to an end, making Sutskever's departure less "unexpected."
Key figures leave OpenAI
A group of OpenAI executives, including Sam Altman, staged a decent "breakup" with Ilya Sutskever.
As co-founder and chief scientist, Ilya Sutskever decided to leave after joining OpenAI for almost a decade. He posted on social platforms that OpenAI's development trajectory is a miracle, and mentioned the current management of OpenAI, saying that under their leadership, OpenAI will create a safe and beneficial AGI.
Sam Altman was quick to respond, calling Ilya Sutskever "one of the most brilliant thinkers of our generation, a leader in our field, and without him, OpenAI wouldn't be where it is today." After acknowledging Ilya Sutskever's competence and expressing his gratitude, Sam Altman announced the successor to the position of Chief Scientist: Jakub Pachocki. Judging from his resume, the scientist joined OpenAI in 2017 and served as OpenAI's research director, and Sam Altman has praised his leadership and technical ability in the development of GPT-4.
It is worth mentioning that shortly after OpenAI CEO Sam Altman and President Greg Brockman left the company in the midst of the vigorous high-level turmoil in November last year, three OpenAI senior researchers also announced their resignations, one of which was Jakub Pachocki.
Looking back now, the development and follow-up of the so-called "Gong Dou" incident a few months ago also laid the groundwork for Ilya Sutskever's departure today.
In mid-November last year, Sam Altman was announced to be dismissed as CEO and leave the board of directors, and Greg Brockman subsequently announced his resignation as president. In the days that followed, under pressure from both external investors and internal employees, the two executives returned to OpenAI and were reinstated.
In this "fight", Ilya Sutskever, who united with the board of directors against Sam Altman, was considered a "talker" by the outside world, and after Sam Altman's return, Ilya Sutskever expressed deep regret for participating in the board action. After the power struggle ended, OpenAI's board of directors was reshuffled from six to three, and Ilya Sutskever no longer served on the board.
Along with this high-level turmoil, there has been a protracted and comprehensive discussion about OpenAI's internal AI technology roadmap. As a leading scholar in the field of machine learning, Ilya Sutskever advocates caution and pays attention to risk and security, and is considered to be a "scientific conservative". In addition, Ilya Sutskever is one of the leaders of a superalignment team within OpenAI that aims to solve the problem of superintelligence alignment.
Ilya Sutskever did not disclose the specific details of her "next home" for the time being, only mentioning that "this project means a lot to me personally". For Sam Altman and OpenAI, as commercial companies such as Google continue to increase their AI and launch highly competitive tools and applications, how OpenAI, as a non-profit organization, handles the original intention of its establishment and commercial development is still a problem that it needs to face.
Google is menacing
If the departure of the chief scientist is expected to be a normal personnel change, then the rapid development of external business giants may pose even more serious challenges to OpenAI.
On May 13, local time, OpenAI held an online live broadcast, without GPT-5 and no search engine, OpenAI released GPT-4o, "o" means "omni", that is, "omni". According to the official introduction, GPT-4o is a step towards more natural human-computer interaction - it accepts any combination of text, audio, images, and video as inputs, and generates any combination of text, audio, and image output.
Compared to existing models, GPT-4o is particularly good at visual and audio understanding. In terms of reaction speed, GPT-4o can respond to audio input in as little as 232 milliseconds, with an average response time of 320 milliseconds, similar to the human response time in a conversation. Prior to GPT-4o, the average latency of users using voice mode to talk to ChatGPT was 2.8 seconds (GPT-3.5) and 5.4 seconds (GPT-4), which shows its leapfrog development. Moreover, the price of GPT-4o is 50% cheaper.
A day later, Google unveiled a series of Gemini-based AI features and products at the developer conference. This can't help but remind the outside world that on February 15 this year, on the same day that Google released the multimodal large model Gemini 1.5 Pro, OpenAI released the Wensheng video model Sora, which almost stole all the limelight from Google.
This time, Google has launched a full-scale counterattack with the launch of Gemini 1.5 Flash, the latest addition to Google's Gemini family of models, and the fastest Gemini model available through APIs, which is lighter than the 1.5 Pro and has powerful multimodal reasoning capabilities when processing large amounts of information. Google has also significantly improved 1.5 Pro, in addition to upgrading the context window from 1 million tokens to 2 million tokens, as well as improving the model's code generation, logical reasoning and planning, multi-turn dialogue, and audio and image comprehension. Today, Google has integrated the 1.5 Pro into Google products, including the Gemini Advanced and Workspace apps.
As for GPT-4o, Google DeepMind also launched the general AI agent "Astra" for the first time. Judging from the video officially displayed by Google, Astra can receive information through the video screen, identify and respond. However, Google also said, "Although we have made amazing progress in developing AI systems that can understand multimodal information, it is still a very difficult engineering challenge to reduce the response time to a point where it can be conversational." ”
In addition, Google announced a series of generative AI tools related to images, music, and videos, including Imagen 3, an AI music sandbox in collaboration with Youtube and musicians, and Veo, the latest video generation model, which is considered to be benchmarked against OpenAI's similar Wensheng graph model DALL.E3 and Wensheng video model Sora.
In the past year, Google has been repeatedly questioned about AI, especially the large model Gemini 1.0 fraud and rollover incident at the end of last year, which caused the company to fall into a crisis of confidence, but Google's continuous investment in AI has not slowed down. According to the latest quarterly earnings report, Google has seen strong growth in search, YouTube, and cloud businesses, and according to Pichai, the company has a clear path to monetize AI breakthroughs through advertising, cloud services, and subscription plans.
At present, OpenAI and Google's AI war is indistinguishable, and there are different opinions about who brings more surprises, but it is certain that the protagonists of this AI competition are not only OpenAI and Google.
In a few days, Microsoft, which has invested in OpenAI, will hold its annual developer conference Build, and AI will undoubtedly be the main event. In addition, on June 10, Apple will hold its annual Worldwide Developers Conference, and the upgrade of Siri's voice assistant is currently the focus of the outside world. Recently, there have been rumors about Apple joining hands with OpenAI and negotiating with Google, and who will win the cake of the hardware king may reshape the competitive landscape of the entire industry.