Wen 丨Congerry
Those who learn from me live, and those who learn from me die.
OpenAI CEO Sam Altman said bluntly at the YC alumni sharing meeting a few days ago: "Shell" OpenAI is doomed!
At the end of last year, the birth of ChatGPT directly set off a new round of AI wave, only two months after its launch, the number of users has exceeded 100 million, and OpenAI as a developer directly exploded around the world.
Sam Altman revealed that ChatGPT was not the direction of OpenAI's "build" from the beginning. The only one undergraduate named Alec Radford who has always adhered to the research line of the big language model has been insisting on it for 7 years.
In addition to these, Sam Altman also revealed the development direction of the next generation model at the sharing session, as well as shared his entrepreneurial experience and suggestions.
Shell OpenAI, you won't end well
Shelling and seams are not the right way.
For companies that simply apply OpenAI models to innovate, Sam Altman offers a clear warning that they have limited space to survive.
Sam Altman cites UI design as an example, if a company only slightly better than ChatGPT in interface design, it is definitely not a good idea. Because OpenAI itself is already continuously improving its small shortcomings in these areas, it is difficult for companies that only put effort into the interface to form a sustainable competitive advantage.
Sam Altman believes that the real valuable innovation in the future should be breakthroughs in technology and models, not simply fixing some small problems with GPT, nor just optimizing the user interface.
For the research direction of AI, Sam Altman also gave his own opinion, AI applications can play a greater role in the medical and education fields.
Specifically, Sam Altman believes that good AI medical assistants can bring great benefits to society.
Previously, Google's blockbuster paper on the medical big model Med-PaLM appeared in the journal Nature, and the model's performance on MultiMedQA reached 92.6% accuracy, exceeding the average human clinician, demonstrating the great potential of large language models in the medical field.
In addition, Altman is also optimistic about the personalized application of AI in the field of education, providing AI tutors with customized teaching and tutoring for each student. This highly personalized AI coach can improve the quality of education on a large scale.
GPT-5 and GPT-6 will have multimodal output capability
At the sharing session, Sam Altman also revealed the news of GPT-5.
Sam Altman said the GPT-5 and GPT-6 would be more reliable and enable better personalization.
According to OpenAI, the basic information of GPT-5 was previously mentioned in the trademark application.
The first function of GPT-5 is the enhancement of multimodal functions, which can process text, images, voice, video and other types of data for translation, recognition, generation and other tasks.
Second, GPT-5 may also have the ability to actively learn, and can select, acquire and process data autonomously according to its own goals and needs, rather than relying solely on data provided by humans.
But even so, there is still some distance from true artificial general intelligence (AGI).
Altman predicts that the future direction of GPT is to reduce costs and improve reliability. The current GPT model has shown scalability, similar to Moore's Law for chip development in the past. The growth of model size and computing power will continue to reduce the API cost of GPT.
Overall, GPT-5 and GPT-6 represent a new stage in the development of large-scale pre-trained language models with stronger performance, but are still some way from true AGI. Over time, the cost of computing power will make these models more widely available.
Sam Altman's entrepreneurial experience
The success of ChatGPT not only set off a wave of AI, but also triggered a wave of crazy startups.
Sam Altman offered a few of his thoughts on startups.
Only by thinking long-term can we go further. OpenAI has gone through many twists. From robotic weapons to nuclear research to gaming. Over the years, it became clear that the problem to be solved was AGI's comprehensive thinking problem, not how to make tendons for the robot.
Repeated failures. Sam Altman's first startup failed, and as a partner at YC, he observed hundreds of startups to get a first-hand view of failure.
Startups experience many near-death experiences; While eventually they will figure it out, it's important to build failed muscles over time.
Recruit, spend time with people who are impossible to join. Sam Altman and leadership didn't clearly realize that this was the right path when hiring AI researchers.
Over time, one researcher made a breakthrough in large language models, and the other teams humbly decided to close their projects and focus their efforts here. This ultimately led to ChatGPT.
That man was Alec Radford.
In 2016, Alec Radford graduated from the Franklin Olin School of Engineering with a bachelor's degree in engineering. He then joined OpenAI, where he continued to study large language models for 7 years, contributing greatly to the development of GPT series models.
Looking back at the development path of OpenAI, the rich founding capital, great vision leadership, visionary and determined innovative ideas have created its success.
If you have anything to say, welcome to leave a message in the comment area in front of the screen to discuss! We will give unlimited red envelopes to students who like, comment and follow~