Just now, "Father of GPT" Alec Radford was reported to have left OpenAI!
No one expected that, as OpenAI's special event on Christmas Day was nearing its end, such a bombshell would arrive.
What OpenAI lost this time was a veteran core member, the true pioneer of the breakthrough that is ChatGPT, the biggest AI breakthrough to date.
After joining OpenAI in 2016, he not only co-authored GPT-1, GPT-2, CLIP, and Whisper, but also provided support for breakthroughs like GPT-3, GPT-4, and DALL-E 2/3, and is the inventor of multiple U.S. patents owned by OpenAI.
He is highly productive in the field of machine learning and deep learning, with over 190,000 citations, and multiple papers accepted by renowned conferences such as NeurIPS, ICLR, and ICML. His collaborators include well-known scholars such as Ilya and Ian Goodfellow.
According to The Information, on Thursday, Radford announced his resignation and plans to pursue independent research. He reportedly plans to collaborate with both OpenAI and other AI developers going forward.
OpenAI's research director, Mark Chen, stated, "We deeply appreciate and thank Alec for his contributions, and look forward to continuing to collaborate with him on projects as he explores independent research."
The "Godfather of Poker", a senior research scientist at OpenAI, once said, "OpenAI really started to emerge when they hired a then-obscure researcher named Alec Radford."
Some have also asked a thought-provoking question: If OpenAI is truly on the edge of AGI, why are these people still leaving?
The True "Father of ChatGPT"
2016 was a turning point for OpenAI.
The 23-year-old legendary AI researcher Alec Radford chose to join, a decision that not only changed his career, but also opened a new chapter in the development of AI.
Some say Radford's role is akin to Google co-founder Larry Page's invention of PageRank.
He is an introverted and low-profile researcher who has rarely appeared in public and never given any interviews.
Previously, Radford had revealed that his greatest interest was to enable neural networks to engage in genuine dialogue with humans.
After joining OpenAI, his first experiment was to train a predictive model using 2 billion Reddit comments, but it ended in failure.
At the time, President Greg Brockman encouraged him, saying, "No problem, try again."
Due to limited computing power, Radford could only experiment in a single domain. Next, he chose Amazon e-commerce platform reviews, trying to let the language model simply predict and generate the next character of user reviews.
This seemingly ordinary choice led to an unexpected harvest: the model not only could predict the next character of the review, but also could independently judge the sentiment tendency (positive/negative) of the review, and even generate reviews of different styles according to requirements.
Article link: https://arxiv.org/pdf/1704.01444
It was this breakthrough that caught the attention of OpenAI's former Chief Scientist Ilya. He then encouraged Radford to further expand the scope of his research - to try again beyond the Amazon review data.
In 2017, the foundational work of TRON was born, ushering in a qualitative leap in Radford's research.
He collaborated with Ilya to create a model with 117 million parameters, which they named "Generative Pre-trained TRON" (GPT).
This model exhibited unprecedented predictive and understanding capabilities in 7,000 unpublished learning books, Quora Q&A, and various exam articles.
Even more impressive is that GPT also has the ability of zero-shot learning, and can reach professional level in untrained domains.
This unexpected capability excited the team and prompted them to deeply reflect on the direction of AI.
Interestingly, when the original GPT was released, Radford and the team chose a low-key approach, publishing the paper directly on arXiv instead of submitting it to major conferences.
a16z's famous venture capitalist Marc Andreessen once described, "Between 2018 and 2019, he was hiding in the corner of artificial intelligence, quietly researching in his own field."
It is this persistence that ultimately changed the trajectory of artificial intelligence development.
This is the moment when the legendary GPT series began.
OpenAI is Losing Core Employees
Alec Radford's departure is like a bombshell, shocking netizens.
Some say the impact of this veteran on OpenAI and the entire AI research has been underestimated. It can even be said that his influence is no less than Ilya's.
Just a few days ago, in Ilya's NeruIPS 2024 speech, he specifically mentioned thanking his former colleague Alec Radford, because he is not only the main author of GPT, but also likely the leader in every important research direction of the company.
UC Berkeley AI PhD Dan Hendrycks stated that in his ranking of academic influence in AI, He Kai-ming and Alec Radford are undoubtedly TOP 2.
CS professor and senior advisor to DeepMind, Jeff Clune, stated that we should rehabilitate Alec Radford.
He is the most important but least known person in the field of AI research, an underestimated person who has gone into hiding after major breakthroughs.
Professor Clune believes that Alec Radford should be recorded in history as the father of modern generative AI.
Many scientists and researchers agree: yes, Alec Radford is the true MVP.
Fortunately, although his departure is a loss for OpenAI, he will certainly not lack GPUs in the future, and if he does some open research, the whole world will benefit.
Some also speculate that whenever OpenAI releases a major model, someone will leave. For example, after the release of the advanced language model Her, CTO Mira Murati left.
According to this rule, perhaps GPT-4.5 will be released tomorrow. (There have been many similar predictions about the release time of GPT-4.5 before.)
Many people understand Alec Radford's departure.
Once a startup goes through its initial rapid growth phase and stabilizes, it is normal to choose to leave the company.
They have connections and experience, and can find the job they want anywhere. They can even start their own companies and get billions of dollars in financing.
Ironically, a year ago, some netizens said that the split of TRON would not have much impact on the trajectory of AI's future, because the real geniuses behind the achievements are scientists like Ilya and Alec Radford.
"As long as they are still there, TRON will not be affected at all."
But now, the foundation of TRON is shaking.
In summary, TRON has lost many veterans in 2024.
Including Chief Technology Officer Mira Murati, Head of Research Bob McGrew, Co-founder and Researcher John Schulman, Vice President of Research Barret Zoph, Head of Safety Systems Team Lilian Weng, Co-founder and Former Chief Scientist Ilya Sutskever, Former Head of Safety Jan Leike, and now Aled Rad.
It can be foreseen that the list of resignations will continue to grow.
References:
https://www.wired.com/story/what-openai-really-wants/
https://www.theinformation.com/briefings/senior-openai-researcher-radford-departs
This article is from the WeChat public account "New Intelligence", author: New Intelligence, editor: Editorial Department JHYZ, authorized by 36Kr for release.
