A few days ago, OpenAI announced an organizational restructuring and split into a for-profit company and a non-profit organization. At the same time, OpenAI CEO Sam Altman thanked many of OpenAI’s veterans in a series of tweets on X.

Later, he sent a separate tweet to express special thanks to an "Einstein-level genius": Alec Radford. Altman also said Radford is his all-time favorite, noting that many of today's advances in AI can be traced to his work. On the 20th of last month, news broke that Radford was about to leave OpenAI to conduct independent research. For details, please refer to the report of the Heart of the Machine. "Just now, OpenAI veteran researcher Alec Radford resigned. He led the research and development of GPT-1 and GPT-2. 》.

According to Google Scholar, Alec Radford’s paper citations now exceed 190,000, and he has several important papers with more than 10,000 citations.

In fact, more than a year ago, someone had already pointed out the huge contribution of this then-unknown researcher: he was the first to publish many foundational research results, including the original GPT paper. One author! What’s even more surprising is that he does not have a PhD or a master’s degree; and many of his research results were first completed in Jupyter Notebook.

At the beginning of the new year, Alec Radford’s story once again became popular on AI social networks. People spoke of him with praise.

Let’s take a brief look at this humble man The life story of a genius.

Alec Radford is a natural language processor and researchers in the field of computer vision. Worked as a machine learning developer/researcher at OpenAI. Prior to that, he served as Director of Research at Indico Corporation.

While working as a researcher at OpenAI, he co-authored a number of papers on language model generative pre-training (GPT), and published many papers in famous conferences and journals such as NeurIPS, ICLR, ICML and Nature. paper.

He was also active on X/Twitter, where he shared his thoughts on various aspects of artificial intelligence.insights and perspectives in the field. Unfortunately, he has not appeared on X since May 2021. His last tweet was in reply to Anthropic researcher Simon Kornblith, explaining why the GPT-1 layer width is set to 768.

LinkedIn shows that from 2011 to 2016, Alec Radford studied at the Franklin W. Olin College of Engineering and received a bachelor's degree. This is a private engineering college located in Needham, Massachusetts, USA. It is known for its very low admission rate and elite education.

The school's curriculum structure is very unique. Its academic system "Olin Triangle" includes basic science and engineering, entrepreneurship, and literature. The school also only offers four degrees: ME (Mechanical Engineering), ECE (Electrical and Computer Engineering), CS (Computer Science), and BCE (Biomedical Engineering).

Olin Engineering attaches great importance to practical education, connecting knowledge with real-life challenges and topics, and encouraging students to pursue their interests and passions.

As an undergraduate, Alec Radford was passionate about machine learning, and successfully participated in the Kaggle competition with his classmates, and eventually received venture capital. In 2013, Alec Radford and his partners founded indico in their dormitory to provide machine learning solutions to enterprises.

As the company’s Director of Research, Alec Radford focuses on identifying, developing and improving promising image and text machine learning techniques and moving them from the research stage into broader industry applications.

There, he conducted research on Generative Adversarial Networks (GAN) and proposed DCGAN to improve the trainability of GAN, which was also an important breakthrough in the field of GAN.

Because the Boston area’s influence in the field of AI has been surpassed by West Coast technology giants, coupled with resource constraints, Radford chose to join OpenAI in 2016.

He described the new job as "similar to joining a graduate program," with an open, low-pressure AI research environment.

Radford is relatively low-key and reluctant to interact with the media. He responded via email to WIRED’s questions about his early work at OpenAI. He says he's most interested in getting neural networks to have clear conversations with humans.

He believed that chatbots at the time, from ELIZA to Siri and Alexa, had limitations. His goal is to explore the possible applications of language models in a variety of tasks, settings, domains, and scenarios.

His first experiment was to use 2 billion Reddit comments to train a language model. Although it failed, OpenAI gives enough room for trial and error, and it also became the beginning of a series of subsequent revolutionary breakthroughs, such as the well-known first-generation GPT, which also led the research and development of GPT-2.

These works laid an important foundation for modern large language models, so much so that "Wired" compared Alec Radford's role in OpenAI to Larry Page's invention of PageRank in a report. Interestingly, although PageRank was the product of Larry Page's Ph.D. studies at Stanford, he never went back to complete his Ph.D.

Not only that, Alec Radford also participated in the writing of the GPT-3 paper and the research on pre-training data and architecture of GPT-4.

At the end of 2024, before the last day of OpenAI’s eye-catching 12-day streak, news broke that Alec Radford was about to leave OpenAI, but there was no news whether this was related to OpenAI’s organizational structure adjustment. .

Unfortunately, all we know about his subsequent career plans is that he will become an independent researcher. Maybe he will enter university and get a PhD, or maybe he will come out with new shocking research results after a period of silence.

Either way, the future that Alec Radford helped create is coming. Regardless of whether AGI will be realized this year as Ultraman once predicted, 2025 is destined to be a crucial year in the field of AI.