This Google research scientist died unexpectedly, leaving a last note two months ago: Working on large model research made me deeply depressed

“Goodbye, Felix!”

This Thursday, the artificial intelligence community said goodbye to a scientist in its own way.
Felix Hill, a research scientist at Google DeepMind, died on December 5, 2024. He had been battling severe mental illness since early 2023.
Before his death, he had worked at Google DeepMind for nearly 9 years.

“It still hurts to think about it now,” said Kyunghyun Cho, a professor of computer science and data science at New York University and co-founder and senior director of the biotech company Prescient Design. He wrote an article to commemorate his acquaintance and research partnership with Felix.

It was 2014, Kyunghyun Cho was a postdoctoral fellow in Montreal, with Yoshua Bengio as his advisor, and Felix was a visiting scholar who had just arrived in Montreal. At that time, Kyunghyun Cho was exploring a neural machine translation system that could handle long texts. When Felix stopped by Kyunghyun Cho’s desk and introduced himself, he left a comment: grammar is not a problem.

This sentence became one of the regular guests on Kyunghyun Cho’s slides in the following years. Now in the era of large language models, we have understood what it means.
Felix and Kyunghyun Cho subsequently collaborated on a series of studies.
It is worth mentioning that in October 2024, Felix left a blog post.
Nvidia research scientist Jim Fan said: This is the most heartbreaking blog I have ever read because it is so real and so close to my heart.