How are academic researchers navigating the tech job market? Premium

Job hunting in the tech industry for PhD students has been a tedious process for years.

On July 25, a seemingly harmless job posting for a research position at NVIDIA lit up microblogging platform X. The position required applicants to have published at least eight research papers in prominent conferences like the International Conference on Machine Learning (ICML), Computer Vision and Pattern Recognition (CVPR), aside from knowing machine learning frameworks like PyTorch and Tensorflow.

How the graphic card-making company arrived at that particular number baffled several academic researchers. The general disapproval from the community forced NVIDIA to eventually apologise calling the bullet point an “error,” and clarifying that they did not want to sacrifice quality for quantity. The incident is but one example of how confusing things can appear for academic researchers looking for a job. 

“Advanced research labs typically have high standards for hiring candidates with research capabilities,” said Ankit Rai, an AI specialist working at an advanced computing lab. “Publications ensure that the candidate has a strong background and is capable of conducting high quality research. However, these high publication requirements may also at times overlook other skill sets like practical experience, teamwork, and set unrealistic expectations at times.”

Arvind Narayanan, professor of computer science at Princeton said, “The NVIDIA requirement of eight publications is obviously silly, especially for a non-academic job. The correlation between paper publishing and real-world skills is minimal.” 

(For top technology news of the day, subscribe to our tech newsletter Today’s Cache)

For those outside of academia, it is hard to understand the rigmarole of having a paper published in a globally recognised conference.

“Submitting your paper with an important conference is a lengthy, tiring process - most deep learning conferences like CVPR have around 10,000 submissions,” explained Abhinav Upadhyay, Technology leader at Accenture Labs.  “A researcher works on a paper for between 10-12 months, and it takes 3 months to review one. So, no, putting down a number as the minimum requirement is not the right way to look at things.”