Dialogue with DeepSeek "Challenger" LI Bojie: This Interview Is Like a Rashomon

A viral complaint post has thrust DeepSeek, a leading domestic AI large model enterprise, into the eye of public controversy.

On July 6, a job seeker posted on social media publicly criticizing DeepSeek's interview process, describing it as slow and cumbersome, and claiming that the interviewer had a disrespectful attitude and accused him of plagiarism. "I felt deeply offended," the poster wrote.

The author of the post is Li Bojie, born in 1992, who holds a Ph.D. in Computer Science from the University of Science and Technology of China and was one of the first selected members of Huawei's "Genius Young" program. According to public information, he has published multiple papers at top-tier conferences including SIGCOMM, SOSP, NSDI, ATC, and PLDI, and has received the ACM China Outstanding Doctoral Dissertation Award and the "Microsoft Scholar" Scholarship.

Currently, Li Bojie still serves as the Chief Scientist of Pine AI.

1

Li Bojie "Pleads His Innocence": It is Completely Normal for Technical Staff to Use Two Screens During Remote Interviews

On July 8, Li Bojie shared more details about this interview with Time Finance. According to his account, the original intention of participating in the interview was to join a cutting-edge platform to expand his expertise in the field of foundational models.

Since officially joining Pine AI in February 2025, Pine's products have experienced rapid growth, with the growth rate in the past two months matching the total growth of the entire last year. As the growth accelerated, Pine AI needed to develop its own models — an area that Li Bojie was not fully proficient in. "Last year we developed some models, but to scale them up further, there are certain aspects that I do not master," he explained.

Li Bojie stated that, as Chief Scientist, he realized he had not fully grasped the evolving trends of large models, and his understanding of some engineering details in model training was not sufficiently in-depth. Against this backdrop, he turned his attention to leading domestic large model enterprises, aiming to acquire more knowledge and continue learning on a more cutting-edge platform.

DeepSeek was one of the targets he chose. "It is arguably the best large model company in China," he said.

Li Bojie mentioned that he considers DeepSeek a remarkable enterprise: the company has developed numerous industry-leading achievements, and has open-sourced many outcomes that should have been kept confidential for commercial profitability, contributing them to the public. "We believe this is a company with deep foundational aspirations and lofty pursuits."

However, the high expectations he held before the interview turned into a sense of disappointment during the process.

The first round of the interview was scheduled in early June.

Li Bojie recalled that about a week after submitting his resume, DeepSeek arranged a written test consisting of standard programming questions, as well as single-choice and multiple-choice questions. He commented: "The coding part in the first round was pretty good. The interviewer was extremely competent — he spotted an error in my code at a glance, which took me a long time to figure out myself."

After the first round, DeepSeek delayed arranging the second interview for a long time. During this period, interviews Li Bojie attended at other companies had already progressed to the offer stage. To avoid missing the opportunity, he followed up with DeepSeek several times, only to be told that "there are too many ongoing interviews, please wait a little longer."

About half a month later, Li Bojie participated in the second round of interview remotely from Beijing. The interviewer failed to log on as scheduled, and subsequent communication broke down due to mismatched expectations for the position — Li Bojie's background is focused on academic research, but DeepSeek assigned him an AI engineer role.

This misalignment further led to the situation where, when Li Bojie discussed cutting-edge AI topics, the interviewer accused him of "digressing," insisting that the focus should be on engineering challenges in the AI field. This made Li Bojie feel that "he showed little respect for my academic research work."

What made Li Bojie feel most offended was the coding segment in the second interview. During this part, Li Bojie used two monitors: one displayed the Tencent Meeting interface showing the interviewer's and his own video feeds, while the other was a shared code editor (Vim) interface.

Since the interviewer's video feed was on the left screen, Li Bojie's gaze occasionally drifted there involuntarily — a behavior the interviewer interpreted as "copying code." Throughout the process, the interviewer repeatedly warned him, saying "I see you've been copying code, and if that continues, the interview cannot proceed."

Li Bojie was furious at the time, but after calming down, he felt the situation was a classic "Rashomon" scenario: the interviewer had no evidence to prove he was plagiarizing, and he also had no way to prove his innocence. He noted that technical personnel often use multiple screens, with many people commonly using one screen for coding and another for Vim, or separating the console and coding outputs.

After the incident, he gave feedback to the hiring HR, stating that he would no longer consider opportunities at DeepSeek.

2

Candidates with Combined "Research + Engineering + Delivery" Capabilities Are Extremely Scarce

So, was this interview a microcosm of arrogance from a large tech firm, or simply a procedural mishap?

Time Finance attempted to contact DeepSeek by phone and send emails to verify the details described by Li Bojie, but no response had been received as of press time.

A software engineer working at a technology company told Time Finance in an interview that during interviews, interviewers typically prioritize two core aspects: first, technical competence, which is a fundamental requirement that varies by position level to ensure the candidate can help the company deliver usable products; second, the candidate's initiative in working on projects and ability to work under pressure — traits that interviewers evaluate using their own sets of criteria.

A human resources industry practitioner commented that this "farce" is more likely the result of information asymmetry and insufficient communication between the two parties; as for the plagiarism accusations during the interview, this depends heavily on the individual style of different companies' interviewers or HR teams.

Multiple HR professionals also reported that the current recruitment landscape in the AI industry presents a paradox: there is an enormous volume of resumes, with some popular positions receiving tens of thousands of applications, but candidates who truly possess the combined "research + engineering + delivery" capabilities are extremely scarce. Overall, the supply-demand ratio for positions in underlying computing power, AI infrastructure, and Agent engineering is severely unbalanced.

It is reported that the mainstream interview process at current AI enterprises typically consists of four stages: first, the initial screening phase, which requires passing an online written test or coding assessment (covering mid-to-high difficulty LeetCode problems and fundamental ML/AI questions); second, the technical evaluation phase, usually 2 to 4 rounds including on-site coding, ML system design, and in-depth technical discussions on Agent/RAG/inference optimization; then the project and behavioral interview, which focuses on evaluating the candidate's past project experience, engineering trade-off decision-making, failure handling capabilities, and team collaboration skills; finally, the final interview and cultural fit assessment, which emphasizes alignment of values and long-term potential.

Time Finance has noticed that many leading industry enterprises have expanded their recruitment plans recently.

For example, in March this year, market sources revealed that to compete with companies like Anthropic PBC and Google under Alphabet, OpenAI plans to expand its workforce from approximately 4,500 to around 8,000 by the end of 2026, with new hires primarily placed in product development, engineering, research, and sales roles.

In late June, DeepSeek launched its largest public recruitment drive since its founding. It posted recruitment notices on its official WeChat account, stating that "as technology evolves, we are working to expand the size of every department by at least double," with open positions covering large model pre-training, post-training, Agent, AI search, inference architecture, AI platforms, data products, AI product managers, product operations, as well as HR, legal affairs, and procurement.

According to data from BOSS Zhipin, as of July 8, DeepSeek had a total of 143 open positions, with technical roles accounting for the highest proportion, most offering 14 to 16 months of annual salary. Among these, the salary range for the Multimodal Understanding (Data/Algorithm) Researcher position based in Beijing is 70K to 100K RMB per month.

Talent is rapidly becoming a critical, scarce "resource" for the development of the AI industry.

This article is from the WeChat Official Account "Time Finance APP" (ID: tf-app), written by Wu Dian and Xin Linlin, edited by Bai Jinlei, and republished with authorization from 36Kr.