2023 has just passed, experiencing some reflection, anxiety, and relief.
Academic Ramblings#
At the end of 2022, the main academic work I was involved in encountered a reviewer with a poor professional attitude, which resulted in a waste of energy and time for myself and my collaborators. This led to a continuous self-doubt and sense of emptiness. Although I had experienced similar situations before, the extent of this one far exceeded expectations: a colleague procrastinated for nearly a year before giving the evaluation, "This work on 'adjective A + noun B' does not fit the scope of the 'Journal of noun B'."
So, in the second year after the initial draft of this work was completed, we made modifications and resubmitted it, despite the explosive development of related technologies in the field. Finally, at the end of 2023, it was accepted. I would like to express special thanks to my mentor and collaborators for their affirmation and support in this work.
Although several other leading and collaborative works did not go as smoothly, they did not exceed expectations. Perhaps it is the fate of researchers to seek a balance between self-confidence and self-doubt on the slope of life. The doctoral stage is likely the last phase of a career where one can fully focus on a single research topic. I hope to enhance my ability to manage multiple parallel projects in the future.
Idle Thoughts#
Time has suddenly fast-forwarded to the potential end of my academic career (perhaps), and I have started receiving frequent invitations from headhunters. Based on my observations, the most urgently sought-after positions this year are in multimodal large models and quantitative trading.
Based on my judgment of the maturity of the domestic financial market, I did not delve deeper into the field of quantitative trading. However, the work on multimodal large models keeps coming at me: from BERT to LLAMA, from ViT to Diffusion, from MoE to RLHF. As someone who does research on fundamental machine learning theory, it feels strange to be asked if I have experience with large models. After all, my work experience is clearly stated, some basic ML theory, some AI+Web3 PoCs, but no work experience at OpenAI or MetaAI (laughs). Perhaps this is the case of "a mountain between professions," where we tend to consider our knowledge in our own field as common sense, while treating some truly essential life knowledge as ancient whispers.
At this year's conference, I saw a lot of work on prompts or tuning for LLMs, which always made me feel like a craftsman facing optimization problems with a weaving machine. The acquisition, retrieval, and distribution of information have always been the core functions of the Internet. Are our technological accumulations, governance rules, and moral standards truly ready to embrace this great transformation? I can't say for sure, but this question will certainly persist for a long time. After all, even the smartest people in the academic world like Ilya Bengio are just beginning to explore.
Ramblings on Leisure#
In 2022, I finished reading about fifty books on WeChat Reading. I haven't kept track of the statistics this year, but I definitely read fewer books. However, I read more articles, blogs, and physical books outside of my research field, so the overall reading volume is not much different.
Because of my work on recommendation systems, I continued to adhere to the principle of "deciding information acquisition for myself," spending little time on the recommended information flow provided by software platforms and not downloading short video apps. Our precious time and attention should not be consumed as commodities, as much as possible.
Machines are learning, humans are hooked. This is not the future that we, as machine learning researchers, strive for, but a dystopian cyberpunk story. This year, I also played the DLC of Cyberpunk 2077, sending the Night City bird to the moon, just like V and Rebecca. The stories of EVA, Ghost in the Shell, 2077, and BioShock intertwined in that moment, and the moon became a faint light called hope.
And then, at the end of 2023, I arrived on the moon (To be continued).