AI on academic research
AI usage on academic research is one of the best reference for AI application.
Terence Tao
The earliest application of AI on academic research is from Terence Tao, who use and share the usage of cutting-edge AI products at early stages of AI development. I kept adobt at that time because AI is till a little bit dump sometimes, Academic research is the hardest and most sacred area. But the following application proved that AI is very useful on academic research. Terence use AI to solve many the most complex issue in mathmatics fields
Tao co-found sair with other professors to scale scientific discovery with AI, and ground AI in science at 2026.
Autoresearch
Another example is autoresearch from karpathy, this project use AI to optimized the parameter, code, even models to automate the iteration process. It’s not a perfect project because it hiden some details in iteration process, some of them might be buggy or be blackbox. But nobody can totally explain how LLM works, it’s still blackbox because it has too many parameters, but it eventuall works. This is a good evidence to provde that autoresearch might works in similar way with LLM.
If we have to talk about white box, we are still able to check the iteration logs to know what happened, to revised it if something wrong. This example shows a big progress of agentic ability of AI.
My research
I was using codex for my research paper as well in the past two months. It can help for brainstorming, investiagte new ideas, coding, validate ideas, write reports. AI cannot make decision for me, I’m the one to point out the direction of next step, I enjoyed the process because it’s smooth. I spent most of the time on thinking, I don’t have to do other works, it’s enjoyable. Of course, the premise is AI can follow my instruction correctly, for example, I found gemini flash 3.0 is much worse than gpt 5.4, gemini 3.1 pro is still a little bit worse.
Summary
AI application on academic research is a evidence that it’s worthwhile to try in any field.
What human should do is thing above the line of AI capability.