Last week, i posted a note on the paradigm shift in white collar work. i have been exploring the new LLM-based tools such as ChatGPT, especially the deep research option, to improve the quality of my analysis and productivity.
Let me explain how –
I search for new ideas using various screens, charts and through general reading. I then review the charts and read a few annual reports and conference call transcripts. At this stage I have a rough idea about the company and know enough to ask the right questions.
Research plan and Autonomous agents
This is a good point for me to create a research plan. This plan has the usual elements of company details, its industry, competitors, management and so on. The nuance is adding company specific questions which are relevant to the idea. For example ā when analyzing a bank i would like to compare it with other banks various metrics such provisions, NPA trends, Loan books etc
I feed this research plan into multiple LLM tools ā Chatgpt, Grok and Gemini etc . I can add all the annual reports, conference call ppt and transcripts to the chat too. The deep research tool uses these uploaded documents as the primary source to generate a detailed report. The beauty of these tools is that if it cannot find the answer, it does extensive search on the web and provides those details with references
The latest reasoning models can understand your questions and reason through the best approach on answering them. It can also figure out which tools to use such as Search, code interpreter and so on. In other words, it is acting as an autonomous agent
The result is often a 30ā40-page detailed report tailored to my questions. This report then sparks more questions which i can google or ask the LLM to find answers
In summary, these tools are like 24/7 analysts, improving at an exponential rate. The analyst is not smart enough to ask the right questions or decide on which companies to research. That is my job.
Are these tools perfect? Of course not. They often get the numbers wrong but by knowing enough beforehand, i find the errors and correct them. As these tools evolve, i expect to use them in more creative ways
