Beyond Generative AI

AI Workflow for Research

ChatGPT, Gemini, Perplexity, Claude, or any other generative AI have introduced pros and cons toward researcher communities. Quoting a line from Jensen Huang (CEO of NVidia) "You're not going to lose your job to an AI, but you're going to lose your job to someone who uses AI". It also resemble for scientist, especially for PhD student an entry step of scientific career. 

Everyone agree we need to put some ethics and wise way to use AI, but how to make it more efficient and make it into seemless workflow. To maximize the efficiency as a PhD researcher, we can treat these AI tools as specialized team members. Each tool excels at a different stage of the scientific process.

1. Step 1: Discovery and Grounding
   We need to know first the starting point of our research from this step. Conventionally, researcher will use review paper to know current state of our research field. But, today with the quick pace of publication, sometimes I feel very exhausted just to find which paper is most related to my research. So, Leapspace and Perplexity are the best choice for this reason. On this step the goal is getting knowledge baseline and some paper as main references

2. Step 2: Personal Knowledge Management
   After get some idea and take some papers as main references, well managed on reference manager, we need to test our understanding and challange the idea of all those references. NotebookLM can act as private "brain" based only on our documents. Ask it to find contradictions between different authors' methodologies. This prevents hallucinations.

3. Step 3: Synthesis and Deep Reasoning
   Now we need to connect the dots. Claude can be used to complex document analysis. It has a high reasoning capability. It is excellent for synthesizing long methodology sections into a single comparison table. Then, Gemini can be used for multi-step investigations. It can perform extensive web searches to find specific material properties or patent filings. It builds comprehensive reports with cited reasoning.

4. Step 4: Drafting and Refinement
   Next after everything has gathered, concluded and ready we need to publish it before anyone else get the same findings. We need to change it into manuscript. As we start from outline or rough notes, chatGPT can help to structurize it and elaborate our idea. Then Claude can act as final polish. It excels at maintaining a professional, academic tone. It ensures your logic remains consistent throughout the paper.

Anchoring Our Research in Truth

AI tools process data with incredible speed. However, they lack the intuition of a doctoral researcher. Hallucination occurs when an AI generates plausible but false information. We must act as the "Anchor" of our own research. This ensures our work remains scientifically sound and ethically responsible.

The Human-in-the-Loop (HITL) Strategy

Human-in-the-loop is a systematic approach to research. We use AI for repetitive tasks. We maintain control over critical decision points. AI excels at processing large datasets. We excel at interpreting complex, field-specific context. 

Practical Habits for the Researcher

• Cross-Check with Authoritative Databases: Use Scopus or Google Scholar to manually find the papers AI suggests. Never cite a paper you have not opened yourself.

• Maintain a Process Trail: Keep your early outlines and messy notes. This "version history" proves your intellectual contribution. It protects you from accusations of AI-only writing.

• Use Domain Logic: AI often misses subtle nuances in materials science. It might suggest a polymer that is incompatible with your specific battery electrolyte. Apply your expertise to filter out these "chemical hallucinations."

• Prompt for Uncertainty: Tell the AI: "If you are unsure or lack data, say 'I don't know'." This reduces the chance of the model fabricating an answer.