Using an LLM as a Thought Partner

In PhDfyi, we are going to use LLM’s to have constructive conversation with, and to help us build, conversationally. LLM’s will act as our personal thought partner, who never sleeps or gets bored. Everyone’s career is unique and the best way to have wide open discussions with no right answer is with a robot thought partner who has digested every corner of the internet.

Each section will have prompts specific to the PhDfyi framework, but here are some general ways to use LLM’s in your routine.

Key Concepts for Using an LLM in Your PhD (stay within all school and journal limitations, of course).

1. Using Projects and Knowledge Bases

Projects (also called Spaces and other similar terms) are organizational tools on all nearly LLM websites that let the user keep related chats together, and give that chat group specific instructions. It’s basically a folder with its own rules.

So if you have a Project only for math problems, you could go to that folder any time you would like to discuss a math problem, and all of those chats could be given a specific instruction like (do not give me the answer until I have tried at least 10 times). That rule would apply to every chat within the Math Project, but not to outside chats.

Additionally, there are usually ‘Knowledge Bases’ within those Projects/Spaces (or similarly termed concepts), that let you upload files that only apply to that folder. It is a set of documents that you would like the robot to keep in mind as it works in that ‘folder with its own rules’. To continue the example, it could be the answer sheet to the math problems, or a specific math textbook pdf.

Note: Please abide by all university policies and common sense when using LLM’s

• Splitting Research into Phases
• Create separate Projects for distinct research phases (surveys, data figures, experiment sets, software programs, literature review, methodology design, data analysis, writing, etc)
• Upload core materials for each phase to create contextual awareness (usually pdfs, images, small Excel files, and text).
• If your LLM supports it, reference materials across Projects when needed for synthesis
• Personal Context Building
• Upload your non-confidential research proposal, study plans, and progress reports (Claude is especially good for this, as the model is not trained on your data. Your school may also have a custom LLM for this purpose.)
• Include relevant background documents about your academic journey
• Add notes about your working style, strengths, and challenges
• Share career aspirations and values to enable personalized guidance

2. Technical Optimization Strategies

• Document Management Hacks
• Use Mac's "Reduce File Size" Quartz filter to compress PDFs before upload, as Knowledge Bases usually have file limits, and this keeps pdfs smaller
• Split larger documents into logical chunks with clear naming conventions
• Prioritize text-searchable PDFs over scanned documents
• Include a "Document Map" file that explains what each document contains and tell your LLM to start here, instead of searching through every document
• Maximum Knowledge Transfer
• Upload methodological guidelines specific to your field
• Include exemplar papers that represent your target publication style
• Add critical feedback you've received on previous work

3. Effective Interaction Patterns

• Prompting Best Practices
• Be explicit about whether you want brainstorming or evaluation
• Specify your current emotional state when seeking feedback
• Clarify if you need field-specific or interdisciplinary perspectives
• Request specific output formats (bullet points vs. paragraphs vs. tables)
• An LLM's Strengths for PhD Work
• Comparing and contrasting arguments across multiple papers
• Generating potential research questions from literature gaps
• Translating technical concepts for different audiences
• Summarizing complex arguments without losing nuance
• Creating structured outlines and frameworks
• Providing alternative perspectives on your research approach
• An LLM's Limitations to Work Around
• May not fully understand very recent or cutting-edge methodologies
• Cannot perform novel scientific calculations or statistical analyses
• Limited ability to judge the scientific significance of very new findings
• Not a replacement for advisor relationships or peer review
• Cannot make ethical research decisions on your behalf

4. Academic Workflow Integration

• Literature Review Enhancement
• Upload multiple papers for thematic analysis
• Ask the LLM to identify conceptual relationships across papers
• Use the LLM to generate synthesis tables of methodological approaches and findings
• Ask for gaps or contradictions in the literature group you've uploaded
• Writing Process Support
• Share drafts for structure and clarity feedback
• Request alternative phrasings for complex concepts
• Use the LLM to help maintain consistent terminology
• Ask for suggestions on strengthening arguments
• Interdisciplinary Translation
• Get help explaining your research to different audiences
• Practice explaining concepts at different levels of complexity
• Generate analogies for technical concepts