Language models can speed up research and analysis, provided you manage their tendency to invent specifics.
Good Uses
- Explaining unfamiliar concepts and terminology.
- Structuring a problem: frameworks, questions to ask, factors to consider.
- Analysing documents and data you provide.
- Comparing options against criteria you define.
- Drafting research plans and interview questions.
Risky Uses
- Specific statistics, dates and quotes from memory.
- Citations and references, which may not exist.
- Recent events after the model's training data.
Ground Research in Sources
Provide the documents, or use tools that search and cite sources, and ask the model to cite what it relies on. Verify citations exist and say what's claimed.
Ask for Structure
"Break this question into sub-questions", "list the assumptions behind this conclusion", "what evidence would change this view?"
Challenge the Analysis
Ask for counterarguments, weaknesses and alternative explanations. Models tend to agree with the framing they're given; explicitly request disagreement.
Separate Facts From Interpretation
Ask the model to label what comes from sources versus its own inference.
Verify Before Relying
Check key facts against primary sources before using them in decisions or publications.