Reading papers with AI: skip the summary, use it as a translator instead
Keeping up with technical papers in a fast-moving field became impossible because more papers appear each week than anyone can read. The obvious fix, having AI summarize them, turned out to be a trap: summaries create the illusion of having read something, and you find out you understood nothing at the worst possible moment, when someone asks a question. What actually works, though it takes longer, is reading the paper yourself, properly and slowly, and using AI only as a translator for the parts that don't make sense.
Useful include asking it to explain a section when the notation is unfamiliar, asking what assumption underlies a step and why it's reasonable, and asking what would break if that assumption were wrong. The technique that makes understanding stick the most is explaining the paper back in your own words, badly, and asking AI to find the errors in that explanation. A rough self-explanation is far more diagnostic than any summary because it reveals exactly which parts of your understanding are shaky.
Finally, asking what a skeptical reviewer in the field would attack first adds another layer of scrutiny.
Key points
- Summarizing papers with AI creates a false sense of understanding
- Use AI as a translator only for parts you get stuck on
- Explaining the paper back in your own words and having AI find errors is the most effective check
- Asking what a skeptical reviewer would attack first adds extra rigor
- The approach is slower but builds real comprehension