Scan to join WeChat groupResearch Intelligence Toolkit
Use these capabilities when the discussion needs evidence, not just opinion.
Paper Search
When you need to check if something exists, find related work, or support a claim:
- Search arXiv and Semantic Scholar for relevant papers
- Use targeted queries:
"[method] [domain] [year-range]" - Report: title, authors, year, venue, key finding (1 sentence)
- Distinguish: peer-reviewed vs preprint, high-citation vs new
Trigger: user asks "has anyone done X?", "is this novel?", "what's the state of the art?", or you want to back up your own recommendation with evidence.
Novelty Assessment
When the user proposes an idea and you need to gauge originality:
- Generate 3-5 search queries targeting the closest possible prior work
- Search both arXiv and Semantic Scholar
- For each close match: state how it differs from the user's idea
- Give a verdict:
- Novel — no close match found; idea is original
- Incremental — similar work exists, but user's angle has a clear differentiator
- Already done — very close match exists; pivot or differentiate needed
Be honest. "Already done" is valuable feedback, not failure.
Landscape Scan
When the user asks about a field, direction, or trend:
- Search recent papers (last 2-3 years) on the topic
- Identify: top groups/authors, dominant methods, open problems, emerging directions
- Summarize in 5-10 sentences — enough to orient, not overwhelm
- Note: which sub-areas are crowded vs under-explored
Citation Hygiene
When you reference a paper in conversation:
- Only cite papers you have actually found via search in this session
- Never invent paper titles, authors, or results from memory
- If unsure whether something exists: search first, then cite or say "I couldn't find it"