VLDB talk: DR Tulu trains deep research agents with rubrics to pick BM25 or vector retrievers
CShorten30 · x · 2026-09-05
- Cites Akari Asai's VLDB talk covering two deep research agent training efforts:
- DR Tulu: trains deep research agents with rubrics to choose between BM25 and vector search retrievers, with strong results.
- AgentIR: trains a dedicated embedding model for the agent's intermediate reasoning and search queries.
- The author suggests combining AgentIR with BM25/ColBERT-style feedback — a meet-in-the-middle between searching by rationale and what the search returns.
- The retweeter notes how basic Deep Research, cutting-edge in early 2025, already looks — a sign of how fast the field iterates.
Related event: VLDB Talk: Deep Research Agents Must Co-Train with Retrievers(3 posts)→
More from coding & agent
- Agents coordinating via a shared message board as persistent state — novasarc01 · 2026-09-05
- Open-Source Writeup on RL Training Knowledge-Work Agents at Scale Goes Viral — heghbalz · 2026-09-05
- Neon Fall: a CRT-style Tetris clone built with GPT few-shot prompting — intellectronica · 2026-09-05
- Supermemory hits ~6-fig MRR, opens affiliate program with 30% recurring commission — julianweisser · 2026-09-05
- After ChatGPT, Claude & Grok All Went Dark, One User's 3-Machine Local AI Lab Kept Running — cocktailpeanut · 2026-09-05
- Almost all agent sandbox breakouts came from explicit hacking goals, argues Rob LeClerc — robleclerc · 2026-09-05