ToolLoop: Three-Stage Reverse Synthesis of Tool-Call Training Data (EMNLP 2026)
jiqizhixin · x · 2026-09-29
Tool calls let LLMs reach external APIs, but synthesizing tool-use training data usually follows a "generate first, filter later" pipeline: the model emits user questions and tool calls in one shot, then rules or a model decide what to keep. Malformed calls often slip past single-pass format checks, and the pipeline optimizes for final pass/fail rather than the actual target function.
ToolLoop, accepted to EMNLP 2026 Main Conference, splits synthesis into three stages — target-function sampling, reverse user-question generation, and forward tool-call generation — with dynamic self-feedback at each stage. It uses explicit intermediate results to constrain later generation, aligning the target function, user intent, and tool calls via generate-verify-correct.
More from Models
- Opus 5.5 still feels unlimited in Claude Code, putting pricing pressure on OpenAI's Dev Day — Angaisb_ · 2026-09-29
- ChatGPT Pro Max tier spotted in development as OpenAI DevDay nears; $2000 price rumored — scaling01 · 2026-09-29
- Sonnet 5.5 one-shots a full $100K/month app in a single prompt — PrajwalTomar_ · 2026-09-29
- Bindu Reddy: OpenAI may not be releasing a new model tomorrow — bindureddy · 2026-09-29
- ChatGPT Pro's subsidized compute era ends: $200 tier usage halved, new $500 plan matches old limits — Norwood_Reaper_ · 2026-09-29
- Opus 5.5 writes perfect HyperFrames videos: lessons from studying its model behavior — toolstelegraph · 2026-09-29