Reproduce “Can You Put it All Together: Evaluating Conversational Agents' Ability to Blend Skills” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Eric Michael Smith, Mary Williamson, Kurt Shuster, Jason Weston, Y-Lan Boureau

The authors train a shared response-retrieval model across three English-language tasks involving persona, knowledge, and empathy, and compare it with specialist models selected by a dialogue manager. On the BlendedSkillTalk test set, the multi-task model reaches 80.1 hits@1, versus 76.8 for the strongest single-skill model; fine-tuning on BlendedSkillTalk raises the score to 83.8. These results measure candidate-reply ranking, not unrestricted text generation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.