A 1.7B model trained with reward-shaped reinforcement learning scores 78.3% on SimpleQA, which the authors claim matches much larger models.
This provided a consistent set of contextual information (tool calls and tool responses) for each question
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CL 1years
2025 1verdicts
REJECT 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Lucy: edgerunning agentic web search on mobile with machine generated task vectors
A 1.7B model trained with reward-shaped reinforcement learning scores 78.3% on SimpleQA, which the authors claim matches much larger models.