In the news
Keyword Harnesses Fail Open: A Cheap Diagnostic Ladder for Tool-Use Claims in Small Language Models
arXiv cs.AI · Published · 3 min read
In 30 seconds
- What happened
- Keyword-matching benchmarks falsely credit small language models for tool use they cannot actually perform, requiring stricter diagnostics.
- Why it matters
- Matters when evaluating tool-use capabilities of small models under 1.2 billion parameters before deployment in production systems.
- Watch out
- The proposed diagnostic ladder is cheap but requires careful implementation; seed sensitivity limits generalizability of some findings across different model architectures.
- language model
- small model
- tool use
- tool-use
- tokenizer
The patterns behind this
Each one covers how the technique works, when it earns its cost, and where it breaks.
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.