Unverified50% confidenceDecision RuleExact time
The tool call parser must match the model family; different model families (Qwen, Llama, Mistral, etc.) have different output formats, and choosing the wrong parser will fail.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIf the data always provides tools that get used, the model learns the erroneous association that it must use a tool whenever it sees one.64% similarUnverifiedWhen multiple tools are semantically similar, the model's tool selection is essentially a multi-class classification problem, and the more candidate classes there are, the higher the misclassification rate.64% similarUnverifiedMost of what goes wrong in LLM features is not in the model itself but in the surrounding code, such as building the messages list, selecting the model ID, parsing JSON, handling an empty choices list, retrying on 429s, and counting tokens.63% similarUnverifiedMisreporting tool execution errors as protocol errors causes the model to miss useful hint information and be unable to self-correct, causing the agent to stop or enter an infinite loop.63% similarUnverifiedIn the summarization task, 4 out of the 8 models failed: the 4-billion-parameter Qwen3.5 fabricated content, LFM2.5 altered facts, Bonsai added nonexistent information, and MiniCPM5 missed key points.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1031395API
curl https://kongchang.com/api/v1/knowledge/claims/1031395MCP
get_claim(id=1031395)