Unverified50% confidenceCautionExact time
At the scale of small models with a few hundred million parameters, reinforcement learning methods such as GRPO may not be applicable, as the models may be unable to produce positive samples of sufficient quality to provide an effective learning signal.
1
Sources
50%
Confidence
Long-term
Relevance
8/29/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAll three models failed to pass the GRPO stage; while GRPO showed slight improvement in teaching the models arithmetic, general capabilities degraded significantly.74% similarUnverifiedIf samples are too easy the model has nothing to learn; if they are too difficult, noisy learning signals may interfere with training; the most effective learning occurs in the region slightly above current capability but still within reach.69% similarUnverifiedIn testing, small models of approximately 1–2B parameters were unable to accurately locate and complete complex tool-calling tasks within lengthy system prompts.67% similarUnverifiedWhen model parameter scale exceeds certain thresholds, capabilities spontaneously appear that were never explicitly required by the training objective, such as few-shot reasoning, code generation, and multi-step planning.67% similarVerified若用含测试集的全量数据计算均值和标准差，会导致预处理泄漏，使评估指标虚高，无法真实反映模型泛化能力65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/819749API
curl https://kongchang.com/api/v1/knowledge/claims/819749MCP
get_claim(id=819749)