arXiv：cs.AI· Hyojung Han, Ryangkyung Kang, Youngeun Kim, Seung-Hun Jeon·· 4 小时前
企业策略决策该用学习还是推理？27B 韩语企业模型上的实证研究
When Should Enterprise Policy Decisions Be Learned, and When Should They Be Reasoned?
AI 导读
研究固定 27B 韩语企业模型，对比生成 JSON 与单一受限 token 两种决策接口及是否先推理：在 2800 条分布内样本上两种部署路径准确率无差异（93.46% vs 94.14%），但重启稳定性相差约十倍（动作翻转 3.00% vs 0.29%）。
正文
Abstract:An enterprise agent that answers under a written policy must choose an action before it says anything: answer, ask for a missing fact, ask for confirmation, call one or several tools, escalate, or refuse. Holding one 27B Korean enterprise model fixed, we vary whether the decision is a generated JSON object or a single constrained token, and whether the model may reason first.
On 2800 in-distribution items we did not detect an accuracy difference between the two deployed paths (93.46% vs 94.14%), while they differ about tenfold in restart stability (3.00% of actions flip vs 0.29%); they also differ in reasoning, so this is an operational comparison.
On a sealed benchmark of 300 items whose labels come from a hidden executable policy, reasoning is worth +17.3 points [+12.3, +23.0] on exact action. With reasoning off, the format difference was 0.0 points [-4.3, +4.0], a failure to detect an advantage for either interface, not equivalence; with reasoning on it is unresolved. Against a token-matched random control, boundary-targeted supervision moves the two covered boundaries by +17.5 and +20.0 points and leaves a held-out boundary unchanged. In a 3-seed training x reasoning factorial, the sealed-rendering interaction is -4.9 [-8.1, -1.7], consistent with partial substitution.
All supervision was Korean. On independently rendered English versions, the model is 0.82 points lower [-2.68, +0.93] with reasoning on, the same order as re-rendering in Korean; the reasoning-off gain from targeted supervision is not stable in either language, so we do not claim transfer.
The original text stays private as the confirmatory set. An open-model re-rendering made after these results reproduced the reasoning effect and the English verdict; the interaction kept its sign but did not replicate under item-clustered intervals. We release the model and that re-rendering with gold labels.
| Comments: | 43 pages, 17 tables, 9 figures |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.17643 [cs.AI] |
| (or arXiv:2610.17643v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.17643 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Ryangkyung Kang [view email]
[v1]
Tue, 13 Oct 2026 03:14:12 UTC (566 KB)
来源：arXiv：cs.AI · arxiv.org