Unverified50% confidenceConceptExact time
Under the DNO framework, alignment is modeled as a two-player game between the policy model and a preference function (or teacher model), ideally reaching a Nash equilibrium when the answer distribution generated by the policy model is such that the preference function can no longer further distinguish better from worse.
1
Sources
50%
Confidence
Long-term
Relevance
11/4/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDNO-Prct uses an iterative process: in each round, online sampling is performed on the policy model, an LLM scores the samples together with the teacher model's golden answers, large-margin preference pairs are selected, and finally a DPO-style contrastive update is performed.75% similarUnverifiedDPO uses the log-probability difference between the policy model and the reference model as an implicit reward, enabling the model to increase the generation probability of better responses while staying close to the reference model.75% similarUnverifiedDecoupling strategy generation from strategy execution brings attributability, making it possible to determine when a strategy loses money whether the problem lies with the strategy itself rather than random model behavior.74% similarVerifiedThe output of a decision model is often a label or probability distribution over a finite action set, similar to a policy network in reinforcement learning, whereas generative models require autoregressive sampling over a large vocabulary.72% similarUnverifiedDPO的核心洞见是语言模型本身可视为隐式奖励函数，因此可直接对偏好对进行对比学习，绕过显式奖励模型72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1134221API
curl https://kongchang.com/api/v1/knowledge/claims/1134221MCP
get_claim(id=1134221)