Unverified50% confidenceOpinionExact time
Even with instruction fine-tuning or RLHF alignment, underlying register bias is difficult to fully eliminate, because alignment data is also predominantly in standard registers.
1
Sources
50%
Confidence
Long-term
Relevance
9/17/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe preference data that RLHF relies on has systematic flaws such as low inter-annotator agreement, position bias, and length bias.72% similarUnverified端到端模型在代表性充足与代表性不足口音之间的WER差距并未按比例缩小，属于数据分布问题而非架构问题70% similarUnverifiedIn large language models, linear layers do not include bias by default (disabled via bias=False), whereas traditional computer vision models such as ResNet typically retain bias.69% similarUnverifiedR²Adapter's rewriting function rewrites borderline queries to more clearly expose implicit multi-hop reasoning requirements, improving retrieval quality without the need for additional supervision signals.67% similarUnverifiedRLHF alignment has a 'one-size-fits-all' problem, where the same safety standards are applied to all scenarios, making it impossible to distinguish between malicious requests and legitimate creative writing needs.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/928508API
curl https://kongchang.com/api/v1/knowledge/claims/928508MCP
get_claim(id=928508)