Unverified50% confidenceFactExact time
Under class-incremental drift, only experience replay preserves the accuracy gain brought by handcrafted features; EWC and naive fine-tuning cannot maintain this advantage.
1
Sources
50%
Confidence
Long-term
Relevance
10/27/2026
First Seen
Sources
Related Entities
Related Claims
Unverified提示词优化的收益受限于底层工程链路的稳定性，若接口不稳定或测试环境不可复现，提示词优化效果将无法被客观衡量67% similarUnverifiedEven in modern embedding models optimized through instruction fine-tuning or contrastive learning (such as BGE and E5), the anisotropy problem has not been completely eliminated, only reduced in degree.65% similarUnverifiedPart of the gain from ensembling comes from multiple sampling itself rather than a substantive improvement in model capability, so directly comparing ensemble results with a model called a single time is methodologically not equivalent.65% similarUnverifiedSwitching to a stronger model yields limited benefit, because capability improvement is not the same as filling in information—what is missing is not language ability but context.65% similarUnverifiedThe formula deliberately omits productivity gains brought by AI tools because such gains are difficult to decouple from the tool versus engineer capability, and including them would introduce a confirmation bias entry point.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1102417API
curl https://kongchang.com/api/v1/knowledge/claims/1102417MCP
get_claim(id=1102417)