Unverified50% confidenceFactExact time
On Gemma-2-9B and Qwen3-14B, the researchers successfully switched the facts read by the model by modifying early internal activation vectors while keeping the written instructions unchanged.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe author used the Qwen3-1.7B model for both generation and attribution analysis tasks.70% similarUnverifiedResearchers validated the approach on models such as SMDM and Qwen3 for new facts and natural instructions, and the directional conclusion that CAGD improves forgetting still holds.69% similarUnverifiedThe researchers validated the parity conclusion on the vision-language model Qwen2.5-VL-3B-Instruct for a single compression ratio using per-scheme learning rate tuning repeated with three random seeds.67% similarUnverified模型权重在推理过程中呈现一次写入、多次读取的特征，与闪存读快写慢的物理特性天然吻合66% similarUnverifiedTEFM was tested on multiple mainstream large language models including Qwen3, Gemma-2, and Phi-4, achieving consistent performance improvements and demonstrating its model-agnostic nature.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1001922API
curl https://kongchang.com/api/v1/knowledge/claims/1001922MCP
get_claim(id=1001922)