Unverified50% confidenceFactExact time
Gender swapping caused 1% to 3.5% of top answers to change, with part of Jev's changes stemming from fluctuation between its calls.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/30/2026
First Seen
Valid until: 1/28/2027
Sources
Related Entities
Related Claims
UnverifiedProbing layer 2 (mean + standard deviation pooling) of EmbeddingGemma-2 raises VoxPopuli gender clustering from 0.05% at the final output to 66.6% (layer 3)/above 65% (layer 2), and IEMOCAP gender classification from 66.4% to 83.0%.64% similarUnverifiedThe confidence thresholds recommended by model manufacturers increased the proportion of stereotypes among retained answers on ambiguous BBQ gender questions, with granite-switch's stereotype proportion rising from 0.76 to 0.85 and pplx's from 0.81 to 1.00.64% similarUnverifiedExperimental results show that answer changes produced by different-category substitutions are 49.7 to 60.9 percentage points higher than those from same-category substitutions.63% similarUnverifiedMerely changing the order of options could shift the share of a certain option for GPT-6-luna from 16% to 40%; changing the internal ids of options affected Jev's answers by 37.9 points (out of 100), almost equal to the effect of removing all hand information (37.1 points).61% similarUnverifiedThe impact of reasoning-chain corruption on final answers was relatively small (2%–30%), and only 2.9%–35.3% of support-judgment flips co-occurred with answer changes.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1115471API
curl https://kongchang.com/api/v1/knowledge/claims/1115471MCP
get_claim(id=1115471)