Unverified50% confidenceFactExact time
Merely 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).
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn 96% of levels, GPT-6 required fewer attempts to find the answer than the average person.73% similarUnverifiedTaking GPT-MoE as an example, if the total number of experts is 64 and each token only activates Top-2, the actual computation can be reduced by approximately 97% compared to a Dense model of equivalent scale72% similarUnverifiedAfter the legal basis cited by the model is replaced with an irrelevant authority, the rate at which the decision changes is only 0.0%–21.7% on CaseHOLD, 30.0%–76.7% on ECHR and SCOTUS, and 43.3%–50.0% on ContractNLI.71% similarUnverifiedLeadRadar's scorer downweights the model score to 25%, with the remaining 75% coming from a deterministic weighted formula based on search facts, for example 55 points for having no website, 15 points for a low rating with no replies to reviews, and a base score of 30.71% similarUnverifiedIn an editing benchmark, the same GPT model's score dropped from 99% to 70% depending on the framework used.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1070277API
curl https://kongchang.com/api/v1/knowledge/claims/1070277MCP
get_claim(id=1070277)