Unverified50% confidenceFactExact time
llm-beliefs did not publish the specific prompts, model version, or parameter settings, resulting in limited interpretability of the results.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/8/2026
First Seen
Valid until: 2/6/2027
Sources
Related Entities
Related Claims
UnverifiedLLMs are architecturally unable to distinguish between trusted instructions and untrusted inputs, and therefore a purely technical solution analogous to parameterized queries has yet to emerge.72% similarUnverifiedThe author believes Jev should not be used as an LLM judge, because it lacks the reasoning ability needed to distinguish between the quality of different LLM outputs.69% similarUnverifiedThe model can write proposals, but the model cannot write approvals.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1149381API
curl https://kongchang.com/api/v1/knowledge/claims/1149381MCP
get_claim(id=1149381)