Unverified50% confidenceEventExact time
Moonlight's first version had gemma3:4b classify each clause into three categories (yes/unclear/no), and all 8 clauses tested returned unclear; after rewriting the prompt and adding two examples, the same model returned yes for all of the same 8 clauses.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/18/2026
First Seen
Valid until: 1/16/2027
Sources
Related Entities
Related Claims
UnverifiedAfter testing, programming blogger Theo pointed out that more than half of Gemini 4's judgments about his codebase were wrong, making it the only one of the 7 tested models with more than half errors.55% similarUnverifiedIn the new version of the 24-point problem (numbers 2, 7, 8, 10, requiring three different solutions), DeepSeek V4.1 Flash got all three solutions wrong, each violating the constraint that each number can only be used once.55% similarUnverifiedIn a test of 12 questions about a 1,061-line pipeline file (three of which asked about features that do not exist), all three models scored a perfect 12/12 and did not invent any features.55% similarUnverifiedIn the REA demonstration, the rewritten function passed 3,205 test cases, and all 63 bytes of the compiled output were identical.54% similarUnverified在285次实测运行中，Gemini取得283次语义通过、2次语义部分通过、0次可复现语义失败54% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1063883API
curl https://kongchang.com/api/v1/knowledge/claims/1063883MCP
get_claim(id=1063883)