Unverified50% confidenceFactExact time
Darwin-397B-ZTC's 93.4% on GPQA Diamond used greedy decoding with no majority voting, which is not under the same inference budget as Darwin-180B-RSI's 16-sample majority voting.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/19/2026
First Seen
Valid until: 1/17/2027
Sources
Related Entities
Related Claims
UnverifiedDarwin ZTC scored 88.0 on the common-sense morality judgment benchmark, ranking first.62% similarUnverifiedOn Track B, the flat RAG baseline can detect 76%–97% of true contradictions, but produces false positives on 16%–43% of surface-similar safe drafts (the specific ratio depends on the backend).62% similarUnverifiedThe GPQA Diamond subset has 198 questions in total, and at a 95% confidence interval the standard error of a single sample is about 2 percentage points, so a difference of 0.7 points is within the noise range.60% similarUnverifiedBoth models expressed almost identical confidence when fabricating versus when answering correctly (Gemini 93.7% vs 95.5%, Gemma 96.3% vs 99.0%), indicating that confidence scores cannot distinguish correct answers from fabrications.60% similarUnverifiedOn a test set of 600 synthetic invoices, the QLoRA fine-tuned model achieved a full-invoice exact rate of 83.0%, an auto-approval rate of 86.8%, and 0 out of 461 validation-coverage field errors in approved invoices (upper bound 0.8%).59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1067096API
curl https://kongchang.com/api/v1/knowledge/claims/1067096MCP
get_claim(id=1067096)