Unverified50% confidenceBenchmarkExact time
On 3,842 held-out questions (random guessing accuracy is 25.0%), the two small encoders using the raw last-token vectors achieved accuracies of 33.1% and 33.0%, respectively.
1
Sources
50%
Confidence
Long-term
Relevance
11/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn a write operation that scales feature encodings by three times, the graph built from previously active features has a median relative error of at least 0.247, while the maximum error of the margin graph is only 0.033.77% similarUnverifiedThe locally run Gemma 26B achieved an accuracy of only 49.2% across 453 scored predictions, while its paired always-predict-up baseline had an accuracy of 53.4%.77% similarUnverifiedAcross a range of question sizes spanning 256x, the error judgment rate of a single run is almost unchanged: 0.531 at T=16 and 0.534 at T=4096.75% similarUnverifiedOn point-in-box accuracy, the 279M ranker reaches 73.4%, zero-shot UI-TARS-2B reaches 29.6%, and Qwen2.5-VL-7B reaches 23.7%.74% similarUnverifiedAfter adopting Cursor, NVIDIA's code commit volume reached 3 times the previous level, while the defect rate remained flat.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1186762API
curl https://kongchang.com/api/v1/knowledge/claims/1186762MCP
get_claim(id=1186762)