Unverified50% confidenceBenchmarkExact time
The researchers constructed a benchmark test set called ALIGNMENTFORECASTBENCH, covering more than 5,000 prediction questions, 17 target models, 32 datasets, and 16 failure modes.
1
Sources
50%
Confidence
Long-term
Relevance
9/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe research team constructed a new evaluation set, MegaRCA-Mix, containing 50 manually annotated failure cases covering long-horizon, execution-intensive tasks.77% similarUnverifiedThe research team selected 8 models spanning different capability tiers and conducted cross-evaluations across 54 benchmarks75% similarUnverifiedThe study employs four complementary causal diagnostic methods, covering four decoder-only architecture models and eight distinct prompt families.74% similarUnverifiedThe research team tested 22 open-source LLMs from six model families72% similarUnverifiedThe research conducted systematic experiments on two datasets, HelpSteer and UltraFeedback, across seven groups of objective pairings.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/961165API
curl https://kongchang.com/api/v1/knowledge/claims/961165MCP
get_claim(id=961165)