Category / T-CAT / RESEARCH
Research
Articles63
This week10
Last updatedJuly 13, 2026
Research / T-2026-3733
Why CORE matters: reasoning improvement without the rollout tax
CORE lets language models improve reasoning with minimal rollouts by generating insights from contrasting success and failure.
SourceCORE: Contrastive Reflection Enables Rapid Improvements in Reasoning
Research / T-2026-6117
Five Frontier LLMs Disagree on 67% of Fact-Check Claims
Five frontier LLMs disagree on 67% of 1,000 real-world fact-check claims, raising questions about their reliability as knowledge tools.
SourceFive frontier LLMs disagree on 67% of 1k real-world fact-check claims
Research / T-2026-0167
Obra's 'Superpowers' Is a Rare Thing: An Agentic Framework That Admits What It Doesn't Know
Obra's open-source agentic skills framework offers a grounded alternative to the industry's sprawling agent benchmarks, focusing on composable, testable capabilities.
Sourceobra/superpowers