CDSE separates an agent skill into planning, experience, tools and scripts
A skill-evolution framework updates four coordinated components instead of treating a reusable skill as one undifferentiated document.
Explore Franklin AI
Understand new AI research with readable explanations and links to the original papers.
A skill-evolution framework updates four coordinated components instead of treating a reusable skill as one undifferentiated document.
Hopper GPU kernels pair FP8 prefill with INT4 decode, reporting speed gains alongside a measured perplexity cost.
A research system develops missing laboratory capabilities in a digital twin and validates workflows through physical experiments.
Per-sample sparse coding improves agreement between independently learned dictionaries in synthetic and model-activation experiments.
A research framework models changing subjective states, reporting results on DynToM and RecToM while leaving real-person validation open.
A drug-discovery harness links scientific operations, evidence and skill review, with benchmark and laboratory results reported separately.
A learned search policy refines same-formula molecular candidates, while fingerprint prediction remains a limit on exact identification.
A single monitor is trained to assess different behaviours in reasoning traces, using broad or specific natural-language factors.
Agents reproduced 77.4% of atomic claims in a 20-paper benchmark, while expert review showed why robustness flags need human assessment.
A temporal locator predicts when relevant evidence occurred, then merges that retrieval channel with ordinary semantic search.
Start with the overview to understand the paper’s main idea, then follow its original paper link for the methods, results and limitations.
Use Submit Research to send the original paper URL and explain why it is useful. An editor reviews the suggestion before publication.