Oct 12, 2026
RATED adds execution reliability to agent tool retrieval
A retrieval framework combines semantic matching with feedback about runtime failures, aiming to reduce selection of unavailable tools.
Explore Franklin AI
AI Research Papers
Understand new AI research with readable explanations and links to the original papers.
Oct 12, 2026
Wikipedia Game study tests language models across languages and literature graphs
A research study compares model navigation with human and TF-IDF baselines, finding performance depends on language and subject.
Oct 11, 2026
OnTrack monitors agent execution graphs before a failing run finishes
A streaming monitor reports early failure-ranking gains while making its reference and schema requirements explicit.
Oct 11, 2026
GeoReform improves geometry solving by revising how diagram facts reach the model
Reflective policy search targets distracting relations and ambiguous grounding, with gains scoped to the tested pipeline.
Oct 11, 2026
Agent study separates answering correctly from handling unresolved uncertainty
Trajectory-level checks show how agents can notice a conflict early yet omit uncertainty from an incorrect final answer.
Oct 11, 2026
BrickBench separates buildable LEGO designs from designs people prefer
The benchmark gives agents construction feedback while measuring physical checks, prompt alignment and design preference separately.
Oct 11, 2026
Legal-reasoning audit finds correct citations do not establish verdict dependence
Authority swaps and prompt-injection tests expose different limits of generated legal explanations.
Oct 11, 2026
AI-agent ecology model shows how collaboration can create a population growth threshold
A toy theory separates individual capability from population-dependent risk and explains the assumptions behind its threshold.
Oct 11, 2026
HANS tests handwriting recognition on corrected, mixed-format mathematics answers
A new answer-sheet dataset and noise-aware model distinguish surviving work from strikethroughs and erasures.
Oct 11, 2026
HRIL preserves capacity for multimodal signals that pairwise alignment can miss
A tensor-based learning method targets joint information and states the limits of its theoretical guarantee.
Good to know
How do I use these AI research explainers?
Start with the overview to understand the paper’s main idea, then follow its original paper link for the methods, results and limitations.
Can I suggest a research paper?
Use Submit Research to send the original paper URL and explain why it is useful. An editor reviews the suggestion before publication.