Explore Franklin AI
Understand new AI research with readable explanations and links to the original papers.
A medical-abstract study compares five explanation methods around a zero-shot DeBERTa classifier. It examines where category ambiguity coincides with unstable attributions, without
The Homomorphic Advantage Operator centers reinforcement-learning targets to help keep polynomial activations within bounds. Its encryption, stability and privacy findings require
HydroJEV combines a training-free probability screen with a rule-tree gate and slower review. Its simulated attribution benchmark tests reviewer savings alongside supervised-learni
Mimir uses simulation to check and revise language-model irrigation proposals. Its retrospective study examines bounded control and memory adaptation without letting the model chan
Hidden-state detectors locate unsupported spans in generated text. The study finds that another model can sometimes detect onsets better than the generator, with important model-ac
A hundred long-form game-design documents support code, replay, and adaptive-play checks tied to the same requirements. The benchmark distinguishes a game that runs from one that f
A runtime-controlled research harness combines complete coding-agent products and preserves experimental evidence. Its matched-budget evaluation reports better execution accuracy,
A training-free research framework measures how memory-conditioned representations respond to hypothetical changes. It aims to distinguish inappropriate reuse of past experience fr
COAT compares ownership allocations across forty-two scenarios, twenty-four model configurations, and 108 participants. Average human-model similarity hides differences in how resp
A human-AI puzzle study separates measuring uncertainty from communicating it. Detailed descriptions and accurately placed warnings improve judgments of wrong moves, but the strong
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.
