The Prompt Knew the Odds. CRISTAL Put Them in Code
CRISTAL shows why analyst systems may need LLMs for interpretation but explicit probabilistic code for evidence weighting, updating, and final decisions.
Cognaptus Insights
Industry thoughts, trends, and commentary.
Fair on Clean Data, Fragile After Fake Profiles
A recommender can pass a clean-data fairness review and still develop larger subgroup gaps after coordinated fake profiles enter its retraining data.
The Answer Looked Clinical. The Critical Steps Were Missing.
A clinician-built evaluation shows why polished medical AI responses can pass presentation checks while omitting the reasoning steps most closely tied to patient harm.
The Smart Part Was the Memory, Not the Controller
Neural Subspace Reallocation shows that continual-learning recovery depends more on storing and retrieving compact task adapters than on training a sophisticated allocation controller.
Fewer Extreme Costs, Higher Average Cost: Risk-Aware Planning Beyond Expected Reward
A practical reading of how risk-aware GUMDPs trade average performance for fewer severe trajectory-level outcomes, and where that framework remains unproven.
One Correction, Every Case: When LLMs Actually Update the Rule
A controlled reversal-learning study shows how to distinguish genuine rule transfer from gradual case-by-case recovery in large language models.
Stocked but Not Synthesizable: URSA Tests the Chemistry Inside the Route
URSA shows why route completion is a weak procurement metric and offers a staged way to screen retrosynthesis systems before chemists commit laboratory effort.
A Full Distribution Is Not a Risk Certificate
A state-level audit shows why distributional reinforcement-learning heads should not drive safety decisions until their strongest risk claims are independently validated.
Five Answers, One Bad Retrieval: When RAG Agreement Misleads
A diagnostic framework for detecting RAG failures that remain invisible when repeated answers agree.
Same Agent, Different Audience: When Social Pressure Changes the Recommendation
A dual-channel benchmark shows how authority, sponsorship, and future dependence can alter an agent’s public recommendation without changing its model or stated task.