Hive Minds and Hallucinations: A Smarter Way to Trust LLMs
A mechanism-first look at how a pharmacy SMS prototype uses deterministic parsing, fuzzy logic, and LLM cross-checking to make hallucination risk operationally manageable.
Cognaptus Insights
Industry thoughts, trends, and commentary.
Sharpe Thinking: How Neural Nets Redraw the Frontier of Portfolio Optimization
A mechanism-first reading of a neural covariance-cleaning model that improves global minimum-variance portfolios without pretending to forecast returns.
Chains of Causality, Not Just Thought
How Causal Influence Prompting turns agent safety from vague warning text into an explicit causal control layer for risky actions.
Chatbot at the Table: Rethinking Group Recommendations with GenAI
A mechanism-first reading of why generative AI may turn group recommenders from ranking engines into decision facilitators.
ChatGPT and the Death of Effort: Is AI Turning Students into Lazy Thinkers?
A small experimental study suggests that ChatGPT-assisted writing can reduce students’ self-reported cognitive engagement, with lessons for how organisations design AI-assisted work.
The Grammar and the Glow: Making Sense of Time-Series AI
A practical synthesis of two arXiv papers showing why time-series AI needs both reusable temporal motifs and human-readable explanations.
Agents Under Siege: How LLM Workflows Invite a New Breed of Cyber Threats
A mechanism-first reading of why agentic AI security is not just prompt filtering, but control over tools, memory, credentials, protocols, and trust boundaries.
Beyond the Pull Request: What ChatGPT Teaches Us About Productivity
Italy’s temporary ChatGPT ban shows that LLM productivity is not only about faster coding, but also about collaboration, learning, onboarding, and organisational resilience.
Grounded and Confused: Why RAG Systems Still Fail in the Enterprise
HERB shows why enterprise RAG failures are less about chatbot polish and more about evidence retrieval across messy, heterogeneous work.
Swiss Cheese for Superintelligence: How STACK Reveals the Fragility of LLM Safeguards
A mechanism-first reading of STACK, a staged attack showing why layered LLM safeguards can still fail when their internal components leak enough information to optimise against them.