If Logic, Then Trouble: Why LLMs Still Miss Human Conditionals
A mechanism-first reading of why LLMs can follow conditional logic yet still fail at the pragmatic reasoning businesses actually need.
Cognaptus Insights
Industry thoughts, trends, and commentary.
Reasonable Doubt: Why LLM Reasoning Needs Process Control
A three-paper synthesis showing why dependable LLM reasoning needs mechanistic caution, multidimensional evaluation, and adaptive scaffold design rather than leaderboard confidence.
Think Longer, Act Smarter: Why Coding Agents Need Behavior-Preserving Reasoning
A mechanism-first reading of M2A, a training-free method for injecting mathematical reasoning into coding agents without breaking their think-act-observe loop.
Do the Math, Not the Mime: Why LLM Reasoning Needs a Verification Pipeline
A mechanism-first reading of why LLM mathematical reasoning fails when fluent explanations are mistaken for verified symbolic work.
Don’t Average the Needle: Spectral Retrieval and the RAG Evidence Problem
A mechanism-first reading of Spectral Retrieval: why dense retrieval can bury localized evidence, how multi-scale sinc convolution tries to recover it, and where the business value actually begins.
Don’t Just Guard the Door: Jailbreak Safety Needs Checkpoints
A practical synthesis of three jailbreak-defense papers showing why AI safety should test the path from prompt to response, not just the prompt itself.
Jailbreak Risk Needs a Stopwatch, Not Just a Scorecard
A business-oriented framework for evaluating LLM jailbreak risk across prompt quality, reasoning traces, and time-to-failure under repeated attacks.
Query the Receipt, Not the Vibe: DualGraph and the RAG Catalog Problem
A mechanism-first reading of DualGraph, SpecsQA, and why semi-structured business QA needs symbolic querying alongside semantic retrieval.
RAG’s Receipt Problem: When Correct Answers Don’t Prove Retrieval
Why enterprise RAG evaluation needs both leakage-resistant benchmarks and internal attribution diagnostics before it can claim evidence-grounded answers.
Read the Receipt: Why RAG Should Highlight Before It Answers
A mechanism-first reading of ACL-Verbatim, showing why trustworthy research QA may need extractive evidence spans before generative answers.