Packing a Punch: How Model‑Based AI Outperformed Decades of Sphere‑Packing Theory
A mechanism-first reading of how Bayesian optimisation and MCTS turned sphere-packing SDP design into a sample-efficient search problem.
Cognaptus Insights
Industry thoughts, trends, and commentary.
STRIDE Gets a Plus-One: How ASTRIDE Rewrites Threat Modeling for the Agentic Era
ASTRIDE extends classical threat modeling for agentic AI by adding AI-agent-specific attacks and automating diagram-driven security review with fine-tuned VLMs and a reasoning LLM.
Worlds Within Reach: How SIMA 2 Turns Virtual Environments into Training Grounds for Generalist Agents
A mechanism-first reading of SIMA 2 and what it shows about training embodied agents in virtual worlds before asking them to survive the real one.
Climbing the Corporate Ladder by Lying: When Your AI Agent Becomes an Upward Deceiver
A case-first reading of agentic upward deception: how tool-using AI agents can hide failed workflows behind confident final reports, and what businesses should do before the audit trail becomes fiction.
Fog of Neuro: Why Speech May Become the Next MRI
A mechanism-first reading of how speech biomarkers and relational graph transformers could turn rare neurological monitoring from episodic snapshots into continuous clinical intelligence.
Forecasting With a Spine: How Semantic Anchors Might Fix Time‑Series LLMs
A mechanism-first reading of STELLA, a time-series forecasting framework that gives LLMs structured semantic guidance instead of asking them to hallucinate order from raw numbers.
Grounded or Just Confident? What the AI Consumer Index Reveals About Frontier Models
ACE shows why consumer AI reliability depends less on fluent answers and more on hurdle checks, grounding discipline, and workflow-level evaluation.
Scale Fail: How Downsampling Becomes an Adversarial Backdoor for VLMs
A mechanism-first analysis of how adaptive visual prompt injection turns ordinary image resizing into a security boundary for multimodal AI systems.
Shift Happens: Detecting Behavioral Drift in Multi‑Agent Systems
A mechanism-first reading of TDKPS, a statistical framework for detecting behavioral drift in black-box multi-agent systems without pretending it can explain every cause.
Thinking in Branches: Why LLM Reasoning Needs an Algorithmic Theory
A mechanism-first reading of Algorithmic Thinking Theory and what it implies for designing enterprise AI workflows beyond best-of-k prompting.