Flame Tamed: Can LLMs Put Out the Internet’s Worst Fires?
A comparison-based reading of new research on LLMs as online mediators, separating moderation, model performance, human style, and practical deployment boundaries.
Cognaptus Insights
Industry thoughts, trends, and commentary.
Prompting on Life Support: How Invasive Context Engineering Fights Long-Context Drift
A mechanism-first reading of Invasive Context Engineering, a training-free proposal for keeping LLM control instructions alive inside long conversations and agentic reasoning loops.
Scan, Plan, Report: When Agentic AI Starts Thinking Like a Radiologist
A mechanism-first look at why Radiologist Copilot matters less as a report generator and more as a workflow engine for high-stakes medical AI.
Stuck on Repeat: Why LLMs Reinforce Their Own Bad Ideas
A mechanism-first reading of Martingale Score, a new unsupervised way to detect when LLM reasoning becomes prior-protecting rather than truth-seeking.
Blunders, Patterns, and Predictability: What n‑Gram Models Teach Us About Human Chess
A mechanism-first look at how skill-specific n-gram models turn chess move prediction from optimal play into human behavior modeling.
Checkmating the Hype: What LLM CHESS Reveals About 'Reasoning Models'
A mechanism-first reading of LLM Chess, showing why interactive benchmarks expose failures that static reasoning tests often miss.
From Building Blocks to Breakthroughs: Why RL Finally Teaches Models to Think
A mechanism-first reading of why reinforcement learning helps models compose memory and context only after supervised training has built the right atomic skills.
Ground and Pound: How Iterative Reasoning Quietly Redefines GUI Grounding
Chain-of-Ground shows that GUI grounding can improve not only by training larger models, but by forcing multimodal models to revisit their own visual hypotheses.
Roots of Understanding: When Transformers Try to Learn the Language of Numbers
A mechanism-first analysis of how a GPT-2-style transformer partially learns arithmetic structure from rooted-tree Dyck words—and why that is a benchmark lesson, not a factoring breakthrough.
Rules of Attraction: How LLMs Learn to Judge Better Than We Do
A mechanism-first reading of learned-rule-augmented LLM evaluators, and why the next AI judge may need better rubrics before bigger brains.