Search for a command to run...
Calibration Collapse After RLHF; and How to Fix It Without Retraining Reinforcement Learning from Human Feedback makes language models more helpful and less harmful. It also makes them systematically
Abstract AI agent products are proliferating faster than the operational frameworks needed to deploy them well. Most enterprise and SME deployments fail not because the underlying models are incapable
The Concentration Problem at the Heart of Vector Search Every major embedding model produces hundreds or thousands of dimensions. You use cosine similarity on them. There is a theorem that says this
The Johnson-Lindenstrauss Lemma Against PCA and Neural Networks A theorem from 1984 claims you can throw a random matrix at high-dimensional data and preserve its geometry almost perfectly, with no
Agentic AI in Credit & KYC and the Governance Gap That Follows Abstract Agentic AI systems are now capable of executing end-to-end credit decisioning and KYC workflows with minimal human input, retri
There is a moment in every production AI system that most engineering teams don't fully design for. The model has run. The prediction is out. And now, something has to happen with it. That moment, the
This week, I noticed a Twitter thread where someone was asking about managing API rate limits when using Google's Gemini API. I've faced similar challenges in my own projects, so I wanted to share my straightforward solution. The Problem If you've wo...
I used AI to help me arrange my thoughts and sentences better It was supposed to be a straightforward project. The client had an existing Laravel 5.7 application that needed a few new features. Simple enough, right? I had just set up my shiny new dev...
Introduction I used AI to arrange my thoughts for this article* So, in the spirit of vibe-coding and thinking of what to use AI for, I was short on ideas and decided to spend a few hours exploring the concept of AI-powered error handling. For starter...