-
Building qareen: My Experience with Multi-Agent Coding
I recently released qareen, a framework designed to solve a specific problem in LLM evaluations: balancing relevance and diversity in few-shot examples. It extends Maximum Marginal Relevance (MMR) to multimodal tasks, helping LLM-as-a-Judge workflows avoid position bias and redundancy.
Read More → -
A Summary of Anthropic’s Persona Vectors Research
In a recent paper, researchers at Anthropic introduced a fascinating new concept called “persona vectors.” This research tackles a critical challenge in AI safety: understanding and controlling the often unpredictable personalities of large language models (LLMs).
Read More → -
Using LangChain to Reduce LOC of a Chatbot
Building a powerful chatbot from scratch can be a complex and time-consuming task. It involves integrating various components, such as Large Language Models (LLMs), prompt engineering, and data retrieval systems. This often leads to a significant amount of boilerplate code, making the development process cumbersome.
Read More → -
Training Deep Neural Networks with Batch Normalization
Since its inception in 2015 by Ioffe and Szegedy, Batch Normalization has gained popularity among Deep Learning practitioners as a technique to achieve faster convergence by reducing the internal covariate shift and to some extent regularizing the network. We discuss the salient features of the paper followed by calculation of...
Read More →