Blog
Latest articles
9 min readReAct: The Prompting Pattern That Powers Modern AI Agents
ReAct interleaves reasoning traces with actions, enabling LLMs to use tools while maintaining a reasoning chain - the foundational pattern behind LangChain agents and modern AI assistants.
5 min readNVIDIA Nemotron-4 340B and Llama-3.1-Nemotron-70B: Enterprise LLMs From NVIDIA
NVIDIA entered the foundation model market with two distinct plays: Nemotron-4 340B for synthetic data generation pipelines, and Llama-3.1-Nemotron-70B-Instruct with an Arena Hard score of 85.1% for enterprise inference.
7 min readOpenAI o3 and o3-mini: The Next Generation of Reasoning Models
OpenAI o3 scores 87.5% on ARC-AGI and 96.7% on AIME 2024. Here's what compute-optimal inference scaling means and how to use o3-mini cost-effectively.
7 min readCogVLM2: Open-Source Video and Image Understanding With Long Context
Zhipu AI's CogVLM2 introduces a Visual Expert Module that gives visual tokens their own weight matrices, enabling richer image and video understanding than shared-weight alternatives.
7 min readVite 6: Environment API, Rolldown Experiment, and What's Next
Vite 6 introduces a unified Environment API for client/SSR/worker builds and ships an experimental Rolldown bundler that is 10x faster than Rollup.
9 min readGEO: Generative Engine Optimization - How to Rank in ChatGPT, Perplexity, and Claude
GEO is the emerging discipline of optimizing content for AI-generated answers rather than traditional search rankings. Here is what the research says actually works.
7 min readMoondream2: A 1.9B VLM That Runs on a Raspberry Pi
Moondream2 is a 1.9B parameter vision-language model that fits in 1.2GB RAM when quantized, enabling image captioning, visual Q&A, and object detection on embedded hardware and edge devices.
9 min readSelf-RAG: Teaching LLMs to Decide When to Retrieve
Self-RAG introduces reflection tokens that let the model decide whether retrieval is needed and evaluate passage relevance and citation support, outperforming standard RAG on factuality benchmarks.
8 min readLlama 3.3 70B: Why Meta's November 2024 Release Rivals 405B
Llama 3.3 70B closes most of the gap with the 405B model through better instruction following data and RLHF improvements - delivering 405B-class performance at a fraction of the serving cost.
7 min readRAGAS: The Standard Framework for Evaluating RAG Pipelines
RAGAS gives you four principled, LLM-computed metrics - faithfulness, answer relevancy, context precision, and context recall - to objectively score your RAG system.
8 min readHaystack 2.0: The Production RAG Pipeline Framework by deepset
Haystack 2.0 from deepset offers a component-based pipeline architecture for building production RAG systems, with YAML-defined pipelines and a built-in evaluation framework.
7 min readPhi-3 Mini: Running a 3.8B Parameter LLM On Your Phone
Phi-3 Mini at 3.8B parameters outperforms Mixtral 8x7B on several benchmarks and runs in browsers via WebGPU or on Android/iOS via ONNX. Here's how.