Explore the fundamentals of coding Large Language Models (LLMs) from scratch with a series of videos. These resources cover setting up a Python environment, text data preparation, coding attention mechanisms, pretraining, fine-tuning for classification, and instruction fine-tuning. A bonus video for paid subscribers discusses the evolution of LLMs from 2018 to 2025, offering insights into the current LLM landscape. The videos are designed to provide a hands-on understanding of LLM mechanics, akin to building a go-kart to grasp car engineering principles.
This blog post explores the use of test-time compute and chain-of-thought (CoT) prompting to improve the performance of large language models. It reviews recent developments in effectively using 'thinking time' and understanding why it helps, drawing parallels to human cognitive processes. The post further discusses methods for improving decoding, such as parallel sampling and sequential revision, and addresses the challenge of ensuring models faithfully represent their reasoning. Finally, it examines scaling laws for thinking time and potential future research directions.
Democratizing GenAI through a Global Enterprise Portal
Generative AI (GenAI) is revolutionizing content creation, but scaling it across global enterprises requires a balanced approach. This blog post introduces a Global Enterprise Portal (GEP) framework that centralizes foundations while decentralizing innovation. Overcoming data quality, asset accessibility, and governance challenges, the GEP framework offers enterprise asset management, a unified portal, and end-to-end governance. The GenAI Adoption Staircase illustrates a progressive four-phase implementation, starting with cloud infrastructure and advancing to diverse GenAI capabilities. A large automobile manufacturer uses GEP to build a chatbot for repair service and to harmonize product specifications.