Hugo’s blog

42 Lessons from a Year of Building with AI Systems

GenAI

LLMs

Findings from a three-hour livestream and two podcast episodes about lessons from building real-world applications on top of LLMs.

Hugo Bowne-Anderson

Fine-Tuning GPT-2 for Spam Classification: A Live Coding Session with Sebastian Raschka

GenAI

LLMs

Discover how to fine-tune a pre-trained GPT-2 model for spam classification in this step-by-step tutorial. Learn data preprocessing techniques, model architecture modification, and the training process using PyTorch. Understand key concepts like freezing layers, cross-entropy loss, and evaluation metrics. See how fine-tuning can significantly improve accuracy from 50% to 95%. Explore ideas for further optimization and adapting the approach to other text classification tasks like sentiment analysis and topic classification.

Hugo Bowne-Anderson and Sebastian Raschka

Developing and Training LLMs From Scratch

GenAI

LLMs

Learn the full lifecycle of building large language models (LLMs) from the ground up. Explore model architecture design, pre-training, fine-tuning, RLHF, and deployment techniques. Discover the skills and hardware needed to work with LLMs, and how to choose between prompt engineering, RAG, and fine-tuning for your use case. Stay up-to-date with cutting-edge research like LoRA, Mixture of Experts, and Direct Preference Optimization. Understand the implications of LLMs for community-driven platforms like Stack Overflow.

Hugo Bowne-Anderson and Sebastian Raschka

10 Brief Arguments for Local LLMs and AI

GenAI

LLMs

I recently hosted a session with Simon Willison for the “Mastering LLMs” conference, discussing local LLM usage with Simon’s CLI utility. Reflecting on why I prefer local LLMs, key benefits include data privacy, performance, cost efficiency, customization, offline capabilities, learning opportunities, open-source support, scalability, ethical considerations, and autonomy. Recommended tools for getting started include Ollama, Simon’s LLM CLI utility, LlamaFile, LM Studio, and Oobagooba’s text generation webUI.

Hugo Bowne-Anderson

Getting Started with Generative AI for Everyone

GenAI

LLMs

With the advent of Generative AI, more people than ever (technical and non-technical) can build interesting, fun, and productivity-increasing AI apps and workflows. In this post, we introduce the GenAI mindset of combining atomic units to build AI-powered apps and workflows

Hugo Bowne-Anderson and Jonathan Whitaker

Boost Your Productivity with ChatGPT in 2024: Simple Steps to Get Started

GenAI

LLMs

Have a 60-page PDF you don’t have time to read and want summarized? A meeting you need action items from? Or an online video you need transcribed and summarized? Would you like them to then generate a PDF, text file, or spreadsheet of the results? You can do all of this with ChatGPT.

Hugo Bowne-Anderson

ChatGPT, Author of The Quixote

GenAI

LLMs

In the era of generative AI, copyright won’t be enough. In fact, it’s the wrong place to look.

Hugo Bowne-Anderson

Lights, GenAI, Action – Building Systems with Generative Video

GenAI

Metaflow

We created a workflow to generate hundreds of videos with Stable Video Diffusion in one command.

What is Causal Inference?

Causal inference

An Introduction for Data Scientists and Machine Learning Engineers.