Try a walking desk while studying ML or working on your projects! https://ocdevel.com/walk
Show notes: https://ocdevel.com/mlg/mla-22
Tools discussed:
Other:
Boost programming productivity by acting as a pair programming partner. Groups these tools into three categories:
• Hands-Off Tools: These include solutions that work on fixed monthly fees and require minimal user intervention. GitHub Copilot started with simple tab completions and now offers an agent mode similar to Cursor, which stands out for its advanced codebase indexing and intelligent file searching. Windsurf is noted for its simplicity—accepting prompts and performing automated edits—but some users report performance throttling after prolonged use.
• Hands-On Tools: Aider is presented as a command-line utility that demands configuration and user involvement. It allows developers to specify files and settings, and it efficiently manages token usage by sending prompts in diff format. Aider also implements an “architect versus edit” approach: a reasoning model (such as DeepSeek R1) first outlines a sequence of changes, then an editor model (like Claude 3.5 Sonnet) produces precise code edits. This dual-model strategy enhances accuracy and reduces token costs, especially for complex tasks.
• Intermediate Power Tools: Open-source tools such as Cline and its more advanced fork, RooCode, require users to supply their own API keys and pay per token. These tools offer robust, agentic features, including codebase indexing, file editing, and even browser automation. RooCode stands out with its ability to autonomously expand functionality through integrations (for example, managing cloud resources or querying issue trackers), making it particularly attractive for tinkerers and power users.
A decision framework is suggested: for those new to AI coding assistants or with limited budgets, starting with Cursor (or cautiously exploring Copilot’s new features) is recommended. For developers who want to customize their workflow and dive deep into the tooling, RooCode or Cline offer greater control—always paired with Aider for precise and token-efficient code edits.
Also reviews model performance using a coding benchmark leaderboard that updates frequently. The current top-performing combination uses DeepSeek R1 as the architect and Claude 3.5 Sonnet as the editor, with alternatives such as OpenAI’s O1 and O3 Mini available. Tools like Open Router are mentioned as a way to consolidate API key management and reduce token costs.
Try a walking desk while studying ML or working on your projects! https://ocdevel.com/walk
Show notes: https://ocdevel.com/mlg/33
3Blue1Brown videos: https://3blue1brown.com/
Background & Motivation:
Core Architecture:
Self-Attention Mechanism:
Masking:
Feed-Forward Networks (MLPs):
Residual Connections & Normalization:
Scalability & Efficiency Considerations:
Training Paradigms & Emergent Properties:
Interpretability & Knowledge Distribution:
Try a walking desk while studying ML or working on your projects! https://ocdevel.com/walk
Discussing Databricks with Ming Chang from Raybeam (part of DEPT®)
Try a walking desk while studying ML or working on your projects! https://ocdevel.com/walk
Conversation with Dirk-Jan Kubeflow (vs cloud native solutions like SageMaker)
Dirk-Jan Verdoorn - Data Scientist at Dept Agency
Kubeflow. (From the website:) The Machine Learning Toolkit for Kubernetes. The Kubeflow project is dedicated to making deployments of machine learning (ML) workflows on Kubernetes simple, portable and scalable. Our goal is not to recreate other services, but to provide a straightforward way to deploy best-of-breed open-source systems for ML to diverse infrastructures. Anywhere you are running Kubernetes, you should be able to run Kubeflow.
TensorFlow Extended (TFX). If using TensorFlow with Kubeflow, combine with TFX for maximum power. (From the website:) TensorFlow Extended (TFX) is an end-to-end platform for deploying production ML pipelines. When you're ready to move your models from research to production, use TFX to create and manage a production pipeline.
Alternatives:
Try a walking desk while studying ML or working on your projects! https://ocdevel.com/walk
Chatting with co-workers about the role of DevOps in a machine learning engineer's life
Expert coworkers at Dept
Devops tools
Pictures (funny and serious)
Try a walking desk while studying ML or working on your projects! https://ocdevel.com/walk
(Optional episode) just showcasing a cool application using machine learning
Dept uses Descript for some of their podcasting. I'm using it like a maniac, I think they're surprised at how into it I am. Check out the transcript & see how it performed.
Try a walking desk while studying ML or working on your projects!
Show notes: ocdevel.com/mlg/mla-17
Developing on AWS first (SageMaker or other)
Consider developing against AWS as your local development environment, rather than only your cloud deployment environment. Solutions:
Connect to deployed infrastructure via Client VPN
Infrastructure as Code
Try a walking desk while studying ML or working on your projects!
Part 2 of deploying your ML models to the cloud with SageMaker (MLOps)
MLOps is deploying your ML models to the cloud. See MadeWithML for an overview of tooling (also generally a great ML educational run-down.)
Try a walking desk while studying ML or working on your projects!
Show notes Part 1 of deploying your ML models to the cloud with SageMaker (MLOps)
MLOps is deploying your ML models to the cloud. See MadeWithML for an overview of tooling (also generally a great ML educational run-down.)
And I forgot to mention JumpStart, I'll mention next time.
Try a walking desk while studying ML or working on your projects!
Server-side ML. Training & hosting for inference, with a goal towards serverless. AWS SageMaker, Batch, Lambda, EFS, Cortex.dev
Try a walking desk while studying ML or working on your projects!
Client, server, database, etc.