Set up a Private ChatGPT with Open WebUI and Gemma 4 With a few simple steps using Docker Desktop and the Open WebUI for Docker Desktop extension

This tutorial is also available on Medium:

medium.com/p/201430bff642

#356 The Forecast for Time Series Forecasts with Rami Krispin, Senior Manager of Data Science, Apple YouTube video by DataCamp

I had the pleasure of talking about my favorite topic - forecasting with Richard Cotton at the DataCamp podcast:
www.youtube.com/watch?v=N0QO...

#forecasting #ai

This is setting is simple and no prior Docker knowledge is needed, all done via the Docker Dashboard

Set up a Private ChatGPT with Open WebUI and Gemma 4 With a few simple steps using Docker Desktop and the Open WebUI for Docker Desktop extension

Setting up a private ChatGPT-like with Open WebUI, Gemma 4, and Docker with a few simple steps👇
theaiops.substack.com/p/set-up-a-p...

#ai #docker #opensource

This will force Anthropic/OpenAI etc., to reduce the token fees.

Long-term, I believe open-source models will win the race.

Now, the question is - can open-source models replace proprietary models such as Opus or GPT?

Probably not in the short term. But you can definitely offset a lot of the regular work to local models and send complex requests to the proprietary models

So, where is the arbitrage opportunity?
Companies will have to make a call between spending $20k/year (or more) on tokens, or buying a MacBook Pro (~$5k) or a powerful Mac Studio (~$9k). When amortizing the cost over multiple years, the cost is even smaller.

The game shifted from how strong your GPU is to how smart your hardware is and to the size of your memory. This is where the Macs come into the picture. The recent MacBook Pro M5 with 128Gb RAM can handle models such as Gemma 4 with 31B and Qwen3.5 with 35 B parameters

I remember trying to run a model I downloaded from Hugging Face on my M1 machine two years ago, and it was very, but very slow.

During the past two years, two things changed:
- LLMs become more efficient due to solutions such as llama.cpp
- Smaller models have higher performance and capabilities

Until recently, running an LLM locally was neither efficient nor practical, as it required a fancy (and expensive) NVIDIA GPU box under your desk.

The token economy went out of control, and there is a huge arbitrage opportunity. The math here is simple:
The token budget per employee is about $100/day
If, on average, there are 220 working days per year, that is a yearly token budget of ~ $20k per employee

Uber CTO says AI spending plans fall short as tools like Claude Code drive costs up Uber's AI strategy hits financial roadblock as costs surge

Yes, I am biased, but this is why, in my opinion, Apple won the AI race without being part of it 🧵👇🏼

"It's the economy, stupid."

www.indiatoday.in/technology/s...

Forecasting with Augmented Dynamic Adaptive Model, Language Modeling from Scratch | Issue 84 A weekly curated update on data science and engineering topics and resources.

My weekly newsletter is out! 🗞️

This week's agenda:
🔹 Open Source of the Week - The Smooth project
🔹 New learning resources
🔹 Book of the week - Forecasting and Analytics with the Augmented Dynamic Adaptive Model (ADAM) by Ivan Svetunkov

substack.com/home/post/p-...

#rstats #python #forecasting

Building a Personal AI Assistant with OpenClaw & Docker Model Runner | Hands-On Workshop Learn How to Build and Run AI Agents Locally with OpenClaw

I am going to run a hands-on live course with Packt about building a personal AI assistant with OpenClaw. More details 👇🏼
ramikrispin.substack.com/p/building-a...

#ai

If you had asked me a few years ago, what would have happened first - me talking to a bot in the morning before leaving for work, and assigning it tasks for the day, or having flying cars, I would say that I prefer the car in white pearl.

Also, me this morning 👇🏼

#ai

Thank you!

The Gemma 4 Project, Cloud DevOps Engineer's Guide, MIT & Stanford New Courses | Issue 83 A weekly curated update on data science and engineering topics and resources.

Issue 83 🗞️ is out 👇🏼

🔹 The Gemma 4 LLM
🔹 New learning resources
🔹 Book of the week - The Cloud DevOps Engineer's Guide

ramikrispin.substack.com/p/the-gemma-...

#ai #datascience #llm

I just installed it 😅

When your AI agent has passive aggressive attitude 🤪

Gemma 4 Just Got Way More Powerful with Open WebUI YouTube video by Teacher's Tech

This is super cool - setting your own personal assistant AI running locally with Gemma 4 and Open WebUI 👇

www.youtube.com/watch?v=aQ1r...

This includes plugging and querying your documentation privately locally.

#ai

Lec 01. Introduction to Deep Learning YouTube video by MIT OpenCourseWare

MIT has released a new introductory course for deep learning. This course covers the fundamentals of deep learning, including both theory and applications.

www.youtube.com/watch?v=6FkR...

#deeplearning #ai #datascience

SQL AI Agents in Production New Course with LinkedIn Learning

I released a new course last week focusing on SQL AI agents in production. More details are available in the following post:
ramikrispin.substack.com/p/sql-ai-age...

#ai #mlops #datascience #sql

The ggauto Project, Transformers in Action, MIT Deep Learning Course | Issue 82 A weekly curated update on data science and engineering topics and resources.

My weekly newsletter is out! 👇🏼

This week's agenda:
🔹 Open Source of the Week - the ggauto project
🔹 New learning resources
🔹 Book of the week - Transformers in Action

ramikrispin.substack.com/p/the-ggauto...

#ai #deeplearning #rstats #datascience #dataviz

#ai #sql #datascience | Rami Krispin I am super excited about the release of my new course with LinkedIn Learning - Build with AI: SQL AI Agents in Production! 🚀🚀🚀 This course, as the name implies, focuses on building SQL AI agents read...

Excited to release a new course focusing on SQL AI agents in production!

More details are available in the following post:
www.linkedin.com/posts/rami-k...

#ai #sql #datascience

The Trackio Project and Deep Learning with PyTorch | Issue 81 A weekly curated update on data science and engineering topics and resources.

My weekly newsletter is out! 👇🏼

This week's agenda:
🔹 Open Source of the Week - The Trackio project
🔹 New learning resources
🔹 Book of the week - Deep Learning with PyTorch, Second Edition by Luca Antiga, Eli Stevens, Howard Huang, and Thomas Viehmann

ramikrispin.substack.com/p/the-tracki...

#ai

This is super cool - running Claude Code from Telegram 👇🏼

github.com/anthropics/c...

#ai

Mastering NLP From Foundations to Agents - Second Edition, the Qlib Project | Issue 80 A weekly curated update on data science and engineering topics and resources.

My weekly newsletter is out! 👇🏼

This week's agenda:
🔹 Open Source of the Week - the Qlib project
🔹 New learning resources
🔹 Book of the week -Mastering NLP From Foundations to Agents by Lior Gazit and Meysam Ghaffari, Ph.D.

ramikrispin.substack.com/p/mastering-...

#ai #datascience

Save 95% of Your Tokens - OpenClaw Full Tutorial YouTube video by Tech With Tim

Optimize OpenClaw 👇🏼

The following tutorial by Tech with Tim focuses on optimization methods to reduce token usage. This includes caching and context-pruning methods, as well as audit token usage.

www.youtube.com/watch?v=lTXF...

#ai

Tell me that you live in the Bay Area without telling me 😅👇🏼

Applied Time Series Analysis for the Social Sciences, the Freestiler Project | Issue 79 A weekly curated update on data science and engineering topics and resources.

My weekly newsletter is out! 👇🏼

🔹 Open Source of the Week - The freestiler project
🔹 New learning resources
🔹 Book of the week - Applied Time Series Analysis for the Social Sciences by Regina M. Baker

ramikrispin.substack.com/p/applied-ti...

#ai #llm #datascience