Build your cost-free, offline AI tool

Demo! ... also, AlphaFold3 open-source, and a 1.7T parameter AI 🧨

Author

Martin Musiol
November 12, 2024 • Estimated Reading Time: 8 minutes

Hey there!

✅ AlphaFold 3 is now open-source!
✅ 1.7T parameters - the largest OS LLM. → Does size matter?
⬇ Demo: Building a secure, offline AI tool from scratch in under 35 minutes. ⬇

🚨 We launched a premium version of the newsletter. Subscribing gives you 100% access to all content, exclusive hands-on demos, and an ad-free experience. 

With Premium, I’m ruthlessly sharing all my secret sauce and exclusive demos — so you can take it, use it, and benefit from it.

Make the smartest use of your time with Brilliant. Thousands of bite-sized, interactive lessons make it easy to sharpen your mind while building real knowledge in everything from math and data analysis to programming, AI, and beyond.

(✨ If you don’t want ads like these, Premium is the solution. It is like you are buying me a Starbucks Iced Honey Apple Almondmilk Flat White a month.)

own animation

AlphaFold 3 is now open-source (OS)!

Don’t underestimate this news. The future will single out AlphaFold.

Abstract 3D visualization of protein folding structures and DNA helix in vibrant colors, featuring intertwined teal protein strands and a prominent magenta DNA double helix. The image showcases molecular biology concepts, representing protein interactions, gene expression, and structural biology. The background gradient features soft blue and pink hues, emphasizing the intricate details of the molecular components. This image is relevant for topics related to biotechnology, genomics, protein modeling, AI in drug discovery, and computational biology.

While DNA is the code of life, proteins are its substrate, the hardware necessary to give the code any meaning. Biotechnology, without the understanding and control of protein, was doomed to a deadlock.

We now know how the linear code works, moving from linear to 3D, from the instructions to function was the missing piece.

Protein design changes everything—health, materials, energy, and more.

Don't be misled by what people are or aren’t noticing. What grabs attention today often fades, while history will remember the transformative.

(If you make it run, reply, and I will feature your progress in the newsletter. 🙌 )

1.7 Trillion parameters - the largest open-source (OS) language model. And: Does size matter?

Near Protocol co-founder Illia Polosukhin: Professional setting with a man sitting comfortably at a glass table in a modern office environment, wearing a black hoodie and smiling at the camera. Behind him, a large window provides a view of a cityscape with high-rise buildings under a clear blue sky. The interior features light gray wall panels, a cream-colored armchair, and a laptop open on the glass table, accompanied by a smartphone and a computer mouse. The image conveys a casual yet professional atmosphere, likely for a business meeting or interview. Ideal for content related to tech, business leadership, or remote work environments.

Near Protocol co-founder Illia Polosukhin. Source: Cointelegraph

Near Protocol aims to build the world’s largest OS AI model with 1.4 trillion parameters.

The project will use crowdsourced R&D via the Near AI Research hub, starting with a 500M model and scaling up. Estimated costs are $160M, funded by token sales and repayments from AI inferences.

They explore decentralized computing as a solution.

Why have larger and larger AI models?

According to AI experts, such as Ilya Sutskever, or Sam Altman (I have also said this repeatedly in my book), our algorithms are good enough, and by scaling them up, they will have more emergent capabilities, even reaching/bringing us to AGI.

Dario Amodei said in yesterday's podcast with Lex Friedman (which I recommend listening to) that scaling data, compute, and model size will lead to Superintelligence.

A major bombshell dropped during the YCombinator interview when Sam Altman casually mentioned that what excites him about 2025 is AGI.

He said it just like that, almost offhandedly. → Interview (min 46:12)

Sam Altman: Close-up image of a man seated in an interview setting, wearing a light blue shirt. He appears to be engaged in conversation, with a calm and focused expression. The background is softly blurred, featuring a dark ambiance with a hint of greenery from indoor plants, creating a casual and relaxed atmosphere. The subject is sitting on a bright orange armchair, adding contrast to the scene. This image could be associated with a discussion on technology, AI, or leadership in a modern, professional context.

In the meantime, a new report (The Information) came out that describes how OpenAI’s next flagship model might not represent as giant a leap forward as its predecessors.

The report also states their coping mechanisms for it. So, it’s all not that easy and straightforward, but it is improbable that the undeniable progress will halt.

Learn AI in 5 Minutes a Day

AI Tool Report is one of the fastest-growing and most respected newsletters in the world, with over 550,000 readers from companies like OpenAI, Nvidia, Meta, Microsoft, and more.

Our research team spends hundreds of hours a week summarizing the latest news, and finding you the best opportunities to save time and earn more using AI.

(✨ If you don’t want ads like these, Premium is the solution. It is like you are buying me a Starbucks Iced Honey Apple Almondmilk Flat White a month.)

own

Demo: How to build a secure, offline AI tool from scratch in under 35 min.

In this short series, I share my frameworks for building a secure offline AI tool that can analyze your bank statements. By following along, you’ll gain the skills to create distributable offline tools that integrate OS LMs.

You don’t need to be a coding expert — just have a clear plan and know how to ask your coding expert (AI) the right questions to guide you through the process.

Four short videos

1/4:

Watch videos 2, 3, and 4 with Premium. Next week, there will be another hands-on video!

That’s a wrap! I hope you enjoyed it.

Martin

Our webpage