Links Justin Sends 3/7

Testified about AI to the NJ Senate, Anthropic Raises 3.5 Billion, and many more updates

Happy Friday!

I had a great time testifying to the NJ State Senate about the intersection of energy and AI earlier this week. Covered an overview of what it takes to bring a model to life and the sectors of industry NJ can or should play a role.

Here’s a brief summary of what I testified πŸ‘‡οΈ 

Phase

Description

Relative Energy Needs

Pre-training

β€’ Gathering, structuring, and organizing data (human-generated and synthetic)

β€’ Preparing data pipelines for model training

β€’ Minimal energy consumption compared to later phases

β€’ GPU inference for synthetic data generation here is only a small fraction of the total energy

Training

β€’ Using massive GPU clusters to train model weights

β€’ Processing large volumes of training data

β€’ Extremely high energy demand due to tens of thousands (or over 100,000) of GPUs

β€’ Typically located in low-density areas with access to cheap, clean energy and affordable land

Inference

β€’ Deploying the trained model for on-demand use

β€’ Running predictions

β€’ Lower energy per operation than training, though overall consumption remains substantial when scaled

Happy to deep dive into more of these in a future post and reach out if you have any questions!

Now on to the links πŸ‘‡οΈ 

πŸ§’ New School Meets Old School: Young engineers are pitching VCs about acquiring old school businesses and applying AI to them to get leverage

🧠 Biological Computer: Cortical Labs launches a biological computer that combines stem cells with silicon

πŸ‰ Dragon Copilot: Microsoft launches drag and co-pilot for the clinician market

πŸ’°οΈ Anthropic Raises $3.5 Billion: Huge funding round at a $61.5 billion post money valuation. Hopefully this results in Anthropic diving more into reasoning and multimodal models.

πŸ¦™ New Models Available on Ollama: Microsoft's Phi4-mini as well as multimodal models like Granite3.2-vision available to run locally on your laptop using Ollama

🐲 Qualcomm βš”οΈ Nvidia: Qualcomm is taking aim at Nvidia with their new Dragonwing chips

πŸ•˜οΈ 96GB VRAM Nvidia 4090: Maybe the rumors are true. 96 GB of VRAM would provide a massive boost to teams who want to run AI locally and affordably

πŸŒ‰ Agent Exchange: Salesforce launches a marketplace to deploy AI agents.

Have a great weekend!

-Justin

aka the guy with great AI links

Co-founder & Head of Technology @ BetterFutureLabs