Links Justin Sends 1/23

DeepSeek and Berkeley NovaSky crushing benchmarks, Hundreds of Billions invested in US AI, and BetterFutureLabs' AI Demo Night

It was a great first BetterFutureLabs AI Demo Night cohosted with TechUnited NJ yesterday!

AI Photo Photo, Jori, and BioProxal gave spectacular demos, my cofounder Aaron and I then wrapped up the night with a fireside chat about the trends we are seeing with new model training and fine-tuning techniques.

Excited to share I’ll be a on Microsoft’s AI Panel next week at their Times Square office! If you’re interested in joining, here’s the registration link: https://www.meetup.com/microsoft-reactor-new-york/events/305132896/?msockid=0748f77c09e66a5a1239e3e608166bc1

Now on to the links 👇️ 

🐋 DeepSeek-R1: Open source model on par with OpenAI’s o1 for a fraction of the cost $2.19 / million output tokens vs OpenAI’s $60.00 / million output tokens.

🌠 Stargate Project: Launched by SoftBank and OpenAI with support from Oracle and Nvidia, Stargate is looking to invest $500 billion into US AI infrastructure.

🏆️ AI Safety Executive Order Revoked: A major victory for the Open Source AI community! This controversial order, which introduced broad definitions for “dual use foundation models” and imposed strict regulatory requirements, faced significant backlash. Prominent tech leaders like Yann LeCun (Chief AI Scientist @ Meta), Marc Andreessen (a16z), Ben Horowitz (a16z), Arthur Mensch (CEO @ Mistral), Garry Tan (President & CEO @ Y Combinator), Clem Delangue (CEO @ Hugging Face), Aravind Srinivas (CEO @ Perplexity), and others voiced their opposition to this Executive Order in a public letter last year.

🧱 Clay Raises $40M: Leading AI powered GTM platform closed a massive funding round at a $1.25 Billion valuation.

🧑‍🚀 Open AI Operator: OpenAI launches ability for an AI model in ChatGPT to interact with a webpage to perform tasks.

⚖️ LinkedIn Sued: Customers claim LinkedIn “disclosed private messages to third parties without permission to train generative artificial intelligence models”

💥 Sky T1-32B Flash: Berkeley’s NovaSky lab has been crushing it these past 2 weeks. This week they introduced a flash variant of their Sky-T1 fine tuned model. Fine-tuning, and inference cost ratio to performance is wildly impressive.

 ByteDance Releases Doubao-1.5-pro: GPT 4o level performance while being exponentially cheaper.

What was the most exciting AI development you’ve learned this week?

-Justin

aka the guy with great AI links

Co-founder & Head of Technology @ BetterFutureLabs