Links Justin Sends 1/17

Big thanks to TechUnited:NJ and Choose New Jersey for hosting BetterFutureLabs along with other AI, Tech, and Government leaders at the Devils game last night, I had a great time!

BetterFutureLabs is also cohosting an AI demo night alongside TechUnited next Thursday at Stevens Institute of Technology — use the code BFL_GUESTS and register here for a complementary ticket (please only register if you are planning on attending).

Now on to the links you came here for 👇️ 

❓️ Biden Administration Launched New AI Regulation — Aims to restrict the use of advanced U.S. AI technologies by geopolitical adversaries through tiered export controls based on country risk levels, regulating access to GPU compute and AI models requiring more than 10^26 FLOPs for training.

Nvidia has strongly opposed this rule, with AI Policy Expert Dean Ball describing it as “an awful lot of authority for the American government to assert, with precisely no Congressional input, over hardware that, by and large, is not manufactured in the United States.“

I’m not best suited to comment on whether the tiered system or the placement of countries within those tiers is appropriate, as I’m far from an expert on geopolitics.

From a technical perspective, regulating the amount of FLOPs during training doesn’t necessarily prevent restricted tiers from accessing cutting-edge U.S. AI technology.

As we talked about in previous Links Justin Sends (New Years Edition, Christmas Edition, and the Sky-T1 model below) — there is a vast pivot in the direct correlation between model size & horsepower and capability with state of the art training and inference cognition techniques.

Therefore, a more effective approach might involve identifying specific capability benchmarks and applying targeted restrictions accordingly.

I’ll publish a more in-depth analysis on what this rule means for AI research and AI startups in the US when I can get through it — it’s a 168-page beast and I would rather finish rereading The Lord Of The Rings ⚔️ 🪄 for the millionth time.

In the meantime here’s a podcast overview I created of the regulation using AI, it’s a very digestible way of understanding the regulation’s high level impacts in under 20 minutes.

☄️ Sky-T1 — The Berkeley NovaSky lab trained a model for $450 that rivals OpenAI o1’s performance while being only 32B Parameters 🤯 — leveraging an internal chain of thought—a technique a part of what I’ve been calling inference-based cognition for the past 8/9 months.

⏰ ChatGPT Tasks — The most popular AI chat client now allows you to schedule reminders with AI — only available with the 4o model for now, interested to try this feature out.

💭 Launch of T3 Chat — ChatGPT might be the most popular AI chat client for now, but T3 looks very interesting — check out the demo.

🤖 Reddit Answers Launch — Interesting way to combine the power of AI with crowd sourced human content to answer questions — test it out here.

💉 21st.dev — An AI first, ShadCN on steroids — allows you to build amazing animations and components leveraging ShadCN with your favorite AI enabled code editor (I prefer Cursor).

🐁 Cursor Raises $105 million — While we are on the topic of Cursor, they just raised their Series B from A16z, Benchmark, and Thrive. I love using Cursor and am looking forward to the new features they can build with this traunch of funding.

Have a great weekend!

-Justin

aka the guy with great AI links

Co-founder & Head of Technology @ BetterFutureLabs

[email protected] | LinkedIn