The AI Daily Brief Newsletter is Back - AIDB Night Desk

Back in 2023 when the podcast started, there was a companion newsletter. The problem was there were about a trillion AI newsletters at the time and I wanted to focus on making the best podcast possible, so I let the newsletter drop.

Now, two years on, AI Daily Brief is the most popular AI podcast and I’m excited to start experimenting to see if we can figure out something value add with the newsletter again.

This first experiment is Night Desk. Every other newsletter comes out in the morning, because that’s what best practices say to do. But AIDB starts the planning process late evening on the East Coast US, so maybe our value could be giving you tomorrow’s news today.

We’re going to do these Monday-Thur for the next couple weeks before the holiday and then assess what’s next.

For now, enjoy tomorrow’s news today.

-NLW

Tomorrow’s Main Episode: New AI Politics at Home and Abroad

AI politics/geopolitics have to be pretty punchy to make the main. Most of you listeners are practical and business-minded and want to know what is going to impact how you do what you do (and maybe even what you will do next).

But a massive shift in US chip policy towards China and a majorly contentious AI regulatory executive order clear the bar.

We’re not just talking about reversing the most recent ban of selling the already-nerfed H20 chips. This allows for the sale of H200’s — a chip more advanced than the workhorse H100’s that trained many of the models you use every day.

On the one hand, the compute constraint in China has been a constraint that bred innovation. In their recent study of 100T tokens, OpenRouter found that Chinese models had gone from around ~1% of tokens to as many as 30% in a given week. What’s more, they’re increasingly used for “serious” b2b type workloads. All this without the benefit of advanced chips.

What’s more, the CCP has been loud about using this as a chance to shore up their domestic chip industry. But still, you gotta think that this news opens up some serious new possibilities for Chinese AI. It also has downstream effects for US-Gulf relations, as much of that collaboration has been shaped by policy towards China, as well.

And then there’s this forthcoming executive order: ONE RULEBOOK FOR AI (as AI czar David Sacks tweeted)

All government actions are a combination of policy and politics and this one has a mess of debate around both. The policy debate has safety folks jockeying with tech pro-innovation folks with another side of the global AI race folks thrown in.

In this case though, the politics might be the even bigger deal, as we had heard that last week, this was scrapped due to internal disagreements within the GOP. We’re starting to see the anti-AI side of the right come out in an increasingly vociferous fashion. The new EO plans have Florida Governor Ron DeSantis out here quoting James Madison.

Feels like a gauntlet-laying moment from the Trump White House. With they line up behind the new policies, or will they make their break with Trump on AI official?

Tomorrow’s Headlines

• Google AI glasses coming next year

• ChatGPT usage stickier than Gemini?

• IBM to buy data streaming platform confluent for $11B

• Cohere getting software margins on enterprise AI biz

Operator Tip

Turns out, prompting the AI to imagine itself as a persona doesn’t meaningfully improve results in the aggregate.

Today’s Episode

Feedback