Not Quite Austerity, OpenAI

Not Quite Austerity

The Fed meeting yesterday left interest rates unchanged at 5.25-5.5%. Two more meetings in 2023 are scheduled for November 1st and December 13th. The most likely outcome is that we won’t have any more rate hikes, and the difference between another pause and another hike is marginal.

More remarkable is that long-duration treasuries are now at or near 20-year highs. TLT, the long-duration treasury ETF, is down -45% from the peak. AGG, the ETF that tracks the aggregate US investment grade bond market, is down -20% from the peak. The total losses in the global bond market due to rate hikes are staggering. I bring this up because the losses in the bond market took down Silicon Valley Bank and threatened the health of the regional banking sector. Bonds are less visible from a headline perspective but represent a far larger total market cap than the S&P 500.

While the Fed works to slow the economy to fight inflation, the US government can’t seem to resist introducing additional financial stimulus. The falling M2 money supply reversed course in March and is now slightly trending up. Total government debt reached a new all-time high of $33T, up 10% or +$3T since we started the rate hikes. The government spending deficit, which will be adjusted once taxes are paid on Oct 15, is currently at -$1.5T. That’s what I mean by “not quite austerity”. Hiking rates to slow the economy, then taking on more debt, and increasing the spending deficit is the whack-a-mole version of austerity. It also helps explain why US equity markets are back up and approaching new all-time highs.

Dall-E-3

Yesterday, OpenAI announced an updated release of Dall-E-3, the AI image generation tool. Two parts of this release stand out to me. The new Dall-E-3 will allow ChatGPT to communicate with images. I use ChatGPT every day, but I rarely use Dall-E. It would be nice if ChatGPT could also communicate via illustrations to illustrate certain concepts.

The second point has to do with prompt engineering. Knowing how to ask the right questions is a fundamentally important life skill. However, the idea that we must treat prompt engineering almost as a new language never sat well with me. We’ll see how well it works, but OpenAI’s latest update says that they have improved the model to interpret questions better and hopefully will move down the path of removing any need to learn the nuances of prompt engineering specific to a particular model. I’m glad to see OpenAI leading the way on this front.

The latest release will be available in October. It’s remarkable how much further ahead OpenAI is in the race to deploy consumer applications. I haven’t been able to get much use out of Google’s Bard or the other competitors, but I get massive value out of ChatGPT. If you have any new favorite AI tools, please let me know!