Weather, GDP, Robots

Weather

Google’s DeepMind had two major releases this week. The first focuses on weather predictions. I wrote recently that I don’t think we are close to running out of data to train AI models, and the weather is a perfect example. This latest model was trained on 40 years of wind speed, temperature and pressure at various altitudes. Weather is an ideal data set for AI, given it’s extremely large and complex, and because it’s physics-based, there are non-random patterns that can be discovered and replicated. This model can generate 15-day prediction windows in roughly 8 minutes that outperform current models. Better forecasts are extremely valuable as they can be used for disaster preparedness or better planning for heating and cooling electricity needs.

The 2nd major release was Genie-2, a highly detailed generative world builder for video games. The post has a ton of details for those who are interested. This model builds entire games from a single image. Consider how video games will change in the next few years if users can design them from their favorite landscapes, characters and prompts. It could also shorten the game development timeline tremendously.

My biggest complaint with these releases is that Google does not release these tools for people to try them. If it weren’t for ChatGPT, Google may have never released its Gemini LLM model. I’m certain Genie-2 would be very popular if people were allowed to try it. Either way, it’s impressive and a view into the future.

Daylight Savings Time (DST)

There is a reference circulating to a 2016 paper published by JPM Chase that shows daylight savings time negatively impacts spending patterns. There have also been studies showing that the time change leads to more accidents and cardiac events. Elon and Vivek may be the ones to end daylight savings time for good, as it’s recently become one of the targets in their DOGE pursuits.

GDP

In 2020, the overwhelming COVID stimulus was the primary driver of the stock market. In the face of the upcoming DOGE program to cut government spending, I was curious to see how government spending aligns with current GDP growth.

GDP has grown 65% over the past 10 years, from $17.8T to $29.3T. Government spending is one of the four main categories tracked by GDP. You can see the full table here. Government expenditures and investment grew from $3.19T to $5.03T, or 58%, over the same 10-year period. At $5T, the government category is 17% of GDP. Given how government spending is discussed in the media, I would have expected that it would be higher than GDP growth, but it has underpaced GDP growth.

I was also a bit surprised to see that the Fed’s balance sheet has been gradually reduced by $2T back to 2020 levels. You can see the Fed balance sheet here. Total government debt is still at all-time highs, and the M2 (the money supply) continues to climb after an uncharacteristic dip.

All of this is a moderately positive sign for markets. The GDP figures tell me the economy is less reliant on government spending than the media has made it seem. Surprisingly, the Fed’s balance sheet has shrunk this much in recent years without impacting markets. I still view M2 as a critical metric. If the money supply stops growing for whatever reason, that will almost certainly align with market prices.

World Robot Population

The IFR, the International Federation of Robotics, tracks the number of commercial robots operating in each country updated annually. There are now 4.28 million robots operating globally, up 10% from the prior year. 40% are in China (1.755m) and 9% are in the US (382k). The highest robot-per-employee ratio is in South Korea, with 1,102 robots per 10,000 employees or over 10%. South Korea also has the lowest birth rate in the developed world and truly alarming population decline statistics. Those stats are not related, but we tend to anthropomorphize robots so it seems relevant to include. Robots are just another form of automation, whether they have arms or not.

4.28 million global robots seems low. Robots are cheaper and easier to program than pre-ChatGPT and pre-COVID. There’s a recent viral video of robots folding towels that gives you a sense of how ubiquitous these will become in our lifetime. These towel robots are remotely controlled, but that function can be fully automated. Robots will replace jobs, but they will also run 24/7 at a marginal cost approaching the cost of electricity, which leads to superabundance.