This week was pretty incredible. One of the coolest companies ever was finally rewarded (probably too well) for years of work.

Nvidia started off as a gaming company and now enables AI. Just a bunch of nerds who are making the world better. I dove into their 10-k to try and get a better understanding of why it works so well.

There were also some fantastic articles about why war sucks (as if we needed more reasons), and Adidas’ issues with Kanye West.

I also learned how persuasive repetition is, and the that this Natural Language Processing stuff is still really hard.

Here is the fitting video for this post!

Nvidia

“The computer industry is going through two simultaneous transitions — accelerated computing and generative AI,” said Jensen Huang, founder and CEO of NVIDIA. “A trillion dollars of installed global data center infrastructure will transition from general purpose to accelerated computing as companies race to apply generative AI into every product, service and business process.”

NVIDIA has Gross Margins of about 70% and is sitting on about $15 billion of cash. They have 2 billion in free cash flow, which is what they keep after all costs. NVIDIA is also one of the few companies that puts more into research than sales. They spend 27 cents of each dollar on research.

Newest product lines are helping companies build their own custom language models. This gets around the privacy issues that come with sending data to a OpenAI, which many businesses struggle with. Trust in AI is becoming one of the central barriers to adoption in enterprise settings.

NVIDIA is an important product for the AI ecosystem, because the GPUs they invented allow for parallelised computation. Additionally, they made parallel processing available for generalised computing. AI as we know it would not exist were it not for these innovations.

What’s more interesting is that they do not manufacture their own products! All they do is design them. This does bring another layer of risk into their supply chain. What’s interesting is that the company has to deal with Geopolitical risk with because of Taiwan SemiConductor and its place in a hectic global environment. NVIDIA has also made a commitment to source 100% of their electricity from renewable sources by 2025, and states that they are not on target to do so. The company is also so large, it probably cannot do any acquisitions. In 2022 regulators stopped it from acquiring ARM, another chip manufacturer. NVIDIA was also affected by the CHIPS bill, which took all its researchers out of China.

Articles

What a “war tax” means for the global economy

War is terrible. You know this as well as I do. Let’s add another reason to hate War. Since 1992 global spending on Murder Technology (AKA Defense) has decreased as a percentage of GDP. The United States wen tfrom spending 6% of its GDP down to 3%. That spending was distirubted into other parts of the economy, which help develop everything else. Last year defense spending increased 4%, defense stocks are outperforming the market and current commmitments will generate over $200 billion in additional defense spending every year.

What if we just spent that money elsewhere, and fixed the problems that plague every soceity across the globe? Say Education, Health, or Agriculture? It’s also quite difficult to understand how much spending is going on, and the levels of corruption in the industry due to its secretive nature.

While this does suck. One benefit we get is that current murder tech gets improved and becomes useful to consumers down the road. Examples include GPS, 5G networks, semi-conductors and more. We also see that as technology improves, defense spending can deflate as shown by spending patterns since the 1960s and 70s.

Putting Cows in Forests Could Prevent Heat-Related Losses

Silvopasture. We’ve spoken about this before. As the climate changes and we see higher rates of heat waves, the livestock industry should expect losses. Cattle quickly fall victim to heat stress and die at high temperatures (Read this article I wrote last year about the topic).

When heat stressed cows eat less, produce less and eventually die. This is sad!

The solution is Silvopasture, a technique where cattle graze among trees and can relax in the shade. The farming method also creates better soil, protects fields from floods and help prevent drought.

Adidas After Yeezy

When Kanye West went on an anti semitic tear last fall, he tore his whole life apart. Adidas, one of his main corporate sponsors, was slower to fire him than the other companies with whom he had strong relationships.

The big issue? Kanye was basically keeping the company afloat on his own. This article is an awesome deep dive into the whole saga.

Low-Carbon Concrete Could Bring a ‘Radical Change’ to the Industry

Cement production is responsible for about 7% of global carbon emissions. It’s also the most expensive piece of manufacturing concrete.

This article details a company, CarbonBuilt, which has created a production technique that uses carbon to strengthen the cement. Because the company is able to sell carbon removal credits, and the economies of scale that come from mass production their cost will continue to diminish. This shows that the right environmentally friendly products can really make a difference.

What I learned

Repetition is persuasive

I had a personal life situation this week. I was trying to make plans with someone and they continually turned my idea down. I was upset, because I knew that this plan was awesome.

Earlier in the week I heard this line. Repetition is persuasive. Not repeating yourself, but repeating an idea. Giving it color. Showing that there are multiple dimensions to it. The more we do this, the more persuasive our idea becomes. Repetition in language gives our ideas rhythm, and texture. It allows us to shape our idea to the audience we are looking to reach.

This works because our brains our lazy and love the past of least resistance. Daniel Kahnemann had this to say about repetition

All in all. It worked. We solved the issue. It worked. Just because I rephrased something a million ways until it worked. This is called the exposure effect.

It’s really hard to fine tune a Llama

Facebook has an OpenSource Language Model they released. It is competitive with OpenAI’s ChatGPT models and can be run on commodity hardware. I tried to find a way to build my own version of it that was going to sound like Homer Simpson. That didn’t work. Open Source models are still quite difficult to work with. That being said, they will be critical to the future development of AI technologies.

Llama in particular is becoming cheaper and cheaper to tune, because of things like LORA. As fine tuning costs decrease, dependence on companies like OpenAI will go down.

Projects / What I’m going to do this week

• Write this newsletter

• Find another client for AI in Education

• Put together a warm up plan for the basketball camp I am coaching