#180: Relationships

S2N Spotlight

I had a bit of fun this morning creating a correlation study I have never done before. I look at the relationship with the S&P 500 on the vertical axis and the US dollar on the horizontal. By the way, its not an error. I had to do quite a bit of checking the S&P 500 and the US Dollar have no correlation, as in 0. Bitcoin also has no relationship with the US dollar, which is strange. It clearly behaves more like a risky asset. I will leave you to observe and draw your own insights.

S2N Observations

I like to get philosophical on a Friday.

I do a lot of work with AI LLM models such as Chat GPT, I tend to use Claude.ai more when it comes to coding than Chat GPT. Yesterday I listened to a fascinating interview with the founder and CEO of Claude on the Lex Fridman podcast. I am not recommending you spend 5.30 hrs listening to it but there was one thing Dario Amodei said that really struck me and made me think deeper on the subject.

Quick background Claude or Anthropic the company already have a valuation north of $40 billion for a 4 year old company. This guy is impressive he was a senior player at Open AI before starting Anthropic with his sister and another industry heavyweight. He really has a gift for explaining the complex. I can see why he is CEO and has managed to raise so much money.

One of his big breakthroughs in the development of Large Language Models that make them so smart, which seems so obvious now, was the following approach Dario implemented.

He said the co-founder of Open AI, Ilya Sutskever, used to say LLM models, love to learn. So Dario decided he would push the envelope and provide much more data than what was considered reasonable and provide the model with more node layers and see where it went. I am talking he introduced super scaling, and it worked. This was a major breakthrough.

To give you an idea of the scale we are talking, the worry now is that there is not enough data for the models to learn from. Essentially the models have learned almost the entire corpus of knowledge that exists. The new frontier is creating synthetic data for the models to learn from.

In essence the bottom line what I took away from the interview is that we are reaching a point possibly within 2 years that the level of smartness these models will possess makes them extremely vulnerable to bad actors. There is a worry amongst the major players about responsibility in the industry. There is a major push to make money from these models and recoup the huge investments being deployed, with this potential comes the need for some level of governance to prevent a situation where the Bots run the world.

Its Friday and I am feeling jaded so let me leave it there before I depress you and myself further.

During October gold was on course to record the best year of performance since 1982. We have slipped a little during November it will be interesting to see if gold gets back onto record pace.