We spoke about the magnificent 7 last month, but perhaps it should really be called the ‘Magnificent AI’. Artificial intelligence (AI) has made some incredible breakthroughs in the last few years, with the pace of change accelerating at warp speed. The step change has largely been the ability to train AI models with extremely large data sets, where the more training data you give it, the better the AI performs.
If you want to understand AI, be prepared to go down a very deep rabbit hole. I have been following this space for years but even experts in the field now agree that the pace of change is taking even them by surprise. If you want to get up to speed, there are a couple of excellent primers on the AI revolution. The first is called ‘The Ai Revolution: The Road to Superintelligence”, written by Tim Urban in 2015 and, more recently “AI-pocalyse Now” by Matt Barrie. I have included links at the end of the newsletter, and they are well worth your time to read.
A major breakthrough in AI occurred in 2017 with a paper titled “Attention is all you need”, which lead to the transformer model and the advent of the LLM (large language models). Designed as next word predictors, when you feed the AI models a bunch of text as training data, and then give in some text input, it will be able to predict the next most likely word. Initially, you could tell the output of these models was clearly written by a computer, but, as they increased the amount of training data, it got better and better and at a non-linear rate.
Something truly incredible happens when you step up the training data in orders of magnitude, which also takes orders of magnitude more computing power. The AI starts getting really good at completing sentences. And if you keep increasing the training data by, for example, using most of the text available on the internet, “suddenly black juju magic starts to happen. It can talk to you in Swahili. It can come in the top decile in SAT math and the Uniform Bar Exam for lawyers. You can tell it to write the next Harry Potter book and it will do it. It can do two-dimensional pattern recognition, balance a pencil on a robot’s finger and control an industrial HVAC air conditioning system.”1
The incredible thing is some amazing and completely unpredicted abilities just suddenly emerged from something that essentially was just designed as a next word predictor. Abilities such as being able to speak different languages and solve complex math problems. “Initially the AI can’t do math, we step up the model an order of magnitude in terms of training data and compute, it still can’t do math, we step up it up again, it can’t do math, we step it up again- bang, all of a sudden it can do university calculus!”1 It is fair to say that most of these abilities were completely unpredicted, even by the experts in the field.
One of the major goals in computing and AI has been the Turing test, a test of a machine’s ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human. We have now clearly sailed past this point, but what is clear is that the more computing power we throw at the problem, the better the AI will get. Hence the mad scramble for AI capable chips, with any company producing such chips trading at truly incredible multiples. Competition will surely come, and just this week Sam Altman of OpenAi is reportedly talking about raising $5-7 Trillion for AI Chip plants. Trillion, not billion…
Many of these AI stocks are clearly exhibiting bubble behavior reminiscent of the tech boom of the late 1990’s and early 2000s. At that time the Nasdaq index rallied 800% into a clear bubble, and then crashed over 78% from its peak in the following years. Clearly there are going to be winners in this AI space, but picking winners isn’t always easy. And even the winners might suffer serious pullbacks in price – In the dot com crash, Amazon’s share price fell 95% from its peak, with Jeff Bezos famously starting his 2000 investor letter with a single word “Ouch.”
Image: An example of AI generated art via Bing.com by the Author.
“Generate an image of AI stocks forming a massive stock market bubble with an oncoming meteor about to hit it”
Helios
I was hoping in this newsletter to be able to give you a full update on Helios, since it’s been 3 months since the AGM and the addition of the new Board members. While our internal research suggests things are progressing behind the scenes, the company still has not made any concrete announcements to the market. The only indication that there are likely to be major changes appears from their latest quarterly activity report, where they mention bringing in a farm-in joint venture partner to participate in the Presidio Oil and gas project. This is the first time they have talked about bringing in a partner, which would be a very welcome addition to de-risk the project, and, more importantly, to get things moving forward.
