Tesla AI

Disclaimer: this is not investment advice. Before taking any investment decision, do your own research.

I’ve been looking at CoPilot and ChatGPT using GPT-4 lately. What I gather is that Tesla is uniquely positioned to step into the LLM market in a way that may be hugely disruptive to big business oriented tech companies (Microsoft, Oracle, Salesforce, AWS.) This are the main points in reasoning.

Language models are important for the problems Tesla is working on

Seamless verbal communication integrated with images and video would be a great addition to future FSM vehicles. Explaining to the car were we want to go and driving preferences instead of using a app UI will be more natural.

But the biggest use of LLM in Tesla would be with integrating robots into factories. Robots that can be explained what to do and iterate with humans on how to improve work will have an exponential productivity curve. It will likely require a very strong control overlay, but the potential es infinite.

Uniquely positioned for large model training

With D1 and Dojo, Tesla has been working on optimizing training cost for a while. Having a new area of AI that can exploit this advantage internally will add value to the investments already made and the ones they are currently working on. Tesla may be able to have LLM with GPT 4 like capabilities in 2–3 years but at a much lower cost.

Tesla OS and a different way of using software

The biggest potential is mixing knowledge models (LLM) with the experience building internal enterprise software using first principle reasoning which would work as the framework for something like Microsoft’s knowledge model that complements 365 CoPilot.

The biggest hurdle with software in most companies is balancing learning how to do what the software provides with adapting existing software to actual needs. Most of the time is wasted learning the capricious ways the software works while failing to make it do what the company wants.

There’s a very large potential in creating a knowledge-based platform that has general capabilities which companies can adapt to their needs with a short, business-oriented customization process where there’s no need for a large team of technical workers (i.e. software engineers.)

Most business software companies will hesitate to create such a platform because it will cannibalize their historical offerings, which creates a competitive advantage for a company, like Tesla, with a lead in AI and experience with enterprise software that is not yet in the market.

Improving the AI problem

Elon Musk talks frequently about his fear on how AI advances. Expanding Tesla’s AI problem set to LLM and other areas may allow him to have some extra influence on where the research goes. By competing with other companies it can help steer the path forward. I believe having large AI companies and non-profits competing and having conversations about how best to solve problems is better than just being paralyzed by possibilities. I see Musk expanding into more AI problems in the future, probably through Tesla.

Tesla Software?

I see a large opportunity for Tesla to get into the software solutions market in the next 2–5 years. This would be a completely separate business unit, so it’s a difficult decision to make for leadership. But Tesla is already doing the core work, so I don’t believe the risk is high and the potential is absurd.

Hope to see something in this direction in the next few years.