As the adoption of AI technology increases so too does the number of IT experts adapting the tech to work better for themselves. Often when projects are made open source it creates communities of extremely passionate individuals who will spend countless hours diving deep into the structure of the technology to try and add their improvements. Often the most popular improvements seen are in the efficiencies of the tech, making it run faster or making it to be able to run on less resource-hungry equipment.
Sure there are always people trying to overclock hardware to get bragging rights of the highest benchmark but what is often more impressive are people who manage to get applications to work on hardware that was never designed for it. Look at the Reddit thread called “Will it run DOOM?” where people try to get doom to run on the most random hardware it was never designed to run on, some good examples are oscilloscopes, smart watches, and a graphics calculator powered by 50Kgs of potatoes.
The great thing about passionate people is they are often very clever, have time to spend on passion projects and they normally look at things from different perspectives. Most AI engineers are very good at their jobs but if you want to make something more efficient let the world look at your code and you will be surprised at what comes back. We have already seen 10x improvements of many AI language models over the months so this means the cost of running such algorithms is now a fraction of what it was and rather than it needing a supercomputer to run on it can start to get down to more manageable higher-end computing.
Matt from Stand-up Maths has a brilliant video about the improvements of code titled “Someone improved my code by 40,832,277,770%” that is well worth a watch as he explains how the code alone can improve by orders of magnitude. A 10x improvement is just the beginning and over time we will see these models get better and better. We have already seen people be able to run ChatGPT on Raspberry Pi, though not very fast, it is a testament to where things can be headed.
As AI becomes more mainstream it is logical to think that some AI will be run on-premises and some will still be run on the cloud. A Tesla vehicle requires the camera’s machine learning algorithms to be running inside the car for real-time processing but they do have expensive computers inside their vehicles specifically for the autopilot and other safety features. It’s not absurd to think that we may unlock the power of our parked cars to run advanced computing.
I don’t think we are going to be running AI in our watches but instead, we can connect to our local processing power and still process it on premises. My gaming computer has 240 Tensor Cores inside the graphics processing unit (GPU) but a Tesla has 640 Tensor Cores, meaning a Tesla V100 is the world’s first GPU to break the 100 teraFLOPS. A large language model could quite happily sit in my car and process things for me when I’m not using them should I not want to use the cloud to crunch the numbers or should I not trust the data being uploaded onto the cloud.
I suppose it comes down to cost, convenience & security. The flexibility of having an alternative to using the cloud is good, at the moment AI is quite cheap on the cloud but the moment people start throwing large data sets onto the cloud the prices are going to go up. We are going to see a paradigm shift soon so we need to be prepared to see the price structure change.
Tesla is still a sleeping giant sitting on a lot of unused technology and even more data. In time we will see Tesla use 2 way charging where you can use your car as a battery to run your house and feed power back to the grid & they have the ability to map the world better than Google. The fact that every Tesla has about 8 cameras plus the most advanced GPU on the market cant be lost on Tesla, they are easily poised to create the best mapping software in the world by going a step beyond and creating a digital twin of the earth so detailed that every GIS (geographic information system) company will be subscribing to their data. And what's great is they will be getting all the data for free and the car owners will even process it for them and even pay for the electricity to do so. If this isn’t your signal to look at buying Tesla stock, I'm not sure what is.
