I’ve been thinking about the future of machine learning after the presentation of GPT-4, a tool that takes advantage of the resources made available to OpenAI by Microsoft to offer far superior capabilities: generation of content up to eight times longer (limit of 25,000 words), knowledge of more languages, the ability to understand images and faster and more accurate responses.
GPT-4 marks the end of the idea of an open tool to improve knowledge and refine the uses of machine learning, to be replaced by the Silicon Valley model of enhancing a specific service based onLarge Language Models or LLMs. In reality, LLMs are simply one of the many varieties of machine learning, and not necessarily the most interesting, but OpenAI has managed to capture it as the first mass-use application and turn it into a viral phenomenon.
What we now have to understand is that to keep on applying LLMs to everything is a waste of time, and instead we must focus on the many other types of algorithms available to create advanced automation models for all types of companies. As I said in 2016 in an article on Forbes, “it is important to understand that it doesn’t matter what you do: the key to your future will be in your company’s ability to become a constant generator of data from all its activities, data that feeds the learning of your algorithms.”
It is time to think about how the ability of companies to do things better than their competitors will depend on how good they are at generating that data, treating it properly and using it to create their own algorithms, something that makes much more sense than trying to use a one-size-fits-all “uber-algorithm” (but that when I asked for my biography, had “married” me to four different women, none of which was mine 🙂
Understanding how an algorithm works and the effect of its training is important. This is why OpenAI opened ChatGPT to the public in December: to incorporate into its training the conversations of all users who started interacting with it, including those who thought they were pulling its leg or driving it crazy. In other words, when I discover that it has no idea about my personal life and tries instead to invent something that looks convincing and correct it, explaining who my wife really is, it not only incorporates it into its training, but also goes on to use it with other users who ask it about me. This is what ChatGPT says:
It’s simply the textbook application of the Silicon Valley philosophy and model to machine learning: first, I’m going to raise venture capital so I can grow very fast and create traction and momentum. And second, I’m going to present a product that is far from finished or perfect, but that users are going to help me improve.
There are other examples. Tesla uses all the information that its vehicles generate — unless the user says they don’t want to share it — which allows it to take into account the myriad situations that can occur on the road. Improving driving safety five- or ten-fold compared to what happens when a human being drives is impressive, although getting a safety agency to approve it as a safer driving method will surely take a couple of years of intense study and lobbying (and even more so in the European Union). But the idea is also to have more unexpected situations under control and more experience in managing them, something that can only be achieved with more training, and to consider reaching the objective of being twenty times safer than a human driver in an equivalent circumstance.
To do that, Tesla needs data: millions of car rides in real circumstances with real drivers sharing them with Tesla in real time, coupled with many tens of millions more in simulated circumstances. At the moment, only Tesla is in a position to generate that data and is experienced enough to process it in real time and improve its algorithms. Other competitors are focused on other areas: Waymo is providing navigation only in areas that have been carefully micro-mapped. Mercedes is improving navigation only when the vehicle is driving behind another vehicle. Achieving full autonomous driving and doing it solely with cameras requires algorithmic work that Tesla’s competitors are many years away from achieving. The key, as we have seen on many other occasions, is in understanding the economies of scale aspect of algorithm training. Economies of scale, once again.
Think about it, and design a machine learning strategy for your company in those terms. Because if you don’t have one, there’s a competitor out there who does.
—
This post was previously published on Enrique Dans’ blog.
***
Join The Good Men Project as a Premium Member today.
All Premium Members get to view The Good Men Project with NO ADS.
A $50 annual membership gives you an all access pass. You can be a part of every call, group, class and community.
A $25 annual membership gives you access to one class, one Social Interest group and our online communities.
A $12 annual membership gives you access to our Friday calls with the publisher, our online community.
Register New Account
Need more info? A complete list of benefits is here.
—
Photo credit: iStock
