Why AI is getting Cheaper & a new challenge :)

Happy New Year!

As a note, in the last email I added the wrong link to a video I had made about RAG (a type of information retrieval). I hope none of you click on it because making videos is scary and really hard.

A new challenge

Over the weekend I programmed a quick telegram bot. Now I’m going to see if I can build a product and get some customers in 2 weeks, doing nights and weekends. Will Keep a running update, here and on linkedin :)

Why AI is cheaper

A few days ago I was having a conversation with a client. He was trying to understand why AI had become so much cheaper over the last few years. His company had built an AI initiative, which had failed because the costs had far outstripped his potential revenues. He (wisely) refused to make the same mistake twice.

After a long conversation, I figured out how to explain it to him. The answer came down to two simple factors; Infrastructure and Talent.

Just as a reminder, we're going to use the following as a definition for AI: Artificial intelligence is the ability of a system to think, learn, and react like humans without being told what to do.

Infrastructure

AI requires large datasets and high powered computer chips, both of which have become more available than ever before. AI was cost prohibitive for many potential users, because of the cost of infrastructure.

Several years ago, university labs and hobbyists realised that they could run AI code on GPUs, and make the training process go way faster.

GPUs are multi-taskers. They can make thousands of simultaneous calculations in parallel, which is perfect for AI. This allows them to be good at a wide variety of highly complex tasks, really quickly.

The GPU market is dominated by a single player (NVIDIA) who charge pretty much whatever they want for their chips. This has kept the price of GPUs quite highand making AI more expensive than it should be.

High GPU prices have catalysed the development of AI specific chips. These will make AI more energy efficient, faster and cheaper. As time goes on, the cost of training models will become negligible from a chip and energy perspective.

There is one important quirk to note about AI. Neural Network quality scales in proportion to the amount of data they have access to. So the most important piece of infrastructure is around the amount of data you can jam into the thing.

As data becomes more valuable, the fitting infrastructure around it will be a more

Talent

Until about 2012 the path forward was muddied. Then in 2012 lab at the University of Toronto won an important Computer Vision challenge. For the first time, someone had shown that deep learning was the best way forward for AI. And with that came a wave of education for aspiring talent.

The most important piece of developing a technology are the engineers behind it. Since 2012 the AI world has shifted towards deep learning as the next generation of software development. The smartest researchers & development talent has begun to transition to building AI vs. regular software development.

One of the critical pieces of this wave was Andrew Ng, a Stanford professor and the cofounder of Google Brain, who believes in education for all. He began teaching online courses (and started coursera), about Deep Learning. These courses were among the first MOOCs about AI, which were highly digestible and widely available.

As time goes on and more engineers realise that there is money to be made in the world of Deep Learning, more will join the field and continue to develop the field. This will continue to snowball until the field has reached a saturation point.

Why does this matter?

Because AI is going to end up everywhere. People who do not have a specialty or serious people skills are going to be caught off guard. Many of the jobs that needed to be done by 5 humans in the past can be done with 1 expert + and AI system.

Many of the menial jobs that aren't that exciting will be gone. Imagine the classic job in an accounting firm. Much of the work that is less skilled will simply be moved away from humans into the hands of a task specific AI.

However, there is one thing that AI will never be able to replace (maybe it will... who knows), human relationships. There is magnetism when you are in the same room as someone else. A certain something that cannot be replaced. That seems like the skill to cultivate.

Plus there are going to be millions of new opportunities created by AI. The invention of the camera didn't put painters out of business. It allowed them to move into expressive vs. representative art. What if everything becomes art? What if you begin going work for the sake of doing it? To me that sounds like a better world, but who knows.