Eric Schmidt is one of the foremost technology authorities there is. He was trained as a coder and eventually became the CEO of Google. The company had a successful run during his tenure, something we have to acknowledge even if one thinks they are the evil empire.
Schmidt has been making the rounds over the past few years, discussing what he is doing with technology and where he sees things going. He does have knowledge of what is going on "under the hood" with different models.
For this reason, he is a good guy to pay attention to. Like all of us, he is apt to be wrong on his forecasts. None of us have a crystal ball.
That said, he has an excellent grasp of trends, both from the 5,000 foot view along with the coding. His outlook is worth some considering to say the least.
Source
AI In A year
To me, this one was of the clearest, most accurate forecasts regarding LLMs I have seen. Whether it is correct or not remains to be seen. However, he has a clear vision where things are going.
Here is what he stated:
Very Large Context Window
He basically said we are dealing with a new form of short term memory. Remember, AI is actually aligned with compute more than the Internet. This is not networking but, rather, processing.
These very large context windows will allow for, according to Schmidt, a 20 volume series of books to be uploaded.
To say that is very powerful is an understatement.
AI Agents
This is something I discussed so I will not dive into it too deeply.
My view is AI agents are going to be task masters. They will fulfill different roles. Jensen Huang called them "skills".
The exponential impact will come when these agents are networked together. This will cause the new "compute" to have its "internet" moment.
Pictures billions of AI agents, interconnected and handle millions of different tasks.
Text-to-Action
This is probably the most impressive of all his forecasts.
Schmidt believes that, in roughly a year, we will simply use text to prompt actions, including coding. In other words, something that you would direct an human to do simply is prompted in a chat window.
From there, the model is able to generate whatever is needed.
Naturally, this action is limited to the digital world, at least in the short term. Perhaps as AI gets embedded, we see this concept expand.
An Entirely New Internet
What we are essentially looking at is an new internet.
If text-to-action occurs, we are basically going to see the average person with fundamental coding skills. It is unlikely the models will compete with the top coders in the world. Nevertheless, some of the basic such as designing an application could be possible.
At the same time, we can use these models to create agents for ourselves. This is one of the reasons why there will be a huge explosion. We could see all kinds of activities automated, once again highlighting how the Internet is going to be AI.
This is Web 3.0.
Schmidt's forecasts are more detailed than what I have stated over the past year or two. That said, it does lead to the same end.
We are about to undergo a radical transformation in society. The digital world is literally going to be put on steroids. How we handle this is going to be a challenge? It is likely, along with massive opportunity, disruption also occurs.
If nothing else, it is going to be a wild ride.