Are we really in an AI bubble?
The AI hype has generated an inevitable fatigue, but this is a symptom that radical changes are about to happen.
Year 1900, and 14 years had already passed since a man named Karl Benz made the first prototype of the car. Meanwhile, a few kilometers away, in the United States, there was Henry Ford, an entrepreneur who was looking for investors for his mechanical workshop. His purpose was to convince these businessmen of his idea about changing the world through automobiles. And of course, in those times, it was unthinkable to believe that the future was these vehicles, as they relied on at least two professional mechanics to operate (one of them at the wheel and the other watching the engine start). All this without mentioning that these vehicles didn’t have all the safety mechanisms we enjoy now like suspension and seat belts and that there were more than 700 car manufacturers, and many of these investors had already lost their savings betting on that same idea Ford was proposing.
If we were to go back to those times and live in the previous context, I know many of us would have doubts about the future of technology and think from a financial point of view that it is a great bubble.
Now let’s go back to 2023, where AI has amazed and panicked us, as we have been told that it will replace humans and might one day be the end of our existence. We’ve also seen that there are people investing time and money in continuing to create more of them.
AI has indeed provided astonishing results; however, this technology still has flaws and those who have used an AI product have experienced them. So can we say that, like the example of the car, AI is another financial bubble being created? Should we assume then that the AI path will end at any moment?
Well, let me tell you that this is only the end of the beginning.
In recent months, we've witnessed an acceleration in the improvement of existing AI products. ChatGPT is a prime example of this, with the launch of GPT-4 introducing new capabilities to its predecessor, enabling more complex tasks like interacting with images and accessing the internet. However, there's already talk of GPT-4.5, which supposedly adds functionalities like video, audio, and 3D, though OpenAI's leaders have not confirmed anything at this time. Moreover, there were rumors about AGI that shook the tech world, which was part of project Q* and allegedly led to Sam Altman's dismissal. On another note, we can discuss the launch of Google Gemini, which only reinforces the feeling that the AI market is becoming saturated with endless improvements, more models, and more people eager to create their own ChatGPT.
But are we truly advancing by leaps and bounds, or are we merely under the false impression of progress because we're in the AI ecosystem?
To better understand this question, let's look at a graph from Silicon Valley University, which explains the Six D's and their function as a roadmap, showing what can happen when an exponential technology emerges.
According to this framework, every exponential change has a deceptive phase where transformations seem to stagnate, and products tend to fail. Currently, we are experiencing this phase, as we witness many AIs being created, maturing, and perishing due to their inherent limitations.
Therefore, I propose the thesis that we are at the end of the beginning of a major transformation of AI models.
So, what's next? Consider the following.
1) User-Oriented AI interfaces
ChatGPT has been the first manifestation of a shift that has impacted us all with its multiple everyday uses in our tasks. But the most significant aspect of this technology is how well it can understand us. This is the most radical change introduced by language models: acknowledging the errors they sometimes generate. Here lies the differential value.
We are currently witnessing what could be called the software revolution, which aims to enhance the understanding of our instructions. Evidence of this includes features in products like
ChatGPT with its custom instructions
Microsoft Copilot with its quick suggestions
Adobe with its text message box that allows designers and artists to better communicate with AI
2 ) AI as a new commodity
A commodity is a category of products with many competitors and little differentiation between them.
Now, applying this concept to the current scenario, anyone is training new language, image, or video models. This gives any entrepreneur or company (large or small) the ability to connect to AI and develop their own chat-based applications.
In the coming months, we'll see the emergence of new models that will become increasingly difficult to distinguish from each other. As a result, many of these new products will end up being available for free. This is what Meta is doing with its Llama 2 model, which everyone can download and use for personal or even commercial purposes.
What will differentiate AI models or companies from others?
A key differentiation factor will be the ability to distribute the product. Giants like Google, Microsoft, X (Twitter), Meta, or Apple have a significant advantage here, as they can launch their AI products and (if they’re good enough) achieve almost immediate success.
Another key differentiation factor will be the data each model is trained on. Here’s where the great battle lies. Companies will choose one model over another.
This explains seemingly nonsensical things like Meta's alliance with Microsoft to distribute the Llama 2 model, even though it directly competes with OpenAI, a company in which Microsoft has made significant investments and with which it is also a partner. All this is to have a wider variety of language models available.
Another point of differentiation is privacy. For example, Microsoft's Bing Chat Enterprise and ChatGPT Enterprise both commit to not using corporate data or user queries, offering complete protection and discretion of information.
3) The use of AI and the types of products it will generate
As the months go by, we will start to see the use cases that AI brings, including the multimodal capabilities that models are currently being trained on. Tools like Copilot are emerging, which promise to enhance the capabilities of people and professionals. Also, there are AI agents that can perform tasks involving multiple steps and decision-making, and let’s not forget about upgrades that can change the market dynamics.
To understand this better, let's use the example of the early days of the Internet when search engines gained popularity in 1995 and 1996. Search engines like Yahoo, Excite, Altavista, and many more emerged and until 1997 everything seemed stable as they competed with each other (just like the current AI models). In that same year, two young students at Stanford, who would later become the founders of Google, presented the results of their new search model to several leading companies, only to face rejection. However, Larry Page and Sergey Brin persisted, eventually bringing their innovation to market and radically altering the landscape as we know it today.
Due to the dynamic nature of the market and the rapid emergence of various technologies, new developments in the field of AI are expected to continue. However, it will not be until we move out of this stagnation phase that the new paradigm will be unleashed.
In conclusion, as I had previously proposed, I believe we are at the end of the beginning. The current models show signs of fatigue, and concurrently, there has been a surge in experimentation with interesting products and others not yet available at the moment.
All that remains is to have patience, adapt, and keep an open mind for this new revolution.