Do you enjoy reading Mark’s articles? Sign up for his founder’s newsletter now and stay up to date with his latest writings.
Since the end of 2022, we have seen A.I. boom and, whenever something new comes along, hype sets in. I’ve been diving deep into AI over the last few months, reading research papers, talking to people who have been working with various forms of AI, and listening to way too many interviews and podcasts with the people building these AI products. So let’s try to separate hype from reality and answer one question: where will AI bring us in the near future?
At its core, AI is built upon prediction models that have analyzed vast datasets to make educated guesses. But prediction on its own is not enough. What helped the accuracy of the AI models is the fact that humans gave feedback on the output these systems gave, something that is called human-reinforced learning or RLHF.
Deep learning models, RLHF, and powerful computer chips have led to the development of tools like ChatGPT and AI art generation program Midjourney. While the output of these tools is impressive, you have to understand one thing: they don’t understand what they are doing.
Midjourney, for example, struggled with drawing hands because it doesn’t comprehend what a hand is and how it works. Similarly, ChatGPT often presents false information because it cannot differentiate between right and wrong.
These limitations are hard to overcome and, for a long time, I assumed this would limit the capacity of these tools. My first response was: “Why don’t you explain how a hand works to the machine?” or “Why don’t you connect ChatGPT to Wikipedia or Wolfram Alpha to teach it about the correctness of information?” But it’s not that simple. It’s simply not how the foundation of the technology works, and I thought that would be the end of its capacity to make big evolutionary leaps. I could see a boom of specific AI tools helping you with tasks like creating a video, summarizing a document, or translating a text. Very useful and efficient and a big productivity bump for the knowledge worker, but it would not be a huge leap forward.
But I was wrong. My whole line of thinking was wrong.
I underestimated tools like ChatGPT because I underestimated something else it can do: write simple computer code.
To understand the power that holds, we have to look at a radical new AI tool called AutoGPT. AutoGPT leverages AI to write code, using user feedback to refine and improve its output.
In doing so, AutoGPT automates every administrative step you would do by hand by writing code to do it better and faster. This means it’s not just one tool optimizing one thing: when an AI generates computer code, it can become any tool to accomplish any range of administrative tasks.
In the near future, AI systems like AutoGPT will become personal or business AI assistants, enhancing productivity and efficiency. However, it is essential to recognize the current limitations of these systems. AutoGPT requires a substantial amount of computer power, making it expensive to run. Moreover, running it effectively necessitates a high level of technical knowledge.
While much of the buzz surrounding AutoGPT may be hype, it does provide us with a better understanding of the transformative power of AI innovation. As these limitations are addressed, anyone working on a computer or laptop will experience a significant productivity boost.
As we venture further into the world of AI, it’s crucial to separate the hype from reality while recognizing the potential of these groundbreaking tools. The journey from ChatGPT to AutoGPT and beyond is a testament to the rapid advancements in AI technology. With time, the limitations of these tools will be overcome, and we will witness a future where AI assistants help us perform tasks with unprecedented efficiency and accuracy.
Now if I’ve learned one thing over the past 20 years is that it’s not technology that holds us back. It’s our inability to change our existing behavior, experiment, and, yes, fail. And I fully understand that learning new things and doing things differently is hard.
But it will be the people and organizations that are fastest to change and embrace these tools that will unlock new doors of possibilities for businesses and individuals alike. A.I. won’t help you there. That’s something you will have to try to do yourself.
On our blog we post about a lot of stuff, just go for it and read some posts for your own fun.
from 31 August 2023
from 3 August 2023