AI for people who don't computer good #3
Where is AI headed? What makes this worth so much damn money?
In Part 1 of this series, we explore how Large Language Models are fundamentally different from the AI systems of yesterday - less "if-then-else" rule followers, more "holy shit they understand me" assistants.
In Part 2, we’ll look at how these AI tools are reshaping work across industries, picking apart specific tasks that AI can handle today (spoiler: it's more than you'd think).
For Part 3, we’ll answer the question “What if AI could work as an entry-level employee in any white-collar job?”
This isn't science fiction - it's where we're headed. And it's happening faster than you think.
A note about Revolutions
Every technological revolution is shaped by its constraints.
The Industrial Revolution (1760-1840) transferred the majority of workers from farms to factories - but factories don't appear overnight. Each one needed transportation hubs, raw materials, specialized equipment, willing consumers, and massive capital investment. You couldn't rush this process: physical infrastructure takes time to build, and even longer to coordinate on a national scale.
The Digital Revolution (1950-2000), however, moved workers from analog to digital technology. Snail mail and fax machines died and were replaced with email and Wi-Fi. This tech didn’t require large infrastructure improvements like factories. Instead we simply piped the sound of beeps and boops over existing phone lines with dial-up modems.
Getting people online was about demonstrating value via convenience and efficiency.
The AI Revolution (2022 - ?), on the other hand, breaks this pattern entirely. It doesn't need new infrastructure - our current hardware works fine. It doesn't need to prove itself - it demonstrates value instantly through complex tasks like customer service and document analysis.
Most importantly, AI tooling accelerates its own development: having better AI lets us build even better AI, creating a self-reinforcing cycle.
The only real bottlenecks? Power and data. That’s why you’re seeing large tech companies make moves like Microsoft turning on Three Mile Island and Meta pirating 81.7 terabytes of copyrighted books.
But, like, how fast is fast?
OpenAI’s launch of the ChatGPT product in November 2022 was a watershed moment. Within two months, it hit 100 million users - a milestone that took TikTok 9 months and Instagram 2.5 years to achieve. This is despite the fact early ChatGPT was far from perfect. It often hallucinated facts, once tried to convince a journalist to leave his wife for it, and when researchers tested it on the bar exam, it scored in the bottom 10th percentile. The system could understand ideas and concepts, but couldn’t use them to do anything more interesting than hold a shallow conversation.
Then came the acceleration. Just three months later, OpenAI released GPT-4. Researchers retested it against the bar exam and this time it scored in the 90th percentile - good enough to practice law in most states. This wasn't a fluke. Across different industry benchmarks, we're seeing the same pattern of rapid improvement. Software developers use SWE-Bench to test AI's ability to fix real-world bugs in code bases. The details here aren’t important, but seeing the huge increase in solved problems in only the past few months gives a clear trend line.
More recently, EpochAI released a new benchmark of expert-level math problems called FrontierMath- hundreds of unpublished, expert-level mathematics problems that human specialists spend days solving. When it launched in November '24, even the best models struggled with the state-of-the-art model scoring 2%. A month later, OpenAI released the o3 model which solved 25% of these problems correctly. Remember: these are problems explicitly designed to challenge human experts by requiring knowledge across multiple math domains.
Where does it lead?
The pattern is unmistakable: each technological revolution has moved faster than the last, but AI's acceleration is unprecedented. The Industrial Revolution crawled, limited by physical constraints. The Digital Revolution walked, constrained by adoption. The AI Revolution? It's sprinting, and picking up speed with every step.
We're rocketing into stage 3 of “hockey stick” growth where capabilities double and redouble in months, not years. When an AI can go from failing the bar exam to outperforming 90% of human test-takers in three months, or tackle math problems that challenge human experts, we're not just seeing incremental improvement. We're watching the emergence of something fundamentally new.
This forces us as society to grapple with questions we thought we had decades to answer. How much decision-making power should we give these systems? What happens to human work when "intelligence" isn't exclusively human anymore? What does it mean for society when cognitive tasks - the last bastion of human uniqueness - become automatable?
The time to tackle these questions isn't some hazy future when AI becomes "advanced enough" to matter. That future is here. We're living in the middle of a transformation that will remake society as fundamentally as the move from farms to factories did - but at a pace that would make our industrial-age ancestors dizzy.
We don't have all the answers. But we're rapidly running out of time to start asking the right questions.
As always, I’d love to hear your thoughts
Your article on AI was incredibly informative and well explained. I really appreciate how you made complex concepts easy to understand for common readers. Looking forward to the next episode