My love for science has been rekindled multiple times throughout my life—calculus, physics, advanced bioengineering, satellite image processing, hyperspectral imaging, high-performance computing, kinetics, psychology, electrodynamics, and, undeniably, AI.
Despite having built & sold a deep-learning company with Gonzo, I was skeptical about our trajectory toward AGI—Artificial General Intelligence. To me, deep-learning was brittle, data-hungry, and an alien imitation of the human brain. I used to spend time trying to imagine something else, do we need to start over? Emergent properties in deep-learning? Not really, underwhelming. Geoffrey Hinton's capsule networks intrigued me, but I couldn't grasp the hype around Large Language Models (LLMs) like those pursued by OpenAI—until ChatGPT arrived. I was shocked, like getting your chair kicked out from under you. It was a game-changer. I messaged my friends in Saudi Arabia, advocating that if there was ever a time to invest heavily in AI, this was it. ChatGPT symbolized a crack in the dam, a sign that the path to AGI was real. The media's initial focus on ChatGPT's occasional missteps missed the point entirely, we’ve entered a new double exponential in AI development, buckle up.
Now, with OpenAI's meteoric rise in revenue ($28M → $1B+) and the advent of GPT-4 and the anticipated GPT-5, the trajectory is clear. Nvidia and Microsoft emerge as victors, providing essential tools for the AI revolution—a revolution in which I still ponder Intel's missed opportunities.
Another breakthrough looms on the horizon, a second dam to crack: LLMs capable of planning and achieving goals. I believe we're witnessing the final year where 'prompt engineering' matters—soon, it will be about 'goal engineering,' with machines transitioning from ideation to execution. The success of any venture will hinge on our ability to define goals and measure outcomes.
When LLMs began to surpass their training I was giddy, welcome to the triple exponential. The traditional path to a computer science PhD—choosing a domain, understanding the knowledge base, experimenting, failing, and eventually publishing a novel discovery—is on the brink of automation. The graph below depicts a level of R&D that no single human could achieve, with no end in sight. To me this plot represents 40yrs of algorithm research completed in less than an hour. Explained: Each bar is an algorithm, red bars are algorithms invented by humans, non-red are generative algorithms invented by LLMs, shorter is better.
By 2024, I anticipate a surge in automated goal fulfillment and computational power. Think compute is high now? Get ready for 1-2 orders of magnitude in inference demand. Nvidia also might face its first real competition in the realm of inference-only chips next year.
J.A.R.V.I.S for everyone:
We're racing towards a future where AI, akin to Iron Man's charming assistant, can accomplish any task we set. Whether it's summarizing emails, optimizing work schedules, devising trading strategies, or enhancing process yields, our desires will be one request away from fulfillment. It makes me wonder about current and future startups, building an application won’t be considered much of a moat anymore. Ultimately access to data will be the moat investors should care about.
In time, AI will even preempt our needs. It will anticipate goals you will like and work on those goals while you sleep. In this new era, humans will retain the helm of exploration, steering the ship even as AI anticipates and addresses foreseeable goals. The prospect excites me—imagine having the innovation resources of the Manhattan Project at your fingertips, enabling breakthroughs in any field you choose.
Fun Task ideas suggested by AI where we can start pulling on these threads together:
Develop a unified theory that reconciles dark matter, dark energy, and visible matter, resolving current contradictions.
Formulate a new philosophy of mind that integrates AI consciousness, human cognition, and the emergent properties of complex systems.
Hoping to announce my next startup soon chasing this new frontier, working to finish raising the initial round with strategic investors and customers. Stay tuned, I could not be more excited for the future of innovation. I see a future where every company has automated IP factories capable of never ending innovation.