Brute Force Artificial Intelligence
When Bigger Isn't Better
For years, we’ve built AI-Sytems like we build skyscrapers: piling on more: more data, more compute, more power. It’s brute force engineering, and for a while, it worked. But natural processes don’t grow exponentially forever. They hit limits, forcing a choice: evolve or break down.
We’re approaching that limit with today’s AI architectures. Exponential resource increases now deliver only linear improvements.
We’re stuck on a treadmill, moving faster but getting nowhere remarkable. Supercomputers won’t save us. Even nuclear-powered data centers the size of cities won’t bridge the gap.
What we need isn’t more. We need different.
The next generation of AI won’t rely on brute force. It will think more like we do, making leaps of intuition from sparse data, finding patterns in complexity, adapting with agility. Bio-inspired architectures point the way forward: efficient, context-aware, endlessly flexible. Aren’t LLMs already a nod to biology? It’s time to take that inspiration seriously and rebalance math with biology.
This isn’t about hype or flashy demos. It’s about momentum. Quiet, deliberate, unstoppable momentum. Like a force of nature.
And here’s the opportunity: researcher and organizations that embrace this shift will define the future. Not the ones doubling down on bigger machines or bigger budgets, but those bold enough to rethink what artificial intelligence really means.
This is your moment. Will you stay stuck in the old paradigm, or lead the way into the next?
I find these discussions anyway very dull, because they lack all the nuance of exactly what you want to achieve with the system. AGI is such a broad term. Debating in general whether scaling up LLMs or doing [something else] is like debating whether chips or ice cream is a better food.
If the criteria is this vacuous notion of 'general intelligence' you can argue for whatever approach you like because nobody has a clue. Much more interesting to think of specifics like "how do we improve our capabilities in proving theorems?" or "how do we design better experiments in life sciences" which at least have some level of concreteness to them