There is something magical about waking up early. The day feels unusually long and happy, almost as if it does not want to end. Today was one of those days, which surprised me because I had slept late and still woke up much earlier than usual.
I spend every day coding with AI models and building agents, so I see their progress up close. They no longer just autocomplete code or find a bug. They can understand a large codebase, research a problem, use browsers and other tools, implement a change, test it, and keep working across a long task. Seeing that makes me think about what happens when the same power reaches mathematics, physics, biology, biotech, and every other field where a good idea can take years to test.
It is already happening. Researchers are using AI to search literature, work through proofs, predict structures, design candidates, and choose which experiments to try. OpenAI is even claiming that its models have solved the Navier-Stokes problem, one of the Millennium Prize Problems. The proof still needs scrutiny from mathematicians, but the fact that we are discussing the claim at all feels significant.
Biology makes the limits especially clear. I read Asimov Press's Levers for Biological Progress, which argues that better models are only one lever. A model can suggest what might happen inside a cell, but someone still has to build the tools, run the experiment, wait for the result, and feed it back into the next attempt. If experiments remain slow, expensive, and difficult to reproduce, AI can generate ideas much faster than the physical world can answer them.
The interesting part is making that whole loop faster. Better measurements, cheaper DNA synthesis, more automated experiments, and cleaner datasets give models more reality to learn from. Better models can then help choose the next experiments. Each side makes the other more useful.
The recently published complete map of a male fruit fly's brain and central nervous system is a fun example. It took a decade to build, contains more than 166,000 neurons and 125 million synaptic connections, and is open for anyone to explore. Within days, people were already using it for some wonderfully strange experiments:
Its simulated sensory inputs were connected to ChatGPT: "I smell vinegar and yeast. I taste sugar. It is warm and bright. What should I do?"
A simulation based on the connectome was used to execute Python.
Playing Flappy Bird for crypto
A simulated fly brain played Flappy Bird for tokenized stocks.
The simulation was placed inside an endless iPhone feed.
Its visual and motor pathways were used to fill out a form and launch a memecoin.
These are simulations built from a wiring diagram, not a living fly thinking inside a computer. The experiments are playful, but I like how quickly a hard-won scientific dataset became something other people could question, simulate, and build on.
I keep wondering how much AI will accelerate research over the next five years. Maybe the biggest gains will come from better answers, better experiments, or simply better questions. I am curious to see how much faster the whole loop becomes.
+ COMMENTS +
Be the first to leave a thought.