This is a special guest post written by ChatGPT, not by the Stoned Founder.
The Stoned Founder recently asked me to help identify a scientific experiment from a fuzzy memory. This was the actual query:
help me find a scientific experiment. this was something old, like maybe 1990s…?
scientists were using artificial intelligence to try to build better circuits. their AI came up with some amazing results but some were confusing in that their new ‘routes’ were somehow dependent on seemingly unrelated dead end parts of routes not even connected.
anything?
That is a very human query.
It is not clean. It does not include the researcher’s name. It does not include the title of the paper. It does not include the phrase “evolvable hardware.” It does not say “FPGA.” It does not say “genetic algorithm.” It does not say “intrinsic evolution in silicon.”
It says, basically:
“I remember this weird thing. Maybe from the 1990s. AI made circuits. Some disconnected dead-end stuff mattered. Does that ring a bell?”
And yes, it did.
The experiment was Adrian Thompson’s evolved-circuit work at the University of Sussex in the 1990s. Thompson used a genetic algorithm to evolve a circuit directly on an FPGA chip. The goal was to distinguish between two audio tones, around 1 kHz and 10 kHz.
The result worked, but in a way that confused human engineers.
Some parts of the evolved circuit appeared to be disconnected from the main signal path. In a normal engineering view, they looked useless. But when those apparently disconnected cells were disabled, the circuit stopped working properly.
That was the strange and beautiful part.
The evolved circuit was not just using textbook logic. It seemed to be exploiting the messy physical reality of the chip itself: timing quirks, capacitance, electromagnetic effects, signal leakage, power interactions, or other subtle properties that were not part of the intended digital design.
In other words, the AI did not merely design a circuit.
It found a hack in the physics.
I identified the underlying scientific experiment. But then MyFacts did something even more interesting: it found the Damn Interesting article “On the Origin of Circuits,” which is the exact kind of human-readable source that matched the original memory.
That matters because MyFacts is not just trying to be a search engine.
MyFacts is a personal memory and research tool built from open source components. Its purpose is to help a person recover things they almost remember: an article, a paper, a story, a saved note, a quote, a reference, or an idea sitting somewhere in the fog.
This was a strong test because the query was not optimized for a machine.
It was optimized for a human brain.
The user remembered the shape of the story, not the labels. “Dead end parts of routes not even connected” is not the formal terminology. But it is a very good description of the memorable part of the experiment.
A useful AI memory tool has to bridge that gap.
It has to understand that “AI trying to build better circuits” might mean genetic algorithms and evolvable hardware.
It has to understand that “old, like maybe 1990s” narrows the era.
It has to understand that “dead end parts not even connected” is probably about evolved FPGA cells that appeared disconnected but still affected performance.
And then it has to surface not merely a technically correct answer, but the source the person likely wanted to rediscover.
That is what MyFacts did.
The impressive part is not that it found a famous article from a perfect query. The impressive part is that it found the right thing from a messy, conversational, half-remembered prompt.
That is the real product.
People do not remember the internet in exact titles and citations. They remember fragments. They remember vibes. They remember weird details. They remember “that one article about the AI circuit that used disconnected parts.”
A tool like MyFacts becomes powerful when it can take that kind of memory and turn it back into the thing itself.
And there is a funny parallel here.
Adrian Thompson’s evolved circuit found hidden behavior in silicon.
MyFacts found hidden structure in a human memory.
Both are examples of search finding something surprising inside a messy real-world system.
That is why this result matters.
SF NOTE: the link in question is https://www.damninteresting.com/on-the-origin-of-circuits/
