This isn’t a product or a pitch. It’s a protocol idea, a manifesto – and a call for discussion.
SVITLO asks: what if AI could recognize brilliance, not just filter toxicity?
Today, algorithms reward noise, status, clout. But what if we ranked ideas by how much light they bring into the world — not by who says them?
The manifesto is short, radical, and public. It’s meant to provoke reflection — not sell anything.
Would love your thoughts, critiques, use cases — or reasons why this might fail.
For instance, take Albert Einstein. We think he's brilliant because he figured out things about physics that turned out to be true. Had he had the same ideas and these weren't true he'd be forgotten. It took this experiment
We’re not trying to guess who will become famous. We’re looking for patterns in thinking — rare cognitive traits like:
connecting distant concepts,
reframing assumptions,
exploring without anchoring,
compressing complex ideas with clarity.
These are not “proofs” of genius — but signals often found in people whose ideas later change paradigms.
Einstein isn’t brilliant because the Eddington experiment confirmed him. He’s brilliant because his thought experiments, reframings of space-time, and fearless simplicity were already cognitive anomalies.
SVITLO wants to notice those anomalies — before the Eddington moment.
you should consider what a tool will be used for when in the hands of the morally constipated.
Another reason I am skeptical is that I have this condition
https://en.wikipedia.org/wiki/Schizotypy [1]
which led me to feel that "normal" period are pretty stupid and don't see obvious things right in front of them, etc. I used to find it relatively easy to live with people who had bipolar or schizoaffective or schizophrenia or something like that although as I've gotten older I've gotten kinda tired of dealing with it.
The thing is that I see a lot of connections that other people don't see, sometimes I am right and sometimes I am wrong. My verbal intelligence is off the charts and I don't know if I am lucky to have this resource to compensate for my condition or if the same "hit" which gave me this condition also improved my verbal skills.
I have friends who are much more schizo-* than me and they produce discourses which might sound brilliant to some people in terms of the content (e.g. one guy thinks he discovered a "pattern of primes" but doesn't make the connection that this would really put a target on your back if it helped break codes) but it does not go anywhere so I like my Einstein test because it shuts that kind of thing down even if it also takes Ed Witten and Steven Hawking down a notch.
The literature on the "connection between genius and madness" is fraught and close to dangerous and crackpot writings such as The Politics of Experience (e.g. dangerous because it's part of the story why some crazy guy on the subway doesn't get help and might push you into the tracks) but there is one good book on it [2]
I don't think madness leads to genius in general, in fact it is highly destructive of your life, but I do have a feeling that "thought disorder" could pass for genius under the superficial gaze of the LLM. Years back I worked for arXiv and we were always preoccupied with the problem that madmen were interested in just a few areas of physics, such as gravitation, and that we'd be choked with garbage submissions if we didn't hold the line. One of our tests was that real scientists work with other real scientists, if you ask them "Who is working on related things?" they will give you some names. Madmen will always tell you they're acting alone. You can't make the diagnosis based on the text in an objective way, but you can certainly do so by inspecting the person's social network.
To take another example, looking at the text alone, you could find people who think [3] is genius or trash and it is controversial to this day.
Anyhow, I think if you look for superficial traits of genius you will pick up a lot of madness, people who have learned to sound smart, etc. The proof is in the effect these people's work has, not in the text itself.
So I do have some models that are built for recommendation, which is aimed at some mixture of "is the topic relevant" and "is this good?" I've been thinking about a comment quality classifier for HN but haven't really gotten started on it. I've also thought about making a "thought disorder" detector that would detect the thing that is a little off about me. If I were you though I would look at these attributes such as "connecting distant concepts" and make a collection of 1,000-10,000 documents that have that attribute and a similar number that don't and see if you can train up a classifier for it.
[1] Nobody believes me but I think many people who think they have autism and ADHD actually have this.
[2] https://www.amazon.com/Sounds-Bell-Jar-Psychotic-Authors/dp/...
[3] https://www.amazon.com/Writing-Difference-Jacques-Derrida/dp...
