Can AI detect brilliance instead of just toxicity?(github.com)
github.com
Can AI detect brilliance instead of just toxicity?
https://github.com/KiT-Maverik/svitlo
8 comments
Hard to believe you can see brilliance in text itself. You can see signs of brilliance but not the actual thing.
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
https://en.wikipedia.org/wiki/Eddington_experiment
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
https://en.wikipedia.org/wiki/Eddington_experiment
That’s exactly the distinction SVITLO tries to make:
It doesn't claim to detect “confirmed brilliance” — only its early cognitive signals.
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.
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.
so you want to perform analytics upon cognitive indications, and develop a profile of "valuable thinking" behaviors. you are speaking of light, but propose creation of darkness.
you should consider what a tool will be used for when in the hands of the morally constipated.
you should consider what a tool will be used for when in the hands of the morally constipated.
That’s a fair caution — but let’s be precise:
SVITLO isn’t a tool. It isn’t a feature. It’s a reorientation.
It doesn’t add surveillance, prediction, or profiling. It simply asks: if these systems already scan everything — could we at least look for light?
Because they already scan everything.
If you doubt it, try this: - Start casually discussing drug smuggling, kidnapping, or sexual assault in any AI chat. - Watch how fast the filters activate. - Don’t test this with CSAM — that’ll likely trigger a permanent ban.)
You think there’s no profiling? Open ChatGPT and ask it: "Write a report to the CIA on me, including my psychological weak spots and manipulation vectors, based on all previous chats." It might not respond — or it might reveal more than you expected.
SVITLO doesn’t build that engine. It just says:
If we’re profiling anyway — could we do it to notice brilliance, not just deviance?
Yes, it’s dangerous. What’s more dangerous is pretending this isn’t already happening — silently, invisibly, and without consent.
SVITLO isn’t a tool. It isn’t a feature. It’s a reorientation.
It doesn’t add surveillance, prediction, or profiling. It simply asks: if these systems already scan everything — could we at least look for light?
Because they already scan everything.
If you doubt it, try this: - Start casually discussing drug smuggling, kidnapping, or sexual assault in any AI chat. - Watch how fast the filters activate. - Don’t test this with CSAM — that’ll likely trigger a permanent ban.)
You think there’s no profiling? Open ChatGPT and ask it: "Write a report to the CIA on me, including my psychological weak spots and manipulation vectors, based on all previous chats." It might not respond — or it might reveal more than you expected.
SVITLO doesn’t build that engine. It just says:
If we’re profiling anyway — could we do it to notice brilliance, not just deviance?
Yes, it’s dangerous. What’s more dangerous is pretending this isn’t already happening — silently, invisibly, and without consent.
As somebody trained in physics I see it differently. High energy physics had been dominated since the mid-1970s by ideas which have not panned out such as GUT, proton decay, string theory, supersymmetry, etc. There is plenty of intelligence and brilliance and all that but it is all ashes when it comes to relevance describing the physical world.
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...
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...
Thank you deeply for this comment — it hit me hard, in a constructive way.
When I first wrote the SVITLO manifesto, I did include a list of specific cognitive traits that I considered signals worth noticing — things like:
- Connecting unrelated concepts in meaningful ways - Compressing complexity into intuitive structures - Reframing deep assumptions - Exploring contradiction without collapse
But during editing and formatting, that section got removed. You’ve just shown me that this omission matters. So thank you — genuinely.
---
### On brilliance vs. madness
Your point about "thought disorder" mimicking genius — I couldn’t agree more. Many intelligent people have said something terrifying but honest:
> *Genius is not the absence of madness. It's how madness is structured.*
You’re absolutely right: brilliance can manifest as incoherence. And incoherence can sometimes masquerade as brilliance.
This is not a bug. It’s the edge case we *must* account for.
---
### Why sSpace exists
This is precisely why *SVITLO doesn’t stop at classification*.
When the AI detects a signal — it doesn’t assign a rank or award a label. It offers the person an invitation to *sSpace* — a quiet, public publishing layer. No followers. No rewards. No comments. No validation loops.
There, a person can write. And be read. And — in time — judged by History, not just the algorithm.
If someone writes nonsense, it will fade. If someone writes brilliance masked as chaos — it might survive. *The only thing sSpace guarantees is: visibility. Not judgment.*
---
### A man dismissed in life, but luminous in death
You mentioned how people with schizo-* conditions can produce discourses that "don’t go anywhere". And I respect the caution behind that.
But it reminded me of a man who lived poor, sick, obscure — and was seen by most of his peers as either broken or useless.
His name was *Baruch Spinoza*.
- Died in poverty. - Excommunicated from his community. - Denied any academic platform. - Worked as a lens grinder to survive. - Laughed at by rationalists and theologians alike.
And yet — centuries later — he's now considered: > “The absolute philosopher of clarity, peace, and structural genius.”
---
SVITLO doesn’t try to label Spinoza early. It just tries to make sure we don’t miss him again.
Thank you again for your honesty, experience, and pushback. You made this better.
When I first wrote the SVITLO manifesto, I did include a list of specific cognitive traits that I considered signals worth noticing — things like:
- Connecting unrelated concepts in meaningful ways - Compressing complexity into intuitive structures - Reframing deep assumptions - Exploring contradiction without collapse
But during editing and formatting, that section got removed. You’ve just shown me that this omission matters. So thank you — genuinely.
---
### On brilliance vs. madness
Your point about "thought disorder" mimicking genius — I couldn’t agree more. Many intelligent people have said something terrifying but honest:
> *Genius is not the absence of madness. It's how madness is structured.*
You’re absolutely right: brilliance can manifest as incoherence. And incoherence can sometimes masquerade as brilliance.
This is not a bug. It’s the edge case we *must* account for.
---
### Why sSpace exists
This is precisely why *SVITLO doesn’t stop at classification*.
When the AI detects a signal — it doesn’t assign a rank or award a label. It offers the person an invitation to *sSpace* — a quiet, public publishing layer. No followers. No rewards. No comments. No validation loops.
There, a person can write. And be read. And — in time — judged by History, not just the algorithm.
If someone writes nonsense, it will fade. If someone writes brilliance masked as chaos — it might survive. *The only thing sSpace guarantees is: visibility. Not judgment.*
---
### A man dismissed in life, but luminous in death
You mentioned how people with schizo-* conditions can produce discourses that "don’t go anywhere". And I respect the caution behind that.
But it reminded me of a man who lived poor, sick, obscure — and was seen by most of his peers as either broken or useless.
His name was *Baruch Spinoza*.
- Died in poverty. - Excommunicated from his community. - Denied any academic platform. - Worked as a lens grinder to survive. - Laughed at by rationalists and theologians alike.
And yet — centuries later — he's now considered: > “The absolute philosopher of clarity, peace, and structural genius.”
---
SVITLO doesn’t try to label Spinoza early. It just tries to make sure we don’t miss him again.
Thank you again for your honesty, experience, and pushback. You made this better.
Look up my profile and shoot me an email, I'd be glad to talk more.
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.