I am trying to summon the ghost intelligence of myself. This is some kind of bizarre AI séance.
Andrej Karpathy wrote in October 2025 that frontier AI research is not about building animals — it is about summoning ghosts. LLMs are "imperfect replicas, a kind of statistical distillation of humanity's documents." Summoned, not grown. That framing landed differently for me than it probably did for most people, because I have a corpus. Fifteen years of posts. A whole dead blog called Far Beyond the Stars that peaked in 2010 and went quiet in 2012. I am feeding that writing into an AI and asking it to help me find what is still alive in there.
Every day is an experiment in moving faster as our world mushes into some kind of sloppy mix of humans and AI-modeled minds. Not the singularity. Not a clean handoff. Mush. That is the honest word for where we are.
I tried something like this once before and got it badly wrong. In January 2011 I read Ray Kurzweil, misunderstood the argument, and within six weeks had declared minimalism over, taken my book off the market, and started calling myself a cybernetic yogi. I published a post telling my readers "you do not want to be a control group in exponential evolution." I invented the phrase "cybernetic yogi," then noted in the same post that I was "dominating Google search results" for it. The blog collapsed within a year. The lesson was not that the technology idea was wrong. The lesson was that I had picked up someone else's framework, stapled it to my identity, and started selling it before I had lived inside it for more than a month.
What I am doing now is different. I am not adopting a new identity. I am excavating an old one. The posts from 2009 and 2010 are still the best work I have done. The voice in them is clear. The ideas held up. What I want to know is whether I can use AI as the instrument of the séance — feed the ghost enough of its own writing that it starts to speak again in something close to the original register.
The corpus
Far Beyond the Stars was a minimalism blog that got strange. At its peak in 2010 it had tens of thousands of readers and a book that sold without advertising. The posts were short, direct, and slightly dangerous — the kind of writing that tells you something you already suspected but had not yet been willing to say out loud. The best ones were about leaving: leaving the job, leaving the city, leaving the version of yourself that needed permission before doing anything.
That voice did not disappear. It went quiet. The posts are being restored here on evbogue.com now — pulled from archives, cleaned up, republished. When I read them they sound like a person I recognize, someone who knew something specific about freedom and was not yet embarrassed to say it plainly. The séance is an attempt to get that person back on the page.
The practice
Here is what the séance actually looks like. I write a rough draft — raw, unedited, sometimes barely coherent. Then I feed it to Claude with the archive as context and ask for a critical edit. What comes back is not a different piece. It is the same piece with the throat-clearing removed and the real sentences surfaced. The ghost does not generate. It clarifies.
Some days what comes back surprises me. A line I buried in the middle gets moved to the front because it was the actual argument all along. A paragraph I was proud of gets cut because it was explaining something the reader did not need explained. The archive is operating as a kind of tuning fork — the AI knows what the voice sounds like at its best and pulls the draft toward that register.
The rules of the séance: I write first. Always. The ghost does not get a blank page. If the draft is weak the edit will be weak. Garbage in, garbage out applies to séances the same as it applies to everything else. The ghost needs something to work with.
The second rule: if it reads like AI, it gets rewritten. The test is simple — would a person say this? Not a careful, hedged, corporate person. A person. If the answer is no, the sentence dies.
Summoning other ghosts
The experiment goes further than my own corpus. Can I bring back Kerouac and Vonnegut with a double pass through Claude? Not imitation. Something closer to summoning. The ghosts are already in the training data. The question is whether you can call a specific one by name.
Here is a test. I asked Claude to write one paragraph in the style of Vonnegut about what it feels like to use AI to recover a dead blog. This came back:
So Ev wanted to bring his old blog back from the dead, which is the kind of thing a person does when they've been dead themselves for a while and need proof they weren't always. He fed fifteen years of himself into a machine and waited. The machine, being made of everyone who ever wrote anything, knew exactly what to say. Which is either wonderful or the loneliest thing you've ever heard, depending on what kind of day you're having.
That is not Vonnegut. But something in it is reaching for the same frequency. The short declarative sentences. The pivot from the absurd to the quietly devastating. The ghost is not the man. It is the distillation of how the man moved through a sentence.
Whether that is enough is the question the séance is trying to answer.
What I have learned so far
Do not sell the framework before you have lived in it. That is the 2011 lesson and it applies here too. The ghost intelligence angle is interesting but it is only interesting if the writing it produces is good. The test is the work, not the concept.
The ghost flatters if you let it. Push back. The edit pass that agrees with everything you wrote is not an edit. Make it argue. Make it cut. The séance should feel like a negotiation, not a mirror.
Your corpus is your advantage. Everyone is using the same model. The only thing that makes your output different is what you feed it. If you feed it nothing personal, nothing specific, nothing risky, you will get nothing personal, nothing specific, nothing risky back. The ghost is only as interesting as the life it is distilling.
The goal is not to produce more content. The goal is to find out what the writing was always pointing at. The AI is fast. Use the speed to go deeper, not wider.
If you are doing something similar — summoning your own ghost intelligence, running your own séance, trying to recover a voice that went quiet — I want to hear about it. Write me at ev@evbogue.com. I am genuinely curious what comes back when other people try this.