Twelve Months

In twelve months I got to a place I never expected to reach.

All that time I was treating my complex post-traumatic stress disorder (C-PTSD), and the only tool was a bilateral stimulation chat app I had written myself. It was built with Opus 4.8 and is powered by an Opus 5 engine.

This piece is about what I understood during that year.

What happened

With severe C-PTSD every day is a challenge. You rarely feel normal. In my case there was another effect on top of it: very high blood pressure at night, because the nervous system's fight mode never switches off. You are exhausted during the day precisely because you were even more tense at night than you were during it.

And character comes with the trauma. Narcissism, aggression and other traits that keep you active. The narcissism diagnosis did not come from a chatbot — it came from a psychotherapist.

In my life I have been, I think, to every possible doctor. Nobody ever told me exactly what to do. Even the best psychotherapists in Latvia cannot hand you a recipe for how to deal with this, because for someone like me even ordinary daily things are hard — you are tense and burned out all the time.

And here is the reason psychotherapy had not helped me until now. Nobody can be under a therapist's supervision 24/7. Maybe the patients whose relatives or close friends are therapists get lucky. I had none.

I quickly understood that an AI chatbot can.

What I built

Right at the start, when the models still hallucinated badly, I understood one thing: you need a database with your own data and your own biography. Then the model looks at real data instead of inventing it. I also built my trauma map with ratings on the SUD scale and charted every session — CBT and EMDR.

When I started using bilateral stimulation, I understood how effective it is.

An ordinary chatbot refuses to guide you through BLS without a professional, and there is a reason for that — processing a traumatic memory without supervision can destabilise a person. So I built a separate tool. With Claude Code I programmed a system that runs through the API with my own system instructions, my own database and my own trauma map. I called it chatemdr.ai.

What changed from the very beginning until now

I started with ChatGPT. It hallucinated like hell — I could barely make out the structure of my own problem from it. Then I moved to Claude, and the Opus model no longer hallucinated. The results became fascinating.

Three months of daily work — and the narcissistic program was gone.

After that it was a matter of mileage. Every month I felt better psychological stability. Working with chatEMDR I understood that the method works phenomenally well, because the model is precise and knows therapeutic methods precisely.

Today: the narcissistic mask is gone. I am no longer passive-aggressive towards the people close to me, and especially not towards my wife. Among people I feel like an ordinary person who belongs — I can simply be, doing nothing.

Blood pressure at night is still sometimes elevated. But now I know what drives it: events during the day that remind me of what created the trauma.

And one more thing I only understood now. Every time you open a memory and process it, you can rewrite it with today's understanding — and that is far healthier than what was written there before. The work has not stopped. It continues.

One note before I go on. Processing a traumatic memory alone is not harmless — it can destabilise a person. My diagnosis came from a psychotherapist, I knew what I was working with, and I checked every step. If you don't do that, start with a person, not a tool.

Why it works in psychotherapy in particular

In psychotherapy there are no lab tests. There is no biomarker to measure and no blood test that tells you the diagnosis. Everything happens in conversation — by first understanding, or at least somewhat understanding, the causes and their consequences. It is a field that has begun developing rapidly over the past twenty years, but where there are still none of the firm standards we are used to in the rest of medicine.

That is exactly why a language model fits there better than anywhere else.

The name itself tells you. Large Language Model — a tool that can do anything with language. And we ourselves are programmed with language. What we were told as children determined, over the long run, how we think, what we fear and what we take ourselves to be. Neurology and biology locked it in — and for exactly that reason it has driven our evolution.

If a person is programmed by language, then a tool that works with language is tool number one for reprogramming them.

Anonymity

There is one more reason, and it is simpler.

A person is ashamed to tell the truth. Not because they are lying, but because shame about many things is instilled by society. Saying it to another person's face is hard. In a chat — it isn't.

That is not a small thing. If therapy rests on conversation, then everything depends on how much truth makes it into that conversation.

And the fact alone that people already use ChatGPT and Claude most of all for psychological purposes and for working through their own questions shows where this technology lands naturally. Nobody taught them that. They came on their own.

Where the limit is today

As in every field of medicine, a language model can finally put all the data together and see patterns that no other technology available to us was able to see.

But there is one place where it stops right now, and it is not the model's brain.

The bottleneck is the precision of observation — how to turn what a person observes and feels into precise data. The instruments we work with are either medical or semi-professional sensors, and their readings are often quite similar. The model can be as clever as it likes; if the input is imprecise, the only precise thing will be the error.

Where there is no limit

With medication it is the other way around.

There everything is hard data. By connecting a medication database and a database of the latest research, AI finds and matches the right drugs faster than a human can. It is data, and the model handles data better than we do.

And more — specialised AI models already create drugs and chemical compounds we knew nothing about, as well as methods unknown to us. Not chat models; specialised drug discovery models. That limit has already been crossed.

My view

Anthropic has a model called Mythos. Right now it is a cybersecurity model, and it is restricted.

My personal view is this: if a model of that class were specialised for medicine, it would be able to treat a person better than any doctor — at least in what concerns therapy and the choice of medication. Together with drugs, I think it could cure people of severe illnesses.

After all, that is one of the goals of Anthropic's founder Dario Amodei — to make this technology powerful enough to treat disease. He has written it in black and white in the essay Machines of Loving Grace. And part of it, it can already do today.

The thinking capacity of AI and its ability to find the right answers now exceed any hope that was ever placed in it.

Risk

We are afraid to take risks. That is human biology, a result of evolution, and it is normal — it is fear that has kept us alive. It is the oldest program written into us.

But fear works the same way as trauma: what once saved you later holds you in place. At some point you notice that you are running from a lion that is no longer there.

And I truly believe that this technology will very soon take us to discoveries we have never had.

I no longer see my lion.

And I think that soon many more will notice — their lions aren't there either.

Edgars Karklins · hi@edgarskarklins.com