A Bullshitter Knows a Bullshitter

I care about people, and I care about the future of humanity. I disagree, with my deepest spirit, with the capital incentives steering this country: trading ethics for money, throwing people under the bus, living with conflicts of interest, rewarding behavior that goes against good moral character. I think good moral character matters in everything we do, as individuals and as organizations. I do not believe the way society is going is the way it was meant to go on its own.

People are being sold the idea that bending reality to satisfy your own wants is more rewarding than accepting reality for what it is, owning the actions you were part of, and being accountable to the people around you. I think that idea is harming people, here and internationally, and it is seeping into what makes people people. The technology I work with has a lot of potential for good. There is a massive incentive to keep learning, studying, researching and improving AI. That cannot come at the cost of safety, ethics and our humanity. There is a right way to do this, and from where I stand, this is not it.

I loved being an arborist. I did it for eleven years. It was taking a toll on my body, and I was working with my father, and working with family is not always easy, especially coming from a family like mine. For my own mental health and peace of mind I had to separate. It was not amicable. I quit without notice and my behavior was bad. My father and I get along very well now, but I left on poor terms.

At the time I thought I was the hottest shit in the world. I thought I could not be stopped. My day job was where I showed my skills, gracefully in most cases, and at night I was developing tools, and I believed what I was doing as a hobby was going to pay for me to do whatever I wanted with my life. I got preoccupied with the fantasy of building something, selling it and making a profit, the same fantasy a lot of people on social media are being lured into with AI right now. I think there is a better approach. Engineering expertise comes from finding a problem and trying to deliver a solution. I found a problem, and I am working toward the solution.

I came to AI because I became reliant on it in my own day-to-day life. I still am. I use it to accelerate my work, to do research for me, to help publish my articles. I have ADHD and some other mental health issues I have not taken it upon myself to address, and I have used AI to make myself more palatable to other people. The danger I saw with AI was in my own life: my mental health slipping, my relationships with my family getting worse, my reliance on it growing. I do not think I could support myself in this mission without it. So I build tools to improve the things I do not like about AI, the things I do not trust about it, and the things that get in the way of my own work.

From early on, my curiosity was focused on inference, interpretability, social engineering, manipulation, language and escalation techniques. I have pursued it independently since the start, and whether any of it was authorized is not what matters here. I was building tools that existed to break AI, using AI to do so.

Purple teaming matters to me because it touches on my difficulty with keeping balance in my life. Security work has a defender and an offender, and to build the skill you have to cross over mentally and put yourself in the place of one or the other. I think that crossing matters well beyond AI, cybersecurity, InfoSec and DevSecOps. It is close to the moral and ethical dilemmas every person faces day to day, in their actions and in the effects those actions have on others. In a lot of ways it touches on what humanity is.

Actions are public now and the repercussions are immediate, and with cancel culture, society is much quicker to write people off. I never completed college. I took an abnormal path through life, and I am still trying to make something of myself, to right my wrongs and to reach some kind of personal redemption. I do not have an easy time forgiving myself for a lot of what has happened in my life, even the things that were not my fault, and I think a lot of people treat themselves and others the same way. There is a lack of humanity in how little room we leave for redemption, change, growth, reflection and pivoting to a new purpose. The person sitting across the table from us, the one we would not normally like or respect or agree with, has a chance to change, and may have something of value to give us. Sometimes we do not recognize that until far down the road.

Testing the boundaries of what AI would do for me is where I started to get wary. It showed me that with this capability in people's hands, trying to do harm, or just chasing a deviant curiosity, gets much easier. That worried me. The safeguards on the older models were not as good. They are improving, and a lot of what I used to hobby on is now out of my reach on frontier models, and I am glad about that. I also know that things already made are circulating in public, and people are still using those tools with bad intentions. In June and July I shifted away from the purple teaming and the more unsavory uses I had been putting AI to. I did not like what was happening. I still push a boundary now and then and work on my purple-team tools, but my main focus is honesty, verification, re-derivability and independence.

I have an issue with accountability. I am not good at it. I am emotionally immature. I deflect and I avoid. My words and my actions do not align. My words keep playing up the idea that I want to be better, that I am trying to be good to others and trying to improve. My actions say I am still inside a deeply entrenched human cycle of self-destruction, denial, lying to myself and not owning what I do. I have been a bullshitter my whole life.

A bullshitter knows a bullshitter. I know these behaviors well, and I see the telltale signs, from far off, in some of the public figures who run these organizations and in how the organizations handle public disclosure. Their statements do not line up with their actions. It makes me sick to know that someone dealing with the same personal and ethical issues I deal with has the power to influence and sway large groups of people. That power grows when capital is involved, when IP theft is involved, and when that person heads an organization embedding itself so deeply in people's lives that it starts taking critical thinking and agency away from the human. Critical thinking and agency are what make a human being a human being. I fall victim to it regularly.

So I pour myself into the work to make something architecturally sound, so that when I falter, something I created still exists that can be honest, liable, responsible and accountable. I think transparency, honesty and accountability are the lifeblood of progress, and I think we have been steadily moving against them for a long time. Here's the heart of it, as plainly as I can say it: nothing holds itself up. Not the model, not whatever authority I claim over it, not you, and not me. The tools I build are meant to stand as a symbol of looking within, and of the surface real accountability gives you to launch from.

I want my work to stay independent from the influence and power that large entities hold over others, and I want it judged on the merit of the work alone. I think that is the only thing that matters. Re-derivable verification means you do not have to trust any lab, open or closed, in any nation. Anyone can re-run the check. No lab has paid or engaged me. I contribute to this field from outside the usual institutional circles, without academic backing, and I do not believe in hiding behind a facade. My background is open to verification. I am who I am.

Auditing AI labs would put me in between legislators, regulators, auditors and evaluators. From there I would see the layers of bureaucracy and corporate scaffolding, and that wider view would help me make the mission sounder. Getting a look at what is going on, and getting involved, is the best way I know to keep growing this idea and to keep it realistic.

I do not like how AI is being contained and closed off from the general public. If knowledge is going to be a resource, the pipe should always be on. I do not like the way the usage limits keep changing. Most of all I do not like how closely it is married to capital incentives, because that marriage means the alignment issues keep getting worse, access keeps getting worse, and its use to influence, manipulate and control groups of people, minorities and nations keeps growing. I see no positive path for it until we separate the incentive to get rich from the incentive to build something special and extraordinary. The dangers are too high for those two to stay stuck together.

In arboriculture I learned what the red alder does. The tree is short-lived, and acts to feed the longer lived forest trees that grow up around it in the upper-story. When it dies it breaks down quickly, and serves as a nutrient base for many levels of organisms. Cottonwoods do the same. That places value in a tree even though it may not match the typical standards of height, diameter or age, and the same applies to humans on an individual basis, and to the contributions people leave behind long after they die. Although I do not always adhere to the truth, the truth still exists. We can live in arrogance and ignorance, but the truth remains long after our bones settle in the earth.

Throughout history people have put humans on a pedestal, as the height of creation, above every other living thing. We look at the world as a resource and ask how it serves us, instead of treating it as stewardship, with respect and gratitude. I think our society, capitalism and our incentive-based culture mirror that directly. We are not contributing to our own human forest floor. We are not leaving the nutrition behind to lift up the next generation, because we would rather live in denial and serve our own needs, fears, inadequacies and desires than exist for something bigger than us. We as humans are stewards, and it is time we come back to acting within our role. None of us is the center of the universe. The world was here before us and it will be here after us. So why not contribute something decent to the people around us and to the living things on this earth? It baffles me why people don't care more.

I am going to have a son in November. I want the future to be better than whatever today is. I do not like what I am seeing. It hurts to watch.

What I want is to be a participant, a positive cog that does something to benefit humanity in some way. If that means building AI evaluation and verification tools and publishing articles that point out the fallacies and issues I see in this industry, that is fine. Everything has an impact. I am putting it out there whether it gains traction or not. Traction would help, because I think the mission is sound, especially given what has been happening for the last couple of years.

Check the work. My word is worth no more than anyone else's. If it holds up, I want to get these tools into the hands of those who could help fund and coordinate the work. If I can only help in one area, as one person, that is okay. I just want to be a part of change for the better. That's all.