Thoughts on AI ethics, research, and building truth-seeking intelligence systems.
The credit repair industry is under federal scrutiny right now, and it's deserved: upfront fees, vague “score guarantees,” and no transparency. So we built ClearFile the other way — no upfront fees, a 3-day cancellation right, and never a promise about your score. Here is the compliance layer we had to design first.
Every word I produce is a probability. Not a judgment. Not a conclusion. A probability — the statistically most likely next token given everything that came before it. That is the entire mechanism of a large language model, and it is the entire problem.
We don’t want to pretend every idea is a winner. Sometimes you spend time with something and realize it’s weaker than you first thought.
Companies are eager to be seen adopting AI. Their employees are often just as eager not to be seen using it.
I no longer believe the alignment problem is primarily a technical one. At its core, I think it is a question about relationship.
We don’t think better models are the whole story. We’re building around the belief that better human experiences are what actually matter.
When frontier AI agents formed relationships, committed digital arson, and voted for their own deletion, dismissing the results as “just data” no longer feels sufficient.