Company

Trimezi researches AI and builds the software that runs on it

Our research is applied. We study where AI genuinely makes someone more capable, and we ship the software that does it.

How we work

Four things decide what we do, in the order they decide it.

  1. The research comes first

    We publish what we found, the method behind it, and the conditions under which it fails. Then we build the products institutions run it in. Doing it the other way produces a claim looking for evidence.

  2. A capability claim carries its limits

    Publishing the failure modes is part of the work, not a caveat at the end of it. A claim without its limits is not a result we are willing to put a name on.

  3. The person using it should end up better

    Every system we ship is measured against whether the person who used it is more capable afterwards. If they are not, we have built the narrower thing.

  4. We would rather be held to it than admired for it

    A position is only binding if it is specific enough to be broken. We would rather say what we will not do and be shown a case where we did it, than describe a value nobody could hold us to.

What we publish

A claim is finished when its limits are attached to it

  1. A claim, broken into the parts that have to be present before it is finished.
  2. What the system can do, how we established it, and where it stops holding.
  3. A claim missing its limits is not a result Trimezi will publish.

The claim

Stated plainly and narrowly. A claim broad enough to be unfalsifiable is not useful to the institution reading it, however good it sounds.

The method

Enough that somebody who knows the field can tell what we actually did, and could argue with the design rather than only with the conclusion.

The limits

The conditions under which the claim fails, published with it rather than after it. This is the part most announcements omit, and omitting it is how a true statement becomes a misleading one.

Without the third

A true statement that will be read as a broader one. We treat this as an unfinished result, not a shorter one.

Where we are

St. Petersburg, Florida
United States

What we publish

Research notes as we finish them, and standing commitments that change rarely. There is no marketing blog, and there will not be one.

Get in touch

For institutions, press, or anything else, write to us directly. A person reads it.

londonc@trimezi.com