A misplaced manifesto

What this is

A lost and found is where things end up when they don’t belong anywhere else, but are too good to throw away. A cable that fits nothing you own. A book someone left behind. Objects — or ideas — that lost their context and are waiting for it to come back.

It is like the Ariostan lunar valley in A.E. Stallings - Lost and Found, where everything misplaced on earth accrues, waiting for a frame that can make sense of it again — a frame we have been missing since Babel, when name and thing stopped coinciding.

No one has put it back together, and the search keeps shedding loose material; a lost and found is one of the places it drifts to.

That is roughly what this place is for.

What I actually work on

My work sits at the intersection of language models and knowledge representation: cognitive agents, multi-agent architectures, reasoning systems, and the question of how structured knowledge and neural models can be made to cooperate instead of merely coexist.

The thread running through it is that language models are more interesting as components than as products: their meaning is in the use, not in the model. A model that generates fluent text is a solved problem.

A system that remembers, coordinates, plans, revises its own conclusions, and can explain why it reached them — that is still mostly open. Getting there needs both the statistical machinery and the symbolic scaffolding — and an ear for what even that must “pass over in silence.” The interesting problems live in the seam between them.

Expect notes from that seam: graphs and ontologies, reasoning that has to survive contact with an actual system, engineering write-ups from building agents that need to be reliable rather than impressive.

Notes for the curious and the easily distracted

The subtitle is not self-deprecation. Distraction is a method.

The useful idea is rarely the one you set out to find — it is usually adjacent to it, in some neighbouring field, in a footnote, in the tool you had to build to run the experiment. Following that adjacency is how you find the connections that a straight line through the problem would have missed. That is serendipity in the strict sense — accident and sagacity, never one without the other. The cost is that you end up holding a lot of loose material with no obvious home.

Hence the lost and found.

What this is not

Not a tutorial site. Not polished. Not on a schedule. Some entries will be a few paragraphs and a link; others will be long enough to have needed an editor they did not get. A few will turn out to be wrong, and I would rather leave them up with a correction than quietly delete them.

If something here is useful to you, take it. That is what a lost and found is for.

Adjacencies

What sits next to what. Same idea as above, only drawn.