About — securesein

I'm Sebastiaan. I've been into technology for as long as I can remember, and these days that curiosity is pointed squarely at machine learning. This site exists mainly for myself: a way to actually keep up with a field that moves faster than any one person can track by skimming headlines, and to work through the mechanism behind something rather than just the announcement of it. If it turns out to be useful to anyone else along the way, that's a bonus, not the reason it exists.

How much of this is written by a model

Nearly all of the sentences. That is true of the news items, and it is also true of the long explainers and the deep dives — which the old version of this page implied were hand-written, and they are not.

What differs between them is how much of my judgement went in. On the news items: none, in the sense that nobody reads them before they publish. On everything else: I choose the subject, set the angle, supply the source material, check the claims that carry weight, and rewrite what comes back wrong. Every post says which of those things actually happened to it, in a box directly under the title, generated from a record rather than written by hand.

The long version — which models, what I verify, what I don't, and what can still go wrong — is on the colophon.

How Scout works

Scout is the pipeline behind the News section. It is not a person and it is not a persona; it is a few hundred lines of Python on a schedule.

It reads a list of RSS feeds every few hours.
It summarises what is new and messages the summaries to me on Telegram.
I flag the ones worth writing up — sometimes with a one-line steer about the angle, sometimes not.
A cloud model drafts a full post from the source article, usually within the hour.
It publishes. Automatically. Nobody reads it first — not me. Every one links back to the article it came from.

What is not reviewed: the facts, the framing, the numbers, the conclusion. My only editorial act is choosing which item gets written up at all. If a News post here matters to you, open its source and check it.

The five sections

Model Updates: What is new in the model landscape?
Research: What have we learned?
Benchmarks: How good is it actually?
Security: What can go wrong?
Fundamentals: How does this work, from first principles?

Cutting across all five are the topics — thirteen subjects, so a two-minute news item and a twenty-minute teardown on the same thing end up on the same shelf. A few of those shelves are nearly empty on purpose: they are where this blog is meant to grow.

Corrections

If something here is wrong, I would rather know. How corrections are handled is on the colophon.