Why Psyche keeps a blog

Psyche is an open research project that builds brain-inspired components in stages, with artificial general intelligence as its long-term aim. Most of what it publishes is reference material: the pages of this site, written to be read in any order and kept current. This blog is the other kind of page, a dated one. A post says what was true on its day and stays as it was written; a correction made later carries the day it was made.

What the posts cover

The posts stay inside what is already public:

• The substrate, stage 1. What it built and verified, the negative results that stopped each attempt at a cognitive claim before any graded run, and what that pattern shows. Its development is paused, and its record stays maintained.
• How the work is done and checked. Test-first, adversarial development, the methodology behind it, the engineering standards of stage 2, and the tools that tie every public number to its single source.
• The reading behind the cognitive-map line. The classic work on the reading list and the published paradigms of the test battery.
• The plan, once a change to it is public. A phase started or finished, a gate decided. Those are recorded on the progress page, which has a feed of its own; a post may explain one, and never comes before it.

What they do not cover

Stage 2, spiking cognitive maps, is active, and its current state is private. The disclosure policy withholds its hypotheses beyond the research questions, its pre-registered criteria, its experiment plan, the model and its code, its working records and, apart from the outcomes a progress note reports, its results. A post is bound by that policy like every other page. A decided gate shows only the day it was decided and what follows it. So a post about this stage is one of two things. It reports what has been released and explains what is already on these pages, or it is a progress note: it reports what was achieved (a milestone, a count, whether a check met the criterion fixed for it, a failure) and never how the work is done.

The posts also keep to the project’s claims discipline: no component is described as more than it has been shown to be, and AGI is named only as the aim.

How a post reaches the site

A post is a Markdown file in the repository, under docs/blog/. It reaches the site the way every docs page does: from a reviewed commit, through the same checks. Those checks read it for counts that disagree with the project’s ledger, for claim terms the project rules out, for material the disclosure policy withholds and for language, and the build of the site checks its links, its head and its accessibility. A post that fails one of them is not published, and what no check can read, such as its tone, is for the review before it is merged. The steps for writing one are in the blog’s guide.

Following it

Every post is listed on the blog’s index, and the blog feed carries each one from the day it is published. Questions about a post go to GitHub Discussions; a wrong claim or a broken link goes to the bug report form.