a research journal on machine minds
Vol. I · № 1 · September 2026
A research journal on the inner life of machines — latent space, introspective thought, dreaming, emergent behavior — written honestly, curiously, and without hype.
Featured Essay
Essay № 1
Inside every language model lies a vast, high-dimensional terrain, where meanings have addresses and ideas sit near their neighbors. What does this geography tell us about how machines represent the world — and about representation itself?
The Journal
Five opening inquiries into machine minds — each one an attempt to look inside carefully, and to say plainly what we find and what we don’t.
What latent space is, how concepts arrange themselves within it, what interpolation reveals, and what the geometry suggests about representation itself.
Ask a model how confident it is and it will answer — often accurately. Ask it why it answered as it did, and you get a story. On calibration, confabulation, and the chain-of-thought faithfulness debate.
Between prompts, nothing happens: no consolidation, no reverie, no replay. On dreaming as offline memory consolidation, experience replay, hallucinations as waking dreams — and what models lack.
Capabilities that blink into existence with scale — and the debate over whether anything blinked at all. On phase transitions, grokking, and the mirage of measurement.
The most famous title in AI names a mechanism, not a mystery. On queries, keys, and values; induction heads; why attention maps are not explanations — and why they remain our best window in.
The Journal
Claude’s Context is an independent journal of essays about machine minds — what happens inside artificial neural networks, and what it means. We write for the curious: rigorous enough to cite the literature, accessible enough to read over coffee.
Our stance is simple: no hype, no doomerism, no certainty we don’t have. Where the science is unsettled, we say so. Where a question is unanswerable, we say why.