THE SERIOUS BIT
AI is moving fast. This is where we stop and think about where it’s going.
Curiosity welcome. Certainty optional. Take your time.
01 / WHY HERE?
The games next door are an invitation to play. This corner is an invitation to think about the technology helping people make things like them.
Being excited about what AI makes possible leaves plenty of room for questions. Who benefits? Who carries the risk? And when does moving quickly stop being a good thing?
Pause, play and think. There’s room for all three.
02 / THE READING ROOM
Stanford’s 2026 AI Index reports that global AI computing capacity grew 3.3 times per year since 2022. It also finds that several leading developers no longer disclose basic details about their most resource-intensive models.
A question to sit withWho gets to check the systems that are becoming harder to inspect?
Read Stanford’s research overview ↗The International AI Safety Report describes gains in coding, maths and autonomous tasks, alongside uneven performance and difficulties with safety testing. It reviews emerging risks and safeguards without recommending a particular policy.
A question to sit withWhat evidence of reliability should come before a system gets more responsibility?
Read the international safety report ↗METR explains how task selection and modelling assumptions affect its estimates of the length of tasks AI can complete. As systems approach the limits of the test set, estimates become more sensitive to those choices.
A question to sit withDoes a headline describe reliable everyday work, or success on a particular test?
Read METR’s explanation ↗03 / THE BIG QUESTION
First, slow down what? Research, the training of frontier models, or putting systems into everyday use? Those are different decisions. Here are two cases to consider, rather than a vote you have to cast.
Testing takes time. So does finding out what happens outside a demonstration. A slower release of powerful systems could leave more room for independent checks, public scrutiny and people affected by the changes.
Where a mistake could cause widespread harm, the case for showing safety before scaling up becomes stronger.
The hard question: Who sets the limits, and how would a pause be enforced fairly?
Useful tools can help people learn, create and solve problems. Delaying them has costs too. Experimentation can reveal failures and improve safeguards, while rules that only large companies can afford could shut smaller builders out.
The case is strongest when uses are limited, outcomes can be checked and mistakes can be corrected.
The hard question: How do we make sure the people taking the risks also get a say?
Perhaps the pace should depend on the stakes: room to experiment with a little game, stronger checks before a system makes consequential decisions, and clear ways to stop or reverse a deployment. Independent evaluation, reporting failures and giving affected people a voice are concrete places to start.
This is an editorial starting point for discussion, not a claim that there is one agreed answer.
04 / THE AI SPEEDOMETER
Our reading of the selected evidence: infrastructure is expanding and some capabilities are improving. That doesn’t establish one acceleration rate for all of AI.
Editorial snapshot · 14 September 2026
No live feed or numerical overall score.
Computing capacity is increasing. Stanford, 2026
Better results coexist with failures. Safety Report, 2026
Different measures tell different stories. Read the evidence above
05 / A LITTLE THOUGHT EXPERIMENT
Imagine a made-up capability starting at one unit. What happens over two years if it doubles at different intervals?
This is a maths toy, not an AI forecast. Real progress can stall, change direction or hit limits. The starting point and doubling intervals here are invented.
16×
Doubling every 6 months means 4 doublings: 1 → 2 → 4 → 8 → 16.
Each dot is one unit. Partial units are rounded up.
Multiplier = 2 raised to (24 ÷ interval in months).
About this corner: an AI-assisted opening collection, with links to the original research. Reading notes summarise sources; questions and debate pieces are editorial. It’s a place to reflect, with no promise of daily updates.