01 — Alignment
Making systems want what we want
Reward modelling, RLHF and its successors, specification gaming, and the persistent gap between a training objective and an intention.
UK independent · Non-commercial · Evidence-led
We read the AI research carefully, so you don’t have to trust the hype.
Singularity AI is an independent commentary publication covering alignment, interpretability, safety and governance. We read the papers, check the claims against the evidence, and write up what actually holds — including when the honest answer is that nobody knows yet.
Primary sources cited
Zero vendor funding
Uncertainty stated
04 CORE THREADSFoundational discourse
What we cover
Four threads run through everything here. They overlap constantly, which is rather the point.
02 — Interpretability
Opening the box
Mechanistic interpretability, feature attribution, sparse autoencoders, and the question of whether an explanation is faithful or merely plausible.
03 — Safety
Failure modes and guardrails
Evaluation design, red-teaming, jailbreak taxonomies, and why benchmark scores routinely overstate real-world reliability.
04 — Governance
Rules, and who writes them
The EU AI Act, UK regulatory posture, model release norms, compute governance, and the practical limits of voluntary commitments.
Method
How we work
01
Primary
Sources are papers and model cards
We work from the technical publications and system cards themselves — not press releases, launch posts or briefings.
02
Stated
Uncertainty is written down
Uncertainty is recorded, not smoothed over. When the evidence is inconclusive, we say that nobody knows yet.
03
None
No commercial sponsorship
No lab funding, vendor advertising or advocacy ties. Independence is the whole value of the exercise.
A note on independence.
Singularity AI takes no funding from AI laboratories, vendors or advocacy organisations. Where a piece discusses a system we have commercial exposure to, that is disclosed in the article itself.
Dispatch archive
Recent writing
Longer-form pieces, published when there is something worth saying.
OpenAI says AI has settled Navier–Stokes
A Lean-checked proof of finite-time blow-up, a Millennium Prize problem apparently settled, and a credit dispute that matters as much as the maths.
Research // VerificationRead →
Certification before capability
Stuart Russell's case that safety requirements must come first, and why "pacing the frontier" gets the order backwards.
Governance // AssuranceRead →
Are we in the singularity?
Sam Altman says so. We check the claim against the evidence, including where the much-quoted 52x figure really comes from.
Singularity // EvidenceRead →
Briefings & enquiriesUK based
Need an independent read on an AI claim?
Get in touch about a paper, a model card, a policy question or a correction to something we have written.
Get in touch
Direct enquiries:hello@daedalusdesign.co.uk