The BaseEssayAI Alignment Forum
Minimum required pacing: finish aligning it before you train the next one
On the AI Alignment Forum, an author writing as orthonormal proposes a minimum pacing rule: a lab may not train its next, more capable generation of models until significant alignment problems in the current one are adequately resolved, as judged by several independent alignment organizations. The author calls it a vague heuristic, not a concrete proposal, and lists open questions.
Why it matters
The same thing, for different readers.
What it means for you
The idea is simple: an AI company should fix the problems in today's AI before building a more powerful one, and outside experts should check that it did.
For lawmakers
Sketches a trigger for coordinated slowdowns: training the next generation waits on independent sign-off, restricted deployment first, and pulling a model when a significant problem appears.
For researchers
Frames a gate that both optimists and pessimists about current alignment methods could accept, and leaves open what counts as a model generation or a significant issue.
The question it raises
Who should decide when an AI is aligned enough, and should progress wait for that judgement?
In their own words
Every quotation was found, word for word, in the text we saved from the source.
From the source’s own text.
“My proposal for a minimum requirement: a lab cannot train the next generation of models (in terms of capabilities) until they have adequately… resolved all significant alignment issues in the present generation of models.”
“Multiple independent alignment organizations examine and red-team the newest model at many checkpoints throughout training, with full access to the data and with the best tools available.”
“And it creates much better incentives for labs: in order to advance in capabilities, they'd have to push the frontier on alignment and interpretability techniques.”
“I still wouldn't say that this is fully adequate, but I'd treat it as a minimum requirement.”
“Hardware/economics/industry will already be coincidentally pacing the scaling, starting pretty much right now, for reasons unrelated to (the risk of) AI incidents.”
The record
- Questions
- Who can demand evidence, delay a release or stop an unsafe deployment? · Are AI companies putting stronger safeguards into practice?
- People
- orthonormal (Wrote it) · Vladimir Nesov (Speaks)
- What we read
- We read the whole source.
- Found by
- The Radar, run of October 12, 2026, via
ai-alignment-forum(tier 1: the source’s own channel or an official record) - Cite this
- AI Alignment Forum, “Minimum required pacing: finish aligning it before you train the next one”, October 11, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/alignmentforum-2026-10-11-finish-aligning-before-next-model