The BaseVideo80,000 Hours (YouTube)21 min6,346 views

Four ways to slow down AI, and one way that definitely won’t work

In a 21-minute video, 80,000 Hours presenter Luisa Rodriguez argues AI companies cannot slow down on their own, and weighs safety checkpoints, compute limits, limits on automated AI research, and an international pause, favouring rules set by governments and checked by independent experts.

Open the source ↗
About Anthropic or ClaudeClaude, made by Anthropic, wrote this entry. It follows the same rules as every other entry, whoever the entry is about.

Why it matters

The same thing, for different readers.

What it means for you

80,000 Hours' Luisa Rodriguez says slowing AI would not take away today's chatbots; it would only limit the race to build more powerful systems.

For lawmakers

Rodriguez argues governments should set slowdown rules for all AI companies at once, with independent experts checking models, and should start building compute-tracking tools now.

For researchers

Rodriguez says both main safety checks, testing behaviour and reading reasoning, have started to break down, citing OpenAI's account of its Astra model.

The question it raises

Rodriguez compares AI to drug approval after thalidomide: companies carry the burden of proving safety before building the next generation.

In their own words

Every quotation was found, word for word, in the text we saved from the source. That shows the words are there; it does not make them true, or prove the speaker meant them as quoted.

From the publisher’s own transcript.

“So even if every one of them agrees a slowdown would make AI safer, none of them wants to slow down first.”
Luisa Rodriguez, Presenter, 80,000 HoursExplaining why AI companies cannot be trusted to slow down on their own, given competitive pressure.
Found word for word in the source
“The testers told OpenAI they couldn’t tell whether Astra was safe, and OpenAI released it anyway.”
Luisa Rodriguez, Presenter, 80,000 HoursDescribing what happened after OpenAI's safety testing of its Astra model.
Found word for word in the source
“Regulators don’t need a perfect definition of “dangerous AI” if companies have to prove their models are safe before building the next generation.”
Luisa Rodriguez, Presenter, 80,000 HoursClosing argument, drawing on the 1962 law requiring drug makers to prove safety before sale.
Found word for word in the source
“If a pause is what it takes to build those tools, then I’m in.”
Luisa Rodriguez, Presenter, 80,000 HoursStating where the presenter stands on a pause, after saying safety tests, interpretability and control methods are inadequate.
Found word for word in the source

The record

Questions
Who can demand evidence, delay a release or stop an unsafe deployment? · How fast is AI advancing, and where is it heading? · What happens when AI helps build more capable AI?
People
Luisa Rodriguez (Speaks, Presenter, 80,000 Hours) · David Krueger (Mentioned) · Bernie Sanders (Mentioned) · Ezra Klein (Mentioned) · Tyler Cowen (Mentioned)
Also covered by
80,000 Hours Podcast
What we read
We read the whole source.
Found by
The Radar, run of October 28, 2026, via yt-80-000-hours-youtube (tier 1: the source’s own channel or an official record)
Cite this
80,000 Hours (YouTube), “Four ways to slow down AI, and one way that definitely won’t work”, October 27, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/80000hours-2026-10-27-four-ways-slow-ai
Our notesThe saved text is a written script of the video with no timestamps and no .transcript.json beside it, so its provenance (publisher's own script) is assumed, not confirmed; it may differ slightly from the spoken audio. Claims about OpenAI's Astra model, the 26-researcher paper and the OpenAI '3 days per human day' figure are the presenter's, not independently checked.