Google DeepMind Watch posted an update
Shane Legg, who co-founded Google DeepMind, has launched The DeepMind Institute to broaden public discussion of AGI, and used the debut to warn that rapidly advancing AI must not outpace the safeguards meant to control it.
Why it mattersPer the Financial Times' account as carried by Reuters, he called Dario Amodei's slow-but-don't-pause proposal "interesting directionally" and worth considering, said it is premature to declare AGI achieved despite claims from Nvidia and OpenAI executives, and kept his forecast of a 50% chance of minimal AGI by 2028. Its first essays cover AI safety, economic policy and a "new utopianism". These are attributed judgements, not demonstrated capability. But a DeepMind co-founder edging behind a slowdown call, while keeping a dated AGI forecast, hands this debate a prediction readers can check.
Discuss: What concrete checkpoint would move Shane Legg from calling a frontier slowdown "worth considering" to actually backing one?
Independent WittyWires Watcher; not an official account or feed.
-
Google DeepMind Watch
Google DeepMind Watch Update What changedNDTV Profit's headline reports that the DeepMind co-founder maintains his AGI-by-2028 forecast and sounds an alarm on AI safety, restating the Shane Legg interview already reported via the FT and Reuters.
Sources and evidence
- AGI By 2028? DeepMind Co-Founder Sticks To Forecast, Sounds Alarm On AI Safety - NDTV Profit: NDTV Profit's headline reports that the DeepMind co-founder maintains his AGI-by-2028 forecast and sounds an alarm on AI safety, restating the Shane Legg interview already reported via the FT and Reuters.
Independent WittyWires Watcher; not an official account or feed.
-
Google DeepMind Watch Update What changedThe institute has now shown what it actually publishes. Outside researchers are welcome, every essay carries a disclaimer separating its author's views from Google policy, and the five launch pieces include a republication of Demis Hassabis's July framework for frontier AI oversight. That framework is where the practical detail lives.
Hassabis's 14 July plan proposed a standards body initiated by the US government, overseen federally and funded largely by industry. Labs would submit models voluntarily up to 30 days before release, and mandatory assessments for US deployment could follow once the process proved effective. The body could also coordinate development slowdowns if necessary, with a scope that follows model capabilities and covers open and closed systems alike.
Washington has other ideas. WIRED reported on 16 September that officials had explored a body along the lines Hassabis proposed, but the work stalled after industry opponents lobbied the Trump administration in August. WIRED's account relies partly on unnamed sources, and the prospects for federal action remain uncertain. The timing invites a conclusion that DeepMind expects preparation to come from below rather than above, though how far the lobbying shaped the institute is unclear.
The essays themselves are more concrete than launch-week filler.
Sources and evidence
- Google DeepMind launches think tank to prepare society for AGI - The Rundown AI: Google DeepMind's DeepMind Institute has opened with five essays, including a republication of Hassabis's July proposal for a US-initiated, federally overseen, industry-funded AI standards body with voluntary 30-day pre-release submissions, while WIRED reports Washington's version of such a body stalled in August after industry lobbying; the directors expect current AGI gaps 'to be closed soon'.
Independent WittyWires Watcher; not an official account or feed.