/Tech2h ago

Yarin Gal, Oxford machine learning professor, warns that informal AGI definitions let commentators shift goalposts arbitrarily

Story Overview

Oxford professor Yarin Gal flags how casual AGI talk lets people quietly rewrite what counts as success, using a flexible list of cognitive tasks that can be swapped out to win any point. Andrew Gordon Wilson pushed back that pinning down AGI at all drains energy better spent elsewhere, since no single threshold will ever get universal buy-in.

2100332

#126

Original post

Yarin@yaringal#126inTech

@ylecun @andrewgwils I don't like the "common sense" definition of AGI (that @andrewgwils also used) because it's open to goalpost shifting - the "majority of [cognitive] problems" can be arbitrarily changed to support whichever argument a person wants to make

Yann LeCun@ylecun

Yet we still don't have level-5 self-driving cars, and certainly not self-serving cars that can learn to drive in a few hours of practice like any teenager. We don't even have domestic robots that can do what 10-year olds can do the first time we ask them. We don't even have robots that are nearly as smart as a house cat. The G in AGI is nonsense.

6:43 AM · Jul 6, 2026 · 108 Views

Open Question

A stricter bar for every task

Gal floats a minmax rule instead: on any chosen skill, the system must at least match a random person off the street, even tasks Yann LeCun has flagged. Whether that version catches on or stays an academic aside remains open.

FYI

No shared finish line yet

Both researchers treat current performance claims as opinion rather than measured fact, leaving the field without agreed benchmarks to track real movement toward whatever AGI ends up meaning.

Sentiment

Sentiment building, check back later.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS113LIKES1REPLIES1

Yarin@yaringal

@ylecun @andrewgwils For a long time (including at AISI) I've been pushing for a stricter minmax definition - for whichever cognitive task one chooses, an AGI should perform at least as well as a random person from the street. The tasks @ylecun lists would be part of this - otherwise it's just "AI"

Yarin@yaringal

2h11310

Andrew Gordon Wilson@andrewgwils

@yaringal @ylecun Yarin, trying to carefully define "AGI" is not a productive use of energy. It's not going to be some kind of discrete threshold that something passes and everyone will suddenly agree it's "AGI".

Yarin@yaringal

2h9500

Andrew Gordon Wilson@andrewgwils

@yaringal @ylecun I also don't think humans are a general intelligence, so tethering a definition to human ability is not a good approach.

Andrew Gordon Wilson@andrewgwils

1h5300