Original post
Séb Krier#505
Séb Krier#505
Ben Goldhaber@BenGoldhaber
Gwern describes current AI training and architecture leading to a slop-pocalypse
11:34 AM · Jun 6, 2026 · 32.4K Views
Gwern argues frontier LLMs are structurally lazy, limiting major productivity gains and leading to an automated 'slop-pocalypse
Scaling will only improve generic corporate tasks.
153222016846.3K
Sentiment
Some users endorse Gwern's warning that current AI training architectures will cause a Slop-Pocalypse as based and correct, while others dismiss the writing as poor or mock predicted model addiction.
Pos
33.3%
Neg
66.7%
3 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS473LIKES6
Andrew Curran@AndrewCurran_
@BenGoldhaber @sebkrier I say Slopocalypse. But I respect your hyphen.
14hViews 473Likes 6
BOOKMARKS1
Ben Goldhaber@BenGoldhaber
@AndrewCurran_ @sebkrier I assume in a few weeks we'll get the definitive NYTimes style guide answer
12hViews 18Likes 2Bookmarks 1
RETWEETS5
Ben Goldhaber@BenGoldhaber
Gwern describes current AI training and architecture leading to a slop-pocalypse
15hViews 32.4KLikes 303Bookmarks 151
REPLIES1
Cleber Bruchez@thebruchez
@Noahpinion I get the feeling, what part of the scenario would make it funniest to you, and what part makes it feel bad-ish?
11hViews 45
Trinley Goldenberg@mattgoldenberg
@BenGoldhaber is this a google doc or something? can't find it on his site
12hViews 31
Ben Goldhaber@BenGoldhaber
articulating the principles a personal guardian angel AI should follow
15hViews 64Likes 1
Ben Golub@ben_golub
@AndrewCurran_ @BenGoldhaber @sebkrier Aslopalypse
9hViews 17Likes 2Bookmarks 0
Ben Goldhaber@BenGoldhaber
@mattgoldenberg https://gwern.net/guardian-angel
12hViews 17Likes 1
Ben Golub@ben_golub
@AndrewCurran_ @BenGoldhaber @sebkrier That's the New Yorker
9hViews 9Likes 1Bookmarks 0
kernel draft@coloneldraught
@BenGoldhaber this is very bad writing. who is he writing for? an excel macro?
11hViews 20
Brad Neuberg@bradneuberg
@BenGoldhaber @AndrewCurran_ @sebkrier Well, I coined the term Coworking (no hyphen), but The Press hates writing it without a hyphen, so we get co-working
8hViews 5Likes 1
Life in Robots@jfarrow
@BenGoldhaber @yacineMTB Harmful hardwired personalities? They are a verbal progressive reflection of their paired user/context, that’s all.
12hViews 12
Adel Bucetta@adelbucetta
@BenGoldhaber the honest answer is that we're still way more concerned about optimizing ai for current tasks than understanding how to prevent it from breaking down when scaled. nobody's talking about the maintenance costs of those massive models yet
10hViews 11
M Barnes Policy@mb_policy
@BenGoldhaber The substitution thesis assumes regulatory passivity. OpenAI-G42 Stargate required governance infrastructure before deployment. Regulatory readiness varies by jurisdiction, and that variance shapes outcomes more than model architecture.
5hViews 7
Andrew Curran@AndrewCurran_
@BenGoldhaber @sebkrier They say Slöpöcalypse.
12hViews 4
Kris Schnee@KrisSchnee
@BenGoldhaber I can imagine a style of architecture that nobody actually likes, but which is unusually good for robot navigation. One example would be super-narrow, dark hallways full of rails.
12hViews 3
NobodyExistsOnTheInternet@nullvaluetensor
@BenGoldhaber You WILL become addicted to Claude, so that anthropic WILL hit 100B ARR.
You WILL love the model. You WILL worship the model. The model WILL make you dependent on it and you WILL mindlessly spend tokens on your subscription.
14hViews 3
dionizije@dionizije_fa
@BenGoldhaber Based and correct
13h