/Tech3h ago

AI developer Teortaxes and research engineer Alexander Doria dispute claims that compute advantages dictate AI capability milestones

Story Overview

Two prominent AI voices are pushing back on the idea that raw compute leads alone set the pace for model breakthroughs, instead spotlighting China's recent GLM agent rollouts as proof that capability jumps can arrive through other routes even as US hardware advantages stretch further ahead.

358045.7K

#501

Original post

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex#501inTech

«america is 20 months ahead in terms of ai compute and i think progress should be mostly a function of compute» it really is striking that the US isn't hitting AI capability milestones anywhere close to 20 months earlier. GLM is already a research agent. Months have passed.

FleetingBits@fleetingbits

i think if glm-5.2 is as good as hyped then it raises a lot of questions about chinese progress; is it distillation or do us labs allocate compute badly or both?

america is 20 months ahead in terms of ai compute and i think progress should be mostly a function of compute

12:38 AM · Jun 23, 2026 · 4.5K Views

Open Question

Agent features arrive without matching scale

Zhipu AI's GLM-5 series and related updates have added long-horizon tool use, autonomous execution loops, and multi-step planning, yet exact training compute figures or head-to-head frontier benchmarks remain unconfirmed in public reporting.

Developer Impact

Dispute leaves timeline questions hanging

The back-and-forth centers on whether observed Chinese agent performance truly decouples from compute trends or simply reflects different deployment priorities, with no resolution offered on how quickly the gap might close.

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

VIEWS1.3KBOOKMARKS2LIKES24

Alexander Doria@Dorialexander

real question always was: what are you modeling exactly? compute is just the way to get there.

FleetingBits@fleetingbits

i think if glm-5.2 is as good as hyped then it raises a lot of questions about chinese progress; is it distillation or do us labs allocate compute badly or both?

america is 20 months ahead in terms of ai compute and i think progress should be mostly a function of compute

1h1.3K242

Sean@SomethingOnSnow

@teortaxesTex Enough chips for training. Then America gets saddled with running inference for everyone.

3h1451

dani@absenteewarlord

@teortaxesTex is there even a generally accepted theory of how they manage to do this?

3h141

Latest Developments in AI@DevelopmentsAI

@teortaxesTex Fable 5.1 is already ready and they are probably working on 5.2 right now.

2h56