Looped/recurrent-depth transformer architectures can improve model quality at a fixed compute budget when the model is large enough.

Added to korrents 20 Sept 2026 · How quotes work · Something wrong? Tell us

Shared subject first, then similar wording. They may agree or disagree.

• Scaling up language models greatly improves task-agnostic few-shot performance, sometimes matching prior fine-tuning approaches. DADario Amodei28 May 2020
• Residual networks are easier to optimize than plain ones and keep gaining accuracy as depth increases. KHKaiming He10 Dec 2015
• Test-time scaling seems to have gained a third axis: latent space reasoning iterations in looped transformers. FCFrançois Chollet5 Sept 2026
• Sequence transduction can use a simple architecture based solely on attention, without recurrence or convolutions. AVAshish Vaswani12 Jun 2017
• Larger language models already have enough capacity that extra per-layer embedding tricks add little benefit to them. SRSebastian Raschka16 May 2026

See this claim on the map