2404.15758

N/A

Let’s Think Dot by Dot: Hidden Computation in Transformer Language Models Jacob Pfau, William Merrill & Samuel R. Bowman Center for Data Science New York University NY, NY 10012, USA {jp6263,willm,bowman}@nyu.edu arXiv:2404.15758v1 [cs.CL] 24 Apr 2024…

related reading

• [2404.15758] Let's Think Dot by Dot: Hidden Computation in Transformer Language Modelsarxiv.org
• Let's Think Dot by Dot: Hidden Computation in Transformer Language Modelsarxiv.org
• LLMs are (mostly) not helped by filler tokens — LessWronglesswrong.com
• 2402.12875arxiv.org
• A Mathematical Framework for Transformer Circuitstransformer-circuits.pub
• [2402.12875] Chain of Thought Empowers Transformers to Solve Inherently Serial Problemsarxiv.org
• Recent LLMs can use filler tokens or problem repeats to improve (no-CoT) math performanceblog.redwoodresearch.org
• Transformer Circuits Threadtransformer-circuits.pub
• dennyzhou.github.io/LLM-Reasoning-Stanford-CS-25.pdfdennyzhou.github.io
• Thinking like Transformersrush.github.io
• Transformers Provably Learn to Internalize Chain-of-Thoughtarxiv.org
• Transformer Explainer: LLM Transformer Model Visually Explainedpoloclub.github.io