LLMs are (mostly) not helped by filler tokens — LessWrong
This work was done at Redwood Research. The views expressed are my own and do not necessarily reflect the views of the organization. Thanks to Ryan Greenblatt, Fabien Roger, and Jenny Nitishinskaya for running some of the initial experiments and to Gabe Wu and Max Nadeau for revising this post. I conducted experiments to see if language models could use 'filler tokens'—unrelated text output before the final answer—for additional computation. For instance, I tested whether asking a model to produce "a b c d e" before solving a math problem improves performance. This mini-study was inspired by Tamera Lanham, who ran similar experiments on Claude and found negative results. Similarly, I find that GPT-3, GPT-3.5, and Claude 2 don’t benefit from filler tokens. However, GPT-4 (which Tamera didn’t study) shows mixed results with strong improvements on some tasks and no improvement on others. These results are not very polished, but I’ve been sitting on them for a while and I thought it was be
x LLMs are (mostly) not helped by filler tokens — LessWrong Language Models (LLMs) AI Capabilities Redwood Research AI Frontpage 68 LLMs are (mostly) not helped by filler tokens by Kshitij Sachan 10th Aug 2023 7 min read 36 68 This work was done at Redwood Research. The views expressed are my own and do not necessarily reflect the views of the organization. Thanks to Ryan Greenblatt, Fabien Roger, and Jenny Nitishinskaya for running some of the initial experiments and to Gabe Wu and Max Nadeau for revising this post. I conducted experiments to see if language models could use 'filler tokens'—u
Explore this link on the map →related reading
- gpt-4.pdfcdn.openai.com
- Patterns for Building LLM-based Systems & Productseugeneyan.com
- State of RL for reasoning LLMs | A. Weersaweers.de
- Transformer Explainer: LLM Transformer Model Visually Explainedpoloclub.github.io
- GenAI Handbookgenai-handbook.github.io
- How LLMs Work, Explained Without Math - miguelgrinberg.comblog.miguelgrinberg.com
- Extending Context is Hard | kaiokendevkaiokendev.github.io
- LLM Inference Performance Engineering: Best Practices | Databricks Blogdatabricks.com
- How LLMs Actually Work | 0xkato0xkato.xyz
- Getting Caught Up to Modern LLM Research | Samarth Goeldev.samarthgoel.com
- The Llama Hitchiking Guide to Local LLMs – hackerllamaosanseviero.github.io
- I blame the tokenizer | David Quareldavidquarel.github.io