Summary of Code Pretraining Improves Entity Tracking Abilities Of Language Models, by Najoung Kim et al.
Code Pretraining Improves Entity Tracking Abilities of Language Models
by Najoung Kim, Sebastian Schuster, Shubham Toshniwal
First submitted to arxiv on: 31 May 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper investigates whether pre-training language models on code improves their ability to track state changes of discourse entities in natural language. The authors compare pairs of language models, including base models and models trained with additional code data, to test this claim. They also examine the effect of math training and alignment tuning on entity tracking performance. The results show that models trained on large amounts of code outperform base models, while math training and alignment tuning do not consistently improve performance across various model families. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Code-trained language models can better track state changes of discourse entities in natural language. Researchers compared pairs of language models to see if adding code data improves their ability to follow these changes. They also looked at the effects of math training and making models easier to use. The study found that models trained on lots of code do a better job than regular models, but extra math practice or making it easier to use didn’t make a consistent difference. |
Keywords
» Artificial intelligence » Alignment » Discourse » Tracking
