Summary of Do Llms Dream Of Elephants (when Told Not To)? Latent Concept Association and Associative Memory in Transformers, by Yibo Jiang et al.
Do LLMs dream of elephants (when told not to)? Latent concept association and associative memory in transformers
by Yibo Jiang, Goutham Rajendran, Pradeep Ravikumar, Bryon Aragam
First submitted to arxiv on: 26 Jun 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Machine Learning (cs.LG); Machine Learning (stat.ML)
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 explores the capabilities of Large Language Models (LLMs) in storing and retrieving facts. Through experiments on open-source models, it is found that LLMs can be easily manipulated by changing contexts without altering the factual meanings. This property is likened to an associative memory model, where specific tokens in the context serve as clues for fact retrieval. The paper mathematically investigates this property using transformers, the building blocks of LLMs, and demonstrates theoretically and empirically that they gather information through self-attention mechanisms. In particular, a one-layer transformer is used to solve a simple latent concept association problem. The findings highlight the potential limitations of LLMs in certain memory tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how Large Language Models (LLMs) remember and recall facts. They found that by changing the context, they can make the model retrieve different facts without actually changing what it knows. This is like a special kind of memory where certain words or phrases are clues to remembering specific information. The researchers studied how this works using simple computer algorithms and showed that LLMs use these algorithms to remember things. |
Keywords
* Artificial intelligence * Recall * Self attention * Transformer
