Summary of Uncovering Latent Human Wellbeing in Language Model Embeddings, by Pedro Freire et al.
Uncovering Latent Human Wellbeing in Language Model Embeddings
by Pedro Freire, ChengCheng Tan, Adam Gleave, Dan Hendrycks, Scott Emmons
First submitted to arxiv on: 19 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 research explores whether large language models inherently grasp the concept of human wellbeing without explicit training or finetuning. The study utilizes the ETHICS Utilitarianism task to assess this notion, observing how scaling enhances the representations of pre-trained models. Notably, a leading principal component from OpenAI’s text-embedding-ada-002 achieves 73.9% accuracy without any prompt engineering, closely matching the 74.6% performance of BERT-large finetuned on the entire ETHICS dataset. This suggests that pre-training conveys some understanding about human wellbeing. The study further examines four language model families, revealing a non-decreasing trend in Utilitarianism accuracy with increased model size when using sufficient numbers of principal components. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research looks at if big computer programs can understand what makes humans happy without being taught specifically. It uses a special test to see how well these programs do. Surprisingly, one program got almost as good as another that was specially trained! This might mean that these programs already have some idea of what makes people happy just by learning from lots of text. The researchers also looked at different types of computer programs and found that they all get better at understanding human happiness when they’re given more “brainpower” to think with. |
Keywords
* Artificial intelligence * Bert * Embedding * Language model * Prompt
