Summary of Nyt-connections: a Deceptively Simple Text Classification Task That Stumps System-1 Thinkers, by Angel Yahir Loredo Lopez et al.
NYT-Connections: A Deceptively Simple Text Classification Task that Stumps System-1 Thinkers
by Angel Yahir Loredo Lopez, Tyler McDonald, Ali Emami
First submitted to arxiv on: 2 Dec 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 presents NYT-Connections, a benchmark designed to assess the ability of Large Language Models (LLMs) to engage in deliberate reasoning. The benchmark consists of 358 simple word classification puzzles derived from the New York Times Connections game, aimed at penalizing quick and intuitive “System 1” thinking. Six recent LLMs, including GPT-4, were evaluated alongside a simple machine learning heuristic and humans across three configurations. The findings show that even top-performing LLMs fall short of human performance by nearly 30%. Additionally, advanced prompting techniques like Chain-of-Thought and Self-Consistency exhibit diminishing returns as task difficulty increases. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about testing how well Large Language Models (LLMs) can think deeply. They created a special set of puzzles to make it hard for the models to just use quick guesses, but instead really understand what they’re doing. The LLMs didn’t do as well as humans on these puzzles, even though some are very good at other tasks. This test is unique because it helps figure out how well the models can think deeply and not just rely on shortcuts. |
Keywords
» Artificial intelligence » Classification » Gpt » Machine learning » Prompting
