Summary of Marple: a Benchmark For Long-horizon Inference, by Emily Jin et al.
MARPLE: A Benchmark for Long-Horizon Inference
by Emily Jin, Zhuoyi Huang, Jan-Philipp Fränken, Weiyu Liu, Hannah Cha, Erik Brockbank, Sarah Wu, Ruohan Zhang, Jiajun Wu, Tobias Gerstenberg
First submitted to arxiv on: 2 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 MARPLE is a benchmark designed to test AI models’ ability to reason across long time horizons and make inferences from multiple sources of evidence. The benchmark features simulated households with procedurally generated environments and agent behaviors, providing visual, language, and auditory cues for agents to infer which one caused a change in the environment. Inspired by “whodunit” stories, participants are asked to identify the culprit as early as possible based on a step-by-step replay of events. The results show that human participants outperform traditional Monte Carlo simulation methods and an LLM baseline (GPT-4) on this task. While GPT-4 struggles with environmental changes, human performance is influenced by factors such as prior knowledge, context, and attention to detail. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine trying to figure out what happened in a mystery story from clues like pictures, words, and sounds. MARPLE is a test for AI models that helps them solve this kind of problem. It’s like a game where the AI has to watch a series of events, listen to sounds, and read words to find out who did something. The goal is to identify the culprit as soon as possible. In tests, humans were better than AI at solving these puzzles. This shows that there are still challenges for AI models when they need to reason over long periods of time. |
Keywords
» Artificial intelligence » Attention » Gpt
