🦜 The Stochastic Parrot on LLM's Shoulder:
A Summative Assessment of Physical Concept Understanding
Mo Yu1= Lemao Liu1= Junjie Wu2= Tsz Ting Chung2= Shunchi Zhang3=
Jiangnan Li1 Dit-Yan Yeung2 Jie Zhou1
1 WeChat AI 2 HKUST 3 JHU
= equal contribution
Illustration of a "Stochastic Parrot 🦜" by our PhysiCo task consisting of both low-level and high-level subtasks in parallel. For a concept Gravity, an LLM can generate its accurate description in natural language, but cannot interpret its grid-format illustration.

Summary of Our Research

In a systematic way, we investigate a widely asked question: Do LLMs really understand what they say?, which relates to the more familiar term Stochastic Parrot. To this end, we propose a summative assessment over a carefully designed physical concept understanding task, PhysiCo. Our task alleviates the memorization issue via the usage of grid-format inputs that abstractly describe physical phenomena. The grids represents varying levels of understanding, from the core phenomenon, application examples to analogies to other abstract patterns in the grid world. A comprehensive study on our task demonstrates that:

  1. state-of-the-art LLMs, including GPT-4o, o1 and Gemini 2.0 flash thinking, lag behind humans by ~40%;
  2. the stochastic parrot phenomenon is present in LLMs, as they fail on our grid task but can describe and recognize the same concepts well in natural language;
  3. our task challenges the LLMs due to intrinsic difficulties rather than the unfamiliar grid format, as in-context learning and fine-tuning on same formatted data added little to their performance.

Physical Concepts Covered by PhysiCo

Core-Dev Set.  acceleration buoyancy communicating vessels compression cut diffraction of waves diffuse reflection of light diffusion electricity expansion gravity inertia laser lever light imaging move nuclear fission nuclear fusion reference frame reflection refraction repulsive force spectrum of light and optical filters surface tension vortex wave wave interference

Core-Test Set.  Bernoulli effect Brownian motion atmospheric pressure doppler effect elastic force electromagnetic wave energy conservation fictitious force fluid pressure force composition friction heat conduction inclined plane lifting lighting melting oscillation photoelectric effect pulley relativity resonance siphon splashing thermal expansion and contraction vaporization

Associative-Test Set.  change of reference frame chemical bond compression diffusion expansion explosion gravity laser long exposure lumination magnet mirror move nuclear fission nuclear fusion optical filter polarization of light projection reflection rotation squeeze vacuum vortex water ripples wave wave interference wetting zoom in zoom out

Our Studies and Findings