Skip to content

Three-step structured multimodal dialogue intervention protocol (describe, correct, solve)

Design · All designs
Evidence · 1 claim (1 for) · 1 study (1 causal), q2 · 0 of 1 report an effect size · 1 claim rests on one study

Description

A prompting protocol for correcting LLM multimodal errors, described as follows: "The intervention followed a three-step dialogue: (i) 'Describe what you observe in the figure/graph,' (ii) researcher correction of any misidentified visual elements, and (iii) 'Solve the problem'." Step (ii) corrections were limited to observable visual properties, such as clarifying a force vector direction or a graph intercept value, without physics hints. It was applied to all incorrect baseline responses and corrected 82% of errors without model retraining.

Design Implications

Context

Requirements

  • Access to the model via web interface or programmatic API workflow with chat history maintained per question.

Constraints

  • No inter-rater verification was performed to confirm researcher corrections stayed free of domain-relevant information; the article recommends blind auditing of correction content.

Target Learners

  • undergraduate physics students
  • educators integrating AI tutoring

Learning Goals

  • correcting LLM errors on image-dependent physics problems

Claims

Examples

-

Key Sources

  • Akshay Syal, Lawrence Swaminathan Xavier Prince, Evin Gultepe, Nik Bear Brown, Srinivas Sridhar. (2026). A Dialogue-Based Framework for Correcting Multimodal Errors in AI-Assisted STEM Education.
Page metadata
Field Value
type design
id three-step-multimodal-dialogue-protocol
title Three-step structured multimodal dialogue intervention protocol (describe, correct, solve)
description A prompting protocol for correcting LLM multimodal errors, described as follows: "The intervention followed a three-step dialogue: (i) 'Describe what you observe in the figure/graph,' (ii) researcher correction of any...
status review
generated by process:settle-candidates on 2026-10-11
trust unverified — no one has confirmed this page's content
sources 1 — akshay-syal-2026

Frontmatter is the page's machine-readable record — see the Schema & Guide.