Skip to main content Skip to navigation

Hints-In-Browser: Benchmarking Language Models for Programming Feedback Generation

Project Overview

The document examines the role of generative AI, particularly large language models, in programming education, emphasizing their ability to provide personalized feedback and hints to students. It introduces an innovative deployment method called 'hints-in-browser,' which enables these models to operate directly within users' browsers, thereby improving cost-efficiency and safeguarding data privacy. A benchmarking study evaluates several language models, including GPT-4 and Llama-3, assessing them based on quality, cost, and performance metrics. The results reveal substantial advantages of utilizing in-browser models for enhancing the educational experience in programming, while also acknowledging challenges such as the limitations posed by model size and consumer hardware capabilities. Overall, the findings underscore the potential of generative AI to transform programming education by delivering tailored support to learners.

Key Applications

Hints-in-Browser workflow for programming feedback generation

Context: Programming education targeting novice learners

Implementation: The model runs in the browser, generating feedback based on the learner's code and requests for hints.

Outcomes: No running costs for educators or learners, complete data privacy, and improved feedback quality with in-browser inference.

Challenges: Model size limitations and hardware requirements for effective inference.

Implementation Barriers

Cost and Time

High running costs for educators to provide feedback generation services using external models, along with long waiting times for learners while feedback is generated, which hampers interactivity.

Proposed Solutions: Utilizing in-browser models significantly reduces waiting times for feedback and eliminates the need for server-side processing, thereby reducing costs.

Data Privacy

Concerns over learner data being sent to external servers for processing.

Proposed Solutions: In-browser models keep user data local, enhancing privacy.

Project Team

Nachiket Kotalwar

Researcher

Alkis Gotovos

Researcher

Adish Singla

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Nachiket Kotalwar, Alkis Gotovos, Adish Singla

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies