AI models' quoting accuracy varies significantly, ELL label impact unclear

A researcher investigated whether labeling a student as an English learner affects how AI models quote their writing, but found insufficient data to draw conclusions. Two small open-source models, Qwen2.5-1.5B-Instruct and SmolLM2-1.7B-Instruct, were tested, with Qwen2.5 generally quoting accurately but exceeding length limits, while SmolLM2 often failed to quote faithfully. The study highlighted that model performance variations are more significant than the impact of the ELL label, and neither model consistently provided the short, exact quotes required for the experiment. AI

IMPACT Highlights significant performance differences between AI models and the limited impact of student labeling on quote accuracy.

RANK_REASON Research study on AI model behavior with insufficient data. [lever_c_demoted from research: ic=1 ai=1.0]

Read on dev.to — LLM tag →

• Qwen2.5-1.5B-Instruct
• SmolLM2-1.7B-Instruct

• model release
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →