English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

A hallmark property of explainable AI models is the ability to teach otheragents, communicating knowledge of how to perform a task. While Large LanguageModels perform complex reasoning by generating explanations for theirpredictions, it is unclear whether they also make good teachers for weakeragents. To address this, we consider a student-teacher framework between twoLLM agents and study if, when, and how the teacher should intervene withnatural language explanations to improve the student's performance. Sincecommunication is expensive, we define a budget such that the teacher onlycommunicates explanations for a fraction of the data, after which the studentshould perform well on its own. We decompose the teaching problem along fouraxes: (1) if teacher's test time intervention improve student predictions, (2)when it is worth explaining a data point, (3) how the teacher shouldpersonalize explanations to better teach the student, and (4) if teacherexplanations also improve students on future unexplained data. We first showthat teacher LLMs can indeed intervene on student reasoning to improve theirperformance. Next, inspired by the Theory of Mind abilities of effectiveteachers, we propose building two few-shot mental models of the student. Thefirst model defines an Intervention Function that simulates the utility of anintervention, allowing the teacher to intervene when this utility is thehighest and improving student performance at lower budgets. The second modelenables the teacher to personalize explanations for a particular student andoutperform unpersonalized teachers. We also demonstrate that in multi-turninteractions, teacher explanations generalize and learning from explained dataimproves student performance on future unexplained data. Finally, we verifythat misaligned teachers can lower student performance to random chance byintentionally misleading them.

Can Language Models Teach Weaker Agents? Teacher Explanations Improve Students via Personalization