Technical testProvisional (rules only)
2026-10-16 · arXiv new: cs.CY, cs.HC, cs.CL, cs.AI · Study (preprint)
Large language models (LLMs) are increasingly being adopted as teaching tools in computing education, particularly for developing AI tutors capable of answering student questions. With the rise in enrollment of international students in computing programs, researchers have begun to explore the difficulties that...
Why this label for Multi-sourced AI Tutor for Non-Native English Speaker Students
- Technical test: provisional rules-only grade, pending grader consensus and review: R4 real-learner gate (F1/F2): Outputs of models, simulations or audits; any humans act as raters, references or data sources, not learners.
- Preprint: not peer reviewed; capped at Level 3
- Tutoring and personalization: matched 'Tutor'
Provisional label from fixed rules applied to the collected text; research items are re-graded by our review pipeline. Method
Level 1: Descriptive or exploratoryProvisional (rules only)
2026-10-13 · arXiv search: AI + education · Study (preprint)
Human-AI Interaction is commonly evaluated by what an artificial agent enables during use, yet interaction can also alter the human who acts next. We introduce Residual Human States, a trajectory-based framework for representing the potentially multidimensional human state carried forward after interaction with AI,...
Why this label for What Human-AI Interaction Leaves Behind: Residual Human States - A Trajectory-Ba
- Level 1: Descriptive: provisional rules-only grade, pending grader consensus and review: R5 design: descriptive (F3): Descriptive, qualitative, pilot or deployment study with real participants.
- Preprint: not peer reviewed; capped at Level 3
- Tutoring and personalization: matched 'tutoring'
- Assessment and grading: matched 'exam'
Provisional label from fixed rules applied to the collected text; research items are re-graded by our review pipeline. Method
Technical testProvisional (rules only)
2026-10-12 · arXiv search: AI + education · Study (preprint)
Large language models can answer the same question in different styles, but users' preferred styles are hidden and vary across users and questions. We study inference-time personalization through online selection from a fixed set of prompts, each representing a persona.
Why this label for Online LLM Personalization from Pairwise Preferences Across Users
- Technical test: provisional rules-only grade, pending grader consensus and review: R4 real-learner gate (F1/F2): Outputs of models, simulations or audits; any humans act as raters, references or data sources, not learners.
- Preprint: not peer reviewed; capped at Level 3
- Tutoring and personalization: matched 'tutoring'
Provisional label from fixed rules applied to the collected text; research items are re-graded by our review pipeline. Method
CommentaryProvisional (rules only)
2026-10-11 · arXiv search: AI + education · Study (preprint)
Large language models can produce fluent tutoring dialogue, but whether they actually help students learn remains unclear. We present a process-centered synthesis of LLM tutoring that follows the path from learner-state inference to pedagogical action, tutor training, evaluation, and student simulation.
Why this label for Do LLM Tutors Actually Teach? A Process-Centered Synthesis of the Tutoring Loop
- Commentary or coverage: provisional rules-only grade, pending grader consensus and review: R3 genre gate (F7): Self-described conceptual, legal, policy or reflective work with no learner data.
- Preprint: not peer reviewed; capped at Level 3
- Tutoring and personalization: matched 'Tutors'
Provisional label from fixed rules applied to the collected text; research items are re-graded by our review pipeline. Method
Technical testProvisional (rules only)
2026-10-08 · arXiv search: AI + education · Study (preprint)
It is assumed that a natural way to improve the pedagogical quality of large language model tutors is to define a metric of instructional performance and fine-tune the model against it. To test this strategy, we designed a metric of pedagogical adaptivity that scores each instructional decision in a learning sequence...
Why this label for The learner who does not learn: when optimizing a pedagogical metric degrades LL
- Technical test: provisional rules-only grade, pending grader consensus and review: R4 real-learner gate (F1/F2): Outputs of models, simulations or audits; any humans act as raters, references or data sources, not learners.
- Preprint: not peer reviewed; capped at Level 3
- Tutoring and personalization: matched 'tutoring'
Provisional label from fixed rules applied to the collected text; research items are re-graded by our review pipeline. Method
Technical testLearners: Adults not in a learning role (raters)Outcome: Model performancePreprintSimulated learners
2026-10-08 · arXiv (back catalog) · Study (preprint)
It is assumed that a natural way to improve the pedagogical quality of large language model tutors is to define a metric of instructional performance and fine-tune the model against it. To test this strategy, we designed a metric of pedagogical adaptivity that scores each instructional decision in a learning sequence...
Why this label for The learner who does not learn: when optimizing a pedagogical metric degrades LL
- Technical test: Two independent AI graders gave the same level.
- Review status: Confirmed against the published method (AI-assisted review): both graders agreed
- Grader A: R4 real-learner gate: educators rate tutor outputs across learner scenarios; no real learners are assessed.
- Grader B: R4 real-learner gate: 2,000 simulated learner scenarios, with educators rating model outputs; no learners, so T
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Tutoring and personalization: matched 'tutoring'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review): both graders agreed Method
Level 1: Descriptive or exploratoryProvisional (rules only)
2026-10-07 · arXiv search: AI + education · Study (preprint)
Corrective feedback is among the best-evidenced drivers of second-language acquisition, yet corrections delivered during lessons rarely accumulate into an actionable view of grammar mastery. Prompted frontier models can provide such a view from learner--tutor lesson transcripts, but they are costly at scale.
Why this label for Grammar Concept Annotation at Scale: Deployed Fine-Tuned Small Language Models O
- Level 1: Descriptive: provisional rules-only grade, pending grader consensus and review: R5 descriptive (F8 vendor self-report): Vendor describes its own internal work without design details; vendor hold applies.
- Preprint: not peer reviewed; capped at Level 3
- Feedback: matched 'Corrective feedback'
- Tutoring and personalization: matched 'tutor'
Provisional label from fixed rules applied to the collected text; research items are re-graded by our review pipeline. Method
CommentaryProvisional (rules only)
2026-10-06 · arXiv search: AI + education · Study (preprint)
In K-12 mathematics tutoring, student-tutor dialogue provides rich evidence of learners' problem-solving processes and sources of difficulty. Learning analytics research increasingly relies on large language models (LLMs) to extract such information from dialogue for a variety of downstream tasks, including knowledge...
Why this label for Agreement Is Not Validity: Cross-Model LLM Consensus in Diagnosing Student Failu
- Commentary or coverage: provisional rules-only grade, pending grader consensus and review: R3 genre gate: No empirical learner data described; conceptual, framework, policy or argument paper.
- Preprint: not peer reviewed; capped at Level 3
- Tutoring and personalization: matched 'Tutoring'
Provisional label from fixed rules applied to the collected text; research items are re-graded by our review pipeline. Method
No study cited
2026-10-05 · Product Hunt · Launch / announcement
An AI tutor that teaches at a whiteboard, not a chat box
Why this label for Albie
- No study cited: launch or announcement; no traceable study
- Tutoring and personalization: matched 'tutor'
Provisional label from fixed rules applied to the collected text; research items are re-graded by our review pipeline. Method
Commentary
2026-09-29 · EdTech Digest · News
The families with the most resources are buying more human support, not less. GUEST COLUMN | by Alicia Kacharia There is a lot of excitement about how, for the first time, every student can access a personalized tutor regardless of income, location, or level.
Why this label for AI Will Work Best for the Students Who Need It the Least
- Commentary or coverage: news, commentary or policy with no study
- Tutoring and personalization: matched 'tutor'
Provisional label from fixed rules applied to the collected text; research items are re-graded by our review pipeline. Method
No study cited
2026-09-29 · Product Hunt · Launch / announcement
An AI tutor on an infinite whiteboard, not a chat thread
Why this label for Gauth Unlimited Digital Canvas
- No study cited: launch or announcement; no traceable study
- Tutoring and personalization: matched 'tutor'
Provisional label from fixed rules applied to the collected text; research items are re-graded by our review pipeline. Method
Level 1: Descriptive or exploratoryLearners: Higher-ed studentsOutcome: Behavior or engagementPreprintResearcher-made measureSimulated learners
2026-09-29 · arXiv (back catalog) · Study (preprint)
When a student is stuck, a tutor faces the assistance dilemma: help given too early can hinder productive struggle, while help withheld too long leaves the student in a frustrating, persistent impasse (i.e., wheel spinning). Generative AI tutors increasingly use guardrails restricting answer-giving, yet little is...
Why this label for Examining Variation in How Guided AI Tutors Resolve Student Impasses
- Level 1: Descriptive or exploratory: Real tutoring logs analyzed descriptively, then simulated conditions; graded on the real arm as 1.
- Review status: Confirmed against the published method (AI-assisted review)
- Grader A: R5 design ceiling: associations in authentic student sessions support a correlational level; simulated tutor conditions do not raise it.
- Grader B: R4 mixed study graded on the real-learner arm only: correlational analysis of authentic sessions (Level 2), downgraded under R6 measurement because the impasse/recovery coding has no reported reliability
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Tutoring and personalization: matched 'Tutors'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review) Method
Level 2: CorrelationalLearners: Higher-ed studentsOutcome: Learning or achievementPreprintPreregisteredSmall or imprecise
2026-09-21 · arXiv (back catalog) · Study (preprint)
Students increasingly study alongside generative artificial intelligence (AI), yet unguided access to fluent answers invites cognitive offloading, and there is little evidence on which configurations of AI tutoring produce learning. Two design margins are usually bundled together: pedagogical structure (how the tutor...
Why this label for When AI Tutors Speak: Evidence from a Randomized Field Experiment
- Level 2: Correlational: Consensus level confirmed against the methodology.
- Review status: Confirmed against the published method (AI-assisted review)
- Grader A: R5 randomized 86 MBA students with a learning outcome starts at Level 3; R6 imprecision lowers it because no interval or power information is reported.
- Grader B: R5: a preregistered randomized field experiment with a learning outcome starts at Level 3; R6 imprecision (n=86, under 100, and only p-values with no interval or power information) lowers it to Level 2.
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Cognitive load: matched 'cognitive offloading'
- Tutoring and personalization: matched 'Tutors'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review) Method
Level 1: Descriptive or exploratoryLearners: Adult learnersOutcome: Learning or achievementPeer-reviewed (journal reference listed)Small or impreciseShort exposureResearcher-made measure
2026-09-01 · arXiv (back catalog) · Study
Does unrestricted AI access bypass the cognitive effort required for learning, or does it streamline knowledge acquisition? This paper reports on a study where we compare three designs for user-AI interaction in a learning context: (1) an unrestricted conversational bot like ChatGPT, (2) a pedagogically constrained...
Why this label for Socrates went Nuclear: Comparing Interaction Strategies for AI systems in a Lear
- Level 1: Descriptive or exploratory: Consensus level confirmed against the methodology.
- Review status: Confirmed against the published method (AI-assisted review)
- Grader A: R5 design ceiling permits Level 2 for a pre/post comparison without stated equivalence; R6 imprecision and measurement lower it to Level 1 for 50 learners and an unvalidated immediate test.
- Grader B: R5 three-condition comparison with pre-test, capped at 3; R7 short-exposure cap applies; R6 imprecision (n=50, d and p only, no interval) lowers to 2; R6 measurement (researcher-made immediate post-test, no reliability) lowers to 1; construct flag: the authors say the chatbot's gains reflect answer retrieval rather than deeper learning
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Cognitive load: matched 'cognitive effort'
- Tutoring and personalization: matched 'tutoring'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review) Method
Level 1: Descriptive or exploratoryLearners: Higher-ed studentsOutcome: Learning or achievementPreprintResearcher-made measureSmall or imprecise
2026-08-25 · arXiv (back catalog) · Study (preprint)
Worked examples are an important part of introductory programming, but reading their expert explanations is passive. Self explanation, students explaining the problem and its solution to themselves with subgoal level analysis, converts passive reading into an active study of worked example, yet it is hard to scale...
Why this label for Self-Explanation Tutor for Active Study of CS1 Worked Examples
- Level 1: Descriptive or exploratory: Consensus level confirmed against the methodology.
- Review status: Confirmed against the published method (AI-assisted review)
- Grader A: R5 design ceiling: the course deployment reports student revisions and learning evidence without a stated comparison group.
- Grader B: R4 mixed study, so only the real-learner arm is graded (the LLM-vs-human-rater arm is T); that arm is a single-group course deployment, so R5 sets Level 1
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Worked examples and explicit instruction: matched 'Worked Examples'
- Tutoring and personalization: matched 'Tutor'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review) Method
Level 1: Descriptive or exploratoryLearners: K-12 studentsOutcome: Behavior or engagementPeer-reviewed (journal reference listed)Vendor-linkedPerception only
2026-08-07 · arXiv (back catalog) · Study
Many AI tutors leverage large language models (LLMs) today. Given that LLMs are opaque black boxes, robust evaluation and live experimentation to measure the impact of every change are essential.
Why this label for Methodologies for Improving the Quality of AI Tutoring in K-12 Education
- Level 1: Descriptive or exploratory: Consensus level confirmed against the methodology.
- Review status: Confirmed against the published method (AI-assisted review)
- Grader A: R5 descriptive ceiling: product metrics and experiments are described without an explicit learner-assignment design or learning outcome.
- Grader B: R5 design ceiling: a vendor (Khan Academy) describes its metrics and unspecified experiments without design or sample details, so it is descriptive; R7 engagement-only cap also applies.
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Motivation and engagement: matched 'student engagement'
- Tutoring and personalization: matched 'Tutoring'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review) Method
Technical testLearners: Simulated or AI-onlyOutcome: Model performancePreprintSimulated learners
2026-08-04 · arXiv (back catalog) · Study (preprint)
Large language models (LLMs) power educational applications from tutoring to essay scoring, but each is a point solution to a single task, and only recently have these point solutions been integrated into agents operating over a learning management system (LMS). Yet tutoring is long-horizon, since a learner improves...
Why this label for EduClaw-Bench: A Long-Horizon Benchmark for Pedagogical LLM Agents with Simulate
- Technical test: Two independent AI graders gave the same level.
- Review status: Confirmed against the published method (AI-assisted review): both graders agreed
- Grader A: R4 real-learner gate: the reported tutoring results come from simulated learners; the classroom check validates the simulation.
- Grader B: R4 real-learner gate: agent tutors are benchmarked on simulated learners; the live-classroom study only calibrates the simulator and reports no learning outcome from real learners.
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Tutoring and personalization: matched 'tutoring'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review): both graders agreed Method
Technical testLearners: Adults not in a learning role (raters)Outcome: Perception or attitudePreprintShort exposurePerception onlySmall or imprecise
2026-08-01 · arXiv (back catalog) · Study (preprint)
This innovative practice full paper presents BoilerSketch, a TA-supervised, diagram-first GenAI practice and tablet interface for providing structured visual explanations in CS1 and early CS2 support settings. Large early computing courses routinely face a support bottleneck during labs and office hours because many...
Why this label for BoilerSketch: A TA-Supervised, Diagram-First GenAI Practice for Structured Diagr
- Technical test: Tool paper; no learner outcomes reported.
- Review status: Confirmed against the published method (AI-assisted review)
- Grader A: R4 real-learner gate: instructional staff evaluated a prototype; no learner outcomes were studied.
- Grader B: R5 design ceiling: one 45-minute expert evaluation with a post-use survey of 21 instructional staff, with no comparison group, so descriptive Level 1 and no learner outcome
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Tutoring and personalization: matched 'tutoring'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review) Method
Technical testLearners: Not reportedOutcome: Model performancePreprintSmall or imprecise
2026-07-24 · arXiv (back catalog) · Study (preprint)
Most educational technology for children is built around visual interfaces, which excludes the many children worldwide who live with visual impairment -- an estimated 1.4 million children are blind and many more have low vision. We present Kutti AI, a voice-first learning companion designed so that audio is the...
Why this label for Kutti AI: A Voice-First, Offline-Capable Learning Companion with Real-Time Strug
- Technical test: Prototype with no evaluation with target children.
- Review status: Confirmed against the published method (AI-assisted review)
- Grader A: R4 real-learner gate: the hackathon prototype has no reported evaluation with its intended child learners.
- Grader B: R5 design ceiling: hackathon prototype with informal qualitative observations and no evaluation with target children yet, so descriptive or pilot Level 1
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Tutoring and personalization: matched 'adaptive learning'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review) Method
Level 1: Descriptive or exploratoryLearners: Higher-ed studentsOutcome: Learning or achievementPreprintSmall or impreciseResearcher-made measureShort exposure
2026-07-17 · arXiv (back catalog) · Study (preprint)
Generative AI (GenAI) is increasingly used by students for programming explanation, debugging, and assignment support. Yet unrestricted large language model (LLM) tutors can hallucinate, contradict course policy, reveal complete solutions, and foster passive dependence.
Why this label for EduGuard: A Safe RAG-Based LLM Tutor for Programming Education
- Level 1: Descriptive or exploratory: Consensus level confirmed against the methodology.
- Review status: Confirmed against the published method (AI-assisted review)
- Grader A: R5 controlled pilot with 10 learners has a Level 2 ceiling; R6 imprecision lowers it to Level 1.
- Grader B: R4 mixed study graded only on the real-learner arm. That arm is a counterbalanced pilot with 10 undergraduates and an immediate post-test only, and pilots are Level 1 under sec. 6. The benchmark part would be T.
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Tutoring and personalization: matched 'Tutor'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review) Method
Level 1: Descriptive or exploratoryLearners: Higher-ed studentsOutcome: Model performancePreprintSimulated learnersSmall or imprecise
2026-07-15 · arXiv (back catalog) · Study (preprint)
This paper is an extension of a paper presented at the ICAART 2026 conference, which introduced LEA (Learning Engagement Assistant), an adaptive AI tutoring agent combining course-specific Retrieval-Augmented Generation (RAG) with structured Knowledge Component (KC) models across integrated Chat, Tutor, and Quiz...
Why this label for Learning Engagement Assistant (LEA): Cross-Course Scalability and Classroom Eval
- Level 1: Descriptive or exploratory: Change to 1. First real classroom deployment with 8 students plus simulation; graded on the real-learner arm as descriptive.
- Review status: Confirmed against the published method (AI-assisted review)
- Grader A: R4 real-learner gate: despite a classroom deployment with eight students, the reported evaluations measure system performance rather than learner outcomes.
- Grader B: R4 mixed study graded only on the real-learner arm, which is a descriptive classroom deployment with 8 students and no comparison group (Level 1). The RAGAS and simulation parts would be T.
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Tutoring and personalization: matched 'Tutoring'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review) Method
Commentary
2026-07-14 · eSchool News · News
In an era defined by rapid technological advancement and an unprecedented explosion of information, public education stands at a pivotal and exciting crossroads.
Why this label for The best time to be a teacher: Leading with AI and technology in the age of pers
- Commentary or coverage: news, commentary or policy with no study
- Tutoring and personalization: matched 'personalized learning'
Provisional label from fixed rules applied to the collected text; research items are re-graded by our review pipeline. Method
Level 1: Descriptive or exploratoryLearners: Higher-ed studentsOutcome: Learning or achievementPeer-reviewed (journal reference listed)Small or imprecise
2026-07-03 · arXiv (back catalog) · Study
While Large Language Models (LLMs) can provide personalized support in learning, several studies have raised concerns regarding their use in education. Importantly, learning depends on how students engage with LLMs.
Why this label for Reflective Dialogue or Prompt Refinement? Effects of Tutor Scaffolding on Studen
- Level 1: Descriptive or exploratory: Two-condition course study, 66 then 52 students, no baseline equivalence, n under 100; 1.
- Review status: Confirmed against the published method (AI-assisted review)
- Grader A: R6 imprecision: the comparison reports no interval or power information and has fewer than 100 learners.
- Grader B: R5 two-tutor comparison over 6 weeks with learning gains (Level 3 ceiling; random assignment not stated), lowered one level by R6 imprecision (n=66, no interval reported)
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Tutoring and personalization: matched 'Tutor'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review) Method
Level 1: Descriptive or exploratoryLearners: K-12 studentsOutcome: Learning or achievementPreprintSmall or impreciseShort exposure
2026-06-26 · arXiv (back catalog) · Study (preprint)
Generative AI systems (GenAI) are increasingly used by students as learning companions, yet little is known about how they regulate these tools in open-ended learning settings, where the goal is to improve understanding for later independent performance. This study examined Grade-9 students' use of a general-purpose...
Why this label for Changes in Help-Seeking Strategies Predict unaided Performance during AI-based M
- Level 1: Descriptive or exploratory: Consensus level confirmed against the methodology.
- Review status: Confirmed against the published method (AI-assisted review)
- Grader A: R6 imprecision: the controlled association uses 89 learners with computable trajectories and reports no interval or power information.
- Grader B: R5 correlational design with pre-test control (Level 2 ceiling) on Grade-9 students, lowered one level by R6 imprecision (analytic n=95/89, no interval reported)
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Metacognition and self-regulation: matched 'self-regulated'
- Tutoring and personalization: matched 'tutor'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review) Method
Level 1: Descriptive or exploratoryLearners: K-12 studentsOutcome: Model performancePreprint
2026-06-17 · arXiv (back catalog) · Study (preprint)
Personalised learning systems often assume that mathematical ability is combined of discrete abilities, acquired sequentially and dependent upon first acquiring foundational abilities, and students often report different strengths. In this work, we explore the validity of these assumptions by applying clustering...
Why this label for Archetypes or ability? Clustering for modelling student mathematical competence
- Level 1: Descriptive or exploratory: Change to 1. Secondary analysis of exam data from 119,034 real students; descriptive and modelling study, not a model benchmark.
- Review status: Confirmed against the published method (AI-assisted review)
- Grader A: R4 real-learner gate: exam results train and benchmark predictive models; no learner outcome is evaluated.
- Grader B: R5 design: exploratory clustering of exam data from 119,034 students with no comparison or intervention, so it is descriptive (Level 1)
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Tutoring and personalization: matched 'Personalised learning'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review) Method
Level 2: CorrelationalLearners: Not reportedOutcome: Learning or achievementPreprintSmall or imprecise
2026-06-16 · arXiv (back catalog) · Study (preprint)
Motivational factors such as self-efficacy and how favorably students feel toward practice play a crucial role in shaping learning, particularly in technology-supported environments. Yet, educational interventions often overlook how these factors interact with practice format.
Why this label for Self-Efficacy and Favorability Shape Learning from Tutoring Systems and Paper Pr
- Level 2: Correlational: Counterbalanced within-subject design with learning outcome; no sample size or interval reported, so held at 2.
- Review status: Confirmed against the published method (AI-assisted review)
- Grader A: R5 design ceiling: counterbalanced within-subject comparison uses matched problem sets and baseline self-efficacy, but Level 4 checks are unavailable.
- Grader B: R5: the headline finding is a moderation link between self-efficacy or favorability and learning gains (correlational, Level 2) inside a counterbalanced within-subject ITS vs paper comparison; n and intervals are not reported, so it stays at 2 pending review (accepted at EC-TEL 2026, not matched)
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Motivation and engagement: matched 'students' motivation'
- Tutoring and personalization: matched 'Tutoring'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review) Method
Technical testLearners: Simulated or AI-onlyOutcome: Model performancePreprint
2026-06-15 · arXiv (back catalog) · Study (preprint)
Large language models are increasingly proposed as educational tutors, yet stronger task-solving ability does not necessarily imply stronger learning support. Motivated by recent calls to measure the social impact of NLP systems in practice, we study whether public LLM tutoring benchmarks distinguish...
Why this label for Beyond Helpfulness: A Teaching-over-Solving Diagnostic for Measuring Educational
- Technical test: Both graders agree; rules version missed it. Consensus level confirmed.
- Review status: Confirmed against the published method (AI-assisted review)
- Grader A: R4 real-learner gate: diagnostic compares model benchmark scores without learner outcomes.
- Grader B: R4 real-learner gate: secondary analysis of tutoring benchmark leaderboards; the only outcome is model performance
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Tutoring and personalization: matched 'Tutors'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review) Method
Technical testLearners: Higher-ed studentsOutcome: Model performancePeer-reviewed (journal reference listed)
2026-06-12 · arXiv (back catalog) · Study
Educational chatbots powered by large language models (LLMs) show promising effects on learning outcomes, yet most systems delegate pedagogical decisions such as content selection and didactic structuring implicitly to the LLM, making tutoring strategies difficult to trace, evaluate, and reproduce. This paper presents...
Why this label for A didactical-driven teacher assistant for a dimensional modeling course
- Technical test: Two independent AI graders gave the same level.
- Review status: Confirmed against the published method (AI-assisted review): both graders agreed
- Grader A: R4 real-learner gate: authentic student questions are test inputs, while the outcome is system detection precision.
- Grader B: R4 real-learner gate: student questions are only test inputs; the only outcome is the system's detection precision
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Tutoring and personalization: matched 'tutoring'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review): both graders agreed Method
Commentary
2026-05-30 · Carl Hendrick (learning science) · Commentary
AI tutoring will work when it stops pretending to be human
Why this label for Spandrels of the Mind: The Case Against Humanlike AI
- Commentary or coverage: news, commentary or policy with no study
- Tutoring and personalization: matched 'tutoring'
Provisional label from fixed rules applied to the collected text; research items are re-graded by our review pipeline. Method
Technical testLearners: Simulated or AI-onlyOutcome: Model performancePreprintSimulated learners
2026-05-12 · arXiv (back catalog) · Study (preprint)
Large language models (LLMs) can fluently generate student-like responses, making them attractive as simulated students for training and evaluating AI tutors and human educators. Yet such simulators are typically evaluated by output similarity to real students, not by whether they behave like students with coherent...
Why this label for Simulating Students or Sycophantic Problem Solving? On Misconception Faithfulnes
- Technical test: Two independent AI graders gave the same level.
- Review status: Confirmed against the published method (AI-assisted review): both graders agreed
- Grader A: R4 real-learner gate: all tested students are LLM simulators.
- Grader B: R4 real-learner gate: an evaluation and post-training of LLM simulated students, with no real learners
- Rule version: METHODOLOGY v0.3, pipeline v1, rules rules-v0.4
- Tutoring and personalization: matched 'tutors'
Graded blind by two AI graders against our published method. Confirmed against the published method (AI-assisted review): both graders agreed Method
