AI Tutor Shows Strong Learning Gains in Dartmouth Course Study

·1 views

A new AI tutoring system has demonstrated notable effectiveness in a Dartmouth college course, achieving an effect size between 0.71 and 1.30 standard deviations. Effect sizes in this range are generally considered large by educational research standards, suggesting meaningful improvements in student learning outcomes. The findings were presented in a paper submitted to an educational technology workshop hosted by Utrecht University in 2026. The study adds to a growing body of research exploring the potential of AI-driven tools to enhance academic performance at the university level.

Read the full story at Hacker News

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

FakerForge Lets You Generate Realistic Dummy Data for Excel Spreadsheets Free

FakerForge is a free online tool that generates realistic dummy data for spreadsheet templates, exporting results as CSV files compatible with Excel, Google Sheets, and similar apps. Users define their columns using plain-language schema descriptions, and the tool automatically maps each column to an appropriate data type — assigning real-looking emails, valid dates, and sensible prices accordingly. Custom value constraints can also be applied, such as limiting a category column to a specific list of options. The free plan supports up to 100 rows per table, which the developers say is sufficient for most mockups, demos, and formula testing. The tool is positioned as an alternative to manual data entry or repurposing real customer data, both of which carry practical and compliance drawbacks.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Cursor 3 Launches as Agent-First IDE with Parallel AI Execution and Rebuilt Interface

Cursor released version 3 on April 2, 2026, describing it as a ground-up rebuild rather than an incremental update to its existing editor. The new release replaces the previous Composer pane with an Agents Window, a full-screen workspace where developers can monitor and manage multiple AI agents running simultaneously. Each agent works independently on an assigned scope — such as refactoring, testing, or documentation — and stages changes in separate worktrees to prevent conflicts. This marks a fundamental shift from the earlier model, where AI assistance was a sidebar tool and agentic tasks ran one at a time in a single blocking session. The redesign positions agent delegation and supervision, rather than direct code authoring, as the primary way developers interact with the tool.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Why Your Scoring System Needs Built-In Explanations, Not Separate Logic

A software developer describes a recurring problem where a lead-scoring system produced numbers that could not be readily explained in review meetings, undermining confidence in the output. The root cause was a common architectural pattern: the scoring function and the explanation function were written separately, each hardcoding the same thresholds independently. When a spend threshold was updated in the scoring function but not in the explanation builder, the system began producing scores and justifications that directly contradicted each other. The proposed fix is straightforward — emit the human-readable reason for each points decision at the exact same line of code where the points are awarded, so both must be updated together. This approach ensures that any change to scoring logic automatically surfaces in the explanation, eliminating silent drift between numbers and their stated rationale.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Per-Seat Pricing Fails AI Agent SaaS as Costs Decouple from User Count

A developer building an AI agent SaaS product discovered that per-seat pricing created unsustainable losses, with some single-user customers costing far more to serve than they paid. Unlike traditional software, AI agent costs are driven by computational usage — such as LLM calls and tool invocations — rather than the number of users accessing the product. This mismatch means a one-seat customer running complex queries can consume more margin than a ten-seat customer with light usage. The volatility of LLM inference costs and unpredictable agent behavior make flat-fee models especially risky for builders. Alternative pricing approaches being explored by the industry include token-based billing, per-action charges, per-task flat fees, and outcome-based pricing, each suited to different product types and customer expectations.

0 comments Read more at DEV Community