Duolingo cut AI inference cost from 30¢ to under 1¢ per call, a 30× reduction, by switching to open-weight models. iabordeducation shipped 45 LLM providers, 316 models on a single grid with one-click swap.
Key points
Cost reduction: inference cost from 30¢ to <1¢ per call, 30× cheaper, using open-weight models.
Provider launch: 45 LLM providers, 316 models, one grid, one-click swap.
Sources
Doug Rohrer's team mixed vs blocked homework, 74% vs 42%
Doug Rohrer's team gave 126 seventh graders identical graph and slope problems for three months, testing mixed and blocked skill assignments. A surprise test 30 days after review showed 74% correct on the mixed skill and 42% on the blocked one.
Key points
126 seventh graders participated in the three-month study.
Mixed-skill homework yielded 74% correct, blocked-skill homework yielded 42% correct.
Sources
Linux Kernel Module Programming Guide now in ChapterPal collection
Linux Kernel Module Programming Guide by Peter Jay Salzman, Michael Burian, Ori Pomerantz, Bob Mottram, and Jim Huang (2026) is now in ChapterPal's collection of open textbooks. The book defines a Linux kernel module (LKM) as a piece of compiled code that can be dynamically loaded into or.
Key points
Book title and authors: Linux Kernel Module Programming Guide by Peter Jay Salzman, Michael Burian, Ori Pomerantz, Bob Mottram, and Jim Huang (2026)
Availability: now in ChapterPal's collection of open textbooks
Definition: A Linux kernel module (LKM) is a piece of compiled code that can be dynamically loaded into or
Sources
e3nn – Graph Neural Networks added to ChapterPal
e3nn – Graph Neural Networks added to ChapterPal notes, a new book titled Equivariant Graph Neural Networks with e3nn by Mohammad Mostafanejad (2026). e3nn is an open-source, modular Python framework for building E(3) equivariant neural networks, developed primarily for PyTorch (with JAX).
Key points
Book: Equivariant Graph Neural Networks with e3nn by Mohammad Mostafanejad (2026)
Framework: e3nn is open-source, modular Python, for E(3) equivariant neural networks, primarily PyTorch (with JAX)
