Student perceptions of AI coding assistants in learning

Scope and rigor of the study

• Several commenters note the study’s very small sample (N=20) and see its findings as unsurprising: AI helps early confidence and implementation but leaves gaps when assistance is removed.
• Some argue the qualitative insights (how students actually use tools) are more valuable than the quantitative claims; others want much larger, more rigorous replications.

Learning, memorization, and syntax

• Debate over whether schools overvalue memorizing syntax vs deeper concepts, abstractions, and readability.
• Some contend you must first master basics to build higher-level skills; others stress that “learning” means generalization, not mere regurgitation.
• There’s concern that AI can create an illusion of understanding when students have not “earned” the knowledge through practice.

AI coding assistants vs calculators and other tools

• Repeated analogies to calculators, typewriters, Google, and high-level languages.
• Key distinction drawn: calculators and compilers are deterministic and logically sound; LLMs are probabilistic, can hallucinate, and outputs are hard to debug.
• Others counter that tools can still be transformative and widely adopted even if they require careful use and produce errors when misused.

Impact on assignments and curricula

• Some argue the particular OOP assignment in the paper is contrived, designed to force in inheritance rather than teach real-world design; in such artificial tasks, AI naturally looks less helpful.
• This is framed as a critique of curriculum design more than of AI’s learning value.

Cheating, grading, and credential erosion

• A long subthread describes how LLMs have “broken the curve”: cheating is easy, online/homework scores are inflated, and diligent students struggle to compete.
• Professors sometimes acknowledge suspected cheaters yet don’t adjust curves or enforce rules, prompting frustration.
• Others note that high exam scores from students in the back row are not always cheating; some simply learn outside lecture.
• Several predict universities and employers will devalue GPAs and rely more on direct assessments and longer, in-house evaluations.

Future of programming and “AI-native” skills

• Some predict that, as AI improves, learning to code “by hand” will become niche, akin to doing integrals manually.
• Critics argue that without a deep mental model (the “tower of knowledge”), students will be unable to handle hard problems or understand/verify AI-generated code.
• There’s speculation about a new divide between skilled AI users who use tools to think better and unskilled users who outsource thinking, with major implications for education and hiring.