Why Students Should Turn Off ChatGPT

Reflections on AI in Education

I have been considering the role of artificial intelligence in educational settings, specifically the unrestricted use of tools such as ChatGPT, Gemini, or Cursor at the college, high school, or middle school levels. This discussion focuses not on assistive tools like Flint, which are controlled by educators, but on freely available independent AI systems.

As an individual whose college experience occurred entirely within the AI era, I believe these tools will represent a net negative for a substantial number of computer science students. This perspective may appear ironic, but I view the optimal application of AI as benefiting those who already possess knowledge of abstractions, engineering principles, and relevant domains.

The Expert Versus the Novice

A key distinction exists in the effective utilization of AI.

For the expert, such as a senior engineer, AI serves as an extension of sophisticated judgment. They specify desired outcomes, outline implementation strategies, and employ AI to produce boilerplate code or integrate elements within an established architecture.

In contrast, the novice, often an inexperienced college student, prompts AI to generate complete solutions from initial problem statements, delegating the entire problem-solving process.

The accessibility of this novice approach poses a significant issue. I estimate that at least 50 percent of computer science students, possibly more, employ AI to circumvent assignment requirements. This practice has become evident. It enables individuals to complete a four-year degree without acquiring coding proficiency, achieving a 4.0 GPA with minimal effort, thanks to advancements in model capabilities.

The Importance of Fundamentals

Critics may contend that such reliance is inconsequential and suggest increasing assignment complexity. I disagree. The purpose of higher education lies in mastering foundational concepts.

Consider the historical shift from manual assembly coding to compiler-assisted development in higher-level languages. While compilers abstract low-level operations, electrical engineering curricula retain instruction in assembly to ensure comprehension of underlying mechanisms, despite infrequent practical application.

Likewise, university assignments, though narrowly defined, demand personal implementation to build essential skills. AI excels at these constrained tasks, which align with introductory building blocks, yet outsourcing them undermines true understanding.

Real-World Limitations

To achieve genuine proficiency, students should disable ChatGPT and similar tools like Cursor until they grasp underlying architectures, system designs, and programming languages sufficiently.

Academic success through AI-assisted shortcuts proves unsustainable professionally. In environments such as legacy organizations with extensive, unstructured codebases, such methods falter. I have observed this repeatedly: students adept at rapid prototyping in controlled settings struggle profoundly when deprived of AI in regulated sectors or with unfamiliar legacy systems. Performance deteriorates markedly, rendering contributions ineffective.

The Risk of Dependency

I remain uncertain about viable applications for AI coding assistance at the undergraduate level, as it often negates educational objectives. Exceptional learners may navigate this differently, but the typical student requires unassisted practice.

A concerning trend emerges: individuals developing a “ChatGPT brained” mindset, wherein AI supplants critical thinking. This proves beneficial only for those with prior expertise. Undergraduates, by definition, seek to acquire such expertise through structured learning.

Optimists predict imminent AI capabilities for generating complex applications in single prompts. I question this timeline. Excessive reliance fosters long-term challenges.

My recommendation remains straightforward: remove ChatGPT and abstain from Cursor. Prioritize conceptual mastery, documentation review, and independent problem-solving from initial exposure.

Transcribed from a voice memo using Gemini 2.5 Pro.