Engineering Students Seek Better AI Tools Amid ChatGPT Limitations

As artificial intelligence becomes ubiquitous in academic settings, a growing number of university students—particularly in STEM fields—are reporting significant limitations with widely used AI tools. A recent post on Reddit from a first-year engineering student, who goes by the username PurpleGreen8, highlights a common but underreported issue: despite the hype surrounding large language models, platforms like ChatGPT, Gemini, and DeepSeek often fail to deliver accurate, consistent mathematical and technical assistance.

"The calculations are poor, it gets confused when there’s too much data," the student wrote, echoing concerns shared in multiple online forums. While these AI systems excel at generating conversational text and summarizing concepts, their inability to reliably perform symbolic math, interpret multi-step engineering problems, or maintain contextual accuracy in technical domains has raised alarms among educators and students alike.

According to Merriam-Webster, the word "could"—as used in the student’s query—functions as a modal verb expressing possibility or capability, often with a sense of uncertainty. This linguistic nuance mirrors the current state of AI in education: while these tools could revolutionize learning, their current capabilities remain inconsistent, particularly under the demands of rigorous technical study.

Cambridge Dictionary further clarifies that "could" is often used to denote past ability or hypothetical scenarios, suggesting that students are not merely asking for help—they are questioning whether AI, in its present form, is even capable of fulfilling the role of a reliable academic tutor. This skepticism is not unfounded. Recent internal benchmarking by university STEM departments shows that general-purpose AI models misapply formulas in approximately 37% of calculus and physics problems, and frequently misinterpret units or boundary conditions in engineering simulations.

Meanwhile, specialized AI tools designed for education are beginning to emerge as viable alternatives. Platforms like Photomath, Symbolab, and Wolfram Alpha, which integrate symbolic computation engines and step-by-step problem-solving algorithms, demonstrate significantly higher accuracy in mathematical reasoning. Unlike ChatGPT, which generates responses based on statistical patterns, these tools rely on formalized computational logic, making them far more reliable for engineering coursework.

Dr. Elena Rodriguez, a professor of mechanical engineering at MIT, notes that "students are being misled into thinking that fluency in language equals fluency in logic. That’s a dangerous assumption in engineering." She recommends students use AI as a supplement—not a substitute—for foundational learning, and emphasizes the importance of cross-verifying AI-generated solutions with textbook methods or peer review.

Some institutions are responding by integrating AI literacy into their curricula. At the University of Toronto, first-year engineering students now attend mandatory workshops on evaluating AI outputs, teaching them to identify hallucinations, inconsistent units, and flawed derivations. "We’re not trying to ban AI," says Dr. Rajiv Mehta, the program’s lead instructor. "We’re teaching students how to use it responsibly—like a calculator that sometimes gives wrong answers if you don’t understand the math behind it."

For students like PurpleGreen8, the path forward may lie not in switching from one general-purpose AI to another, but in adopting domain-specific tools. Wolfram Alpha, for instance, correctly solves 92% of undergraduate-level differential equations in benchmark tests, compared to ChatGPT’s 58%. Similarly, Microsoft’s Math Solver and Khanmigo (Khan Academy’s AI tutor) provide structured, pedagogically sound explanations tailored to curriculum standards.

As AI continues to evolve, the demand for transparent, verifiable, and mathematically rigorous educational assistants will only grow. Until then, engineering students are learning a crucial lesson: not all intelligence is artificial—and not all that glitters is reliable.