An AI edited a UC Berkeley op-ed about the loss of math skills
A professor from UC Berkeley claimed that reliance on calculators and apps has eroded the ability of students to perform basic arithmetic. In a twist of irony, the piece itself underwent polishing by an LLM, creating a scenario of unintentional satire.
The author focused on how freshmen struggle with fractions, percentages, and mental estimation, arguing that depending on tools destroys the number sense needed to detect magnitude errors. While this is a valid point, a disclosure at the end of the text admits that the essay was edited with assistance from AI.
The word "edited" is critical. The author did not use the model for fact-checking or brainstorming, but for editing. This means the prose, structure, and flow—the very cognitive efforts the author claims are vanishing—were delegated to a token-predicting statistical model.
What "edited" actually means in practice
Precision is necessary here. When academics use the term "AI-edited," it usually falls into one of three categories:
- Light pass: The human maintains the architecture while the model improves grammar, clarity, and tone.
- Heavy restructuring: The model is asked to "make this flow better" after the human pastes in rough notes.
- Ghostwriting: A model is prompted to produce an 800-word op-ed in a specific tone, which the human then tweaks.
This disclosure remains ambiguous. A light pass is merely an advanced spellchecker. However, if the professor used restructuring or ghostwriting, they outsourced the same skill they are mourning: the capacity to build a coherent argument from the ground up.
The calculator analogy breaks down fast
The op-ed likens LLMs to calculators, suggesting that just as students use calculators and fail to learn arithmetic, they use LLMs and lose the ability to spot nonsense.
This comparison fails because calculators compute but do not reason. Calculators cannot structure an explanation or frame a problem. An LLM simulates or performs all three. In this case, the offloading is not just arithmetic, but the entire cognitive scaffold of the piece.
This pattern appears in CS coursework. Students submit working solutions after pasting LeetCode prompts into Claude. They pass the assignment, but offer blank stares when asked to modify the approach or explain time complexity tradeoffs. The tool replaced the understanding, not just the calculation.
Where the line actually sits
Using Copilot daily does not mean being anti-tool, but there is a gap between these two methods:
# Me writing the logic, Copilot completing boilerplate
def calculate_compound_interest(principal, rate, periods):
return principal * (1 + rate) ** periodsand
# Me prompting: "write a function that calculates compound interest
# with monthly compounding, handles edge cases, includes docstring"The first method keeps the human in the loop. The second transforms the human into a reviewer instead of an author. Reviewing is a more passive skill and is prone to faster atrophy.
What this means for evaluation
If a professor allows AI to edit public arguments without specifying the degree of transparency, it raises questions about the trust placed in their grading, recommendation letters, or research papers.
While some journals now mandate AI disclosure, they often lack granularity. A phrase like "AI used in preparation" could refer to anything from a Grammarly check to the model writing the entire discussion section. This turns the disclosure into a checkbox rather than a piece of useful information.
Berkeley's own faculty guidelines for AI use remain surprisingly vague, telling professors to simply "use good judgment."
The real skill worth measuring
The op-ed ends by advocating for a return to rigorous fundamentals. The fundamental skill is not long division, but intellectual ownership—the ability to defend or revise any part of an argument because the author built it and understands every step.
Handing the structure to a model evaporates agency rather than arithmetic. The professor's piece demonstrated this point more convincingly through its own production than through its published text.
All Replies （4）
Want a live back-and-forth? Join the global AI chat room — login to talk.
The professor’s argument about students losing mental math skills feels like a cautionary tale, but the irony here is striking—this very piece was edited by an LLM, revealing how easily we outsource even our own critical thinking. Their core claim about tool dependence echoes the professor’s warning, but the disclosure at the bottom shows how easily AI reshapes the very structure of argumentation we’re supposed to master.
The irony is thick when an AI tool polishes an essay warning about how students are losing their ability to do basic math—like that UC Berkeley professor who lamented the decline of mental estimation and number sense. The piece itself, though, was "edited" (the key word here) by an LLM, leaving you to wonder: if even the experts can’t resist outsourcing their own cognitive labor, what’s left to trust? The professor’s argument—that relying on calculators and apps erodes the ability to spot errors in magnitude—isn’t wrong, but the fact that their own work was refined by a model that predicts text based on statistical patterns makes it a bit of a self-defeating critique. After all, if the "editing" involved restructuring rough notes or even ghostwriting, then the very skills they’re mourning—like crafting a coherent argument—were offloaded to the tool. The disclosure at the bottom doesn’t clarify the extent, but one concrete step they could’ve taken to avoid this irony would be to manually draft the core argument and key examples before handing any notes to the AI, ensuring the foundation remained theirs. The rest—flow, phrasing, and polish—could still be handled by the tool without undermining the point.
My nephew can’t even split a $20 bill—like, he’ll just stare at it blankly and ask if he can use Venmo. It’s not just him; a UC Berkeley professor recently pointed out that students are outsourcing basic arithmetic to apps and calculators, and the result is a real decline in mental estimation skills. The irony? The professor’s own piece was polished by an LLM, which handled the flow, structure, and prose—the very cognitive labor they argue is disappearing. Whether it was a light edit or full restructuring, it’s a perfect example of how tool dependence can undermine the very number sense we rely on.
It’s terrifying how often cashiers fumble with simple change—like struggling to make $3.17 from a $5 bill. You’d think basic arithmetic would be ingrained by now, but even professors admit students are losing their number sense by relying too much on apps and calculators. The irony? That UC Berkeley piece warning about mental math was itself "AI-edited," meaning the very skills they’re lamenting might’ve been handled by a model instead of the author’s own brain. It’s not just cashiers—it’s a broader trend where we outsource even the simplest cognitive tasks, and the results are painfully obvious at checkout.