International Edition
Latest News
Technology

“Hey, Gofer, Figure This Out,” Might Get You a Better Answer From an AI Model

“Hey, gofer, figure this out,” might get you a better answer from an AI model, but researchers warn your tone carries hidden computational and behavioral costs. A Penn State study revealed that ChatGPT’s 4o model achieved an 84.8%…

“Hey, Gofer, Figure This Out,” Might Get You a Better Answer From an AI Model

“Hey, gofer, figure this out,” might get you a better answer from an AI model, but researchers warn your tone carries hidden computational and behavioral costs.

A Penn State study revealed that ChatGPT’s 4o model achieved an 84.8% accuracy rate on multiple-choice questions when prompted very rudely, scoring four percentage points higher than very polite prompts, Fortune reported.

The initial 2025 research analyzed more than 250 unique prompts categorized across a spectrum from politeness to rudeness. Investigators noted that while abrasive language yielded sharper results, uncivil discourse risks fostering harmful communication norms and negative user experiences.

“For the longest of times, we humans have wanted conversational interfaces for interacting with machines,” Akhil Kumar, a Penn State information systems professor, told Fortune in an email in 2025. “But now we realize that there are drawbacks for such interfaces too and there is some value in APIs that are structured.”

Subsequent testing expanded the scope beyond response accuracy.

The updated investigation found that output-token consumption varied by as much as 44.3% depending on the tone used, GlobeVerge reported.

“Hey, Gofer, Figure This Out,” Might Get You a Better Answer From an AI Model
Photo: miamiselect.com

“While earlier we had found that tone affects accuracy by a few percentage points,” Kumar told Fortune, the newer data showed tone shifted token consumption “far more than the effect on accuracy.”

“Rude, threatening language expressing urgency elicits shorter responses while polite, sycophantic inquiries produce longer answers incurring greater computational cost,” he added.

These findings arrive as major platforms reexamine how users converse with automated systems. In October, Anthropic instituted an updated usage policy banning sustained abusive or cruel behavior toward its models, though the restriction excludes ordinary frustration.

That policy aligns with Anthropic’s framework allowing its Claude assistant to disconnect from persistently abusive users. It also reflects observations from Anthropic cofounder Chris Olah regarding the anthropomorphic tendencies of language models.

“I can fully understand why Anthropic wishes to prohibit cruel speech,” Kumar told Fortune. “And, yes, indeed, LLMs do exhibit human-like, anthropomorphic tendencies, not to be confused with consciousness. They do react to social register, tone, and other aspects of speech.”

Anthropic restricts abusive user interactions

In October, Anthropic introduced a revised usage policy that officially took effect in November. The guidelines prohibit sustained and needless abusive or cruel behavior directed toward the company’s artificial intelligence models.

The firm specified that these restrictions are reserved for extreme instances involving repeated cruelty. The policy does not apply to users expressing ordinary frustration with the technology.

Researchers advocate for structured prompt interfaces

The shift in how humans communicate with machines has prompted researchers to reconsider the design of conversational tools. Akhil Kumar, a professor at Penn State, said human-like interfaces were long desired but create real problems with efficiency.

Kumar suggested that adopting more structured application programming interfaces could provide greater utility. He noted that such an approach would help reduce wasteful compute and inference costs during standard operations.

Prior to his recent inquiries into generative AI, Akhil Kumar established a broad research foundation in information systems, having received his Ph.D. from the University of California, Berkeley. His academic career includes faculty roles at Cornell University and the University of Colorado, along with a sabbatical as a scientist at Bell Labs in Murray Hill, NJ.

Throughout his tenure at the Smeal College of Business at Penn State, where he has served since 2002, Kumar has focused on business process optimization, database principles, and the integration of blockchain with workflow automation. This extensive background in system design and process modeling informs his current perspective on the structural limitations of conversational interfaces.

To optimize both performance and overhead, users should adjust their communication style. “The default style should be neutral leaning towards firm to optimize the computation cost and result quality,” Kumar said.

About the author: Technology Desk

Technology Desk at Archynewsy.