Josh Fruhlinger
Contributing Writer

Learning how to measure genAI’s impact

analysis
Apr 18, 20252 mins
Business IntelligenceGenerative AIMachine Learning

Measuring the performance boost from a new CPU is pretty straightforward, but how to measure the performance of an AI model? Developers, tech, and business leaders are being asked to quantify generative AI’s return on investment. Our top stories this month have answers.

A graph on a blackboard shows an upward growth trajectory with the words 'Generative AI'
Credit: Linaimages / Shutterstock

Top picks for generative AI readers on InfoWorld

AI will require more software developers, not fewer
Data show AI assistants make it significantly faster and easier to develop software, spurring companies to tackle more software projectsβ€”and (surprise!) hire more developers.

Five questions to answer before you deploy an LLM
Before you jump on the AI bandwagon, you’ll want to answer a few key questions. Maybe a good place to start is, β€œAm I about to get sued?”

Measuring success in dataops, data governance, and data security
How should data and AI leaders measure the effectiveness of their dataops, data governance, and data security practices? We asked the experts.

Let’s get real about AI model performance
AI vendors tout rapid advances with each iteration of their models, but they’re often hard to quantify beyond just β€œvibes.” Enter the Vector Institute for Artificial Intelligence, a new project to establish independent benchmarks for AI.

More good reads and generative AI updates elsewhere

When it comes to language models, smaller might be better
LLMs are touted as the route to a human-level (or superhuman) artificial general intelligence. But when it comes to practical applications, small language models may pack more punch.

When will AI be smarter than people? That’s the wrong question
AGI, or artificial general intelligence, is the buzzword du jour in AI-boosting circles. But this idea mostly ignores the fact that human intelligenceβ€”the benchmark AGI is measured againstβ€”isn’t really β€œgeneral.”

Is ChatGPT addiction about to be a problem?
A joint study conducted by OpenAI and MIT Media Lab found that a subset of very regular ChatGPT users may develop a parasocial relationship with AI. And these users really don’t appreciate changes in model behavior.

Josh Fruhlinger

Josh Fruhlinger is a writer and editor who has been covering technology since the first dot-com boom. His interests include cybersecurity, programming tools and techniques, internet and open source culture, and what causes tech projects to fail. He won a 2025 AZBEE Award for a feature article on refactoring AI code and his coverage of generative AI earned him a Jesse H. Neal Award in 2024. In 2015 he published The Enthusiast, a novel about what happens when online fan communities collide with corporate marketing schemes. He lives in Los Angeles.

More from this author