Readers rated AI-generated poems as more human than real ones
A Scientific Reports study found non-expert readers misjudged AI poems as human more often than they misjudged real poems as AI.
3 min · source-led record↗
Explore / Read the evidence
What did the researchers actually measure?
Separate a study’s result from the bigger claim made about it.
No matching record. Try fewer words or reset the filters.
One writing assistant nudged many writers toward similar ideas
A 33-participant study found ChatGPT users produced less distinct ideas than users of a non-AI tool, and AI-written book reviews cluster narrowly too.
3 min · source-led record↗
Most Americans expect AI to make creative thinking worse
A representative Pew survey found 53 percent of U.S. adults expect AI to worsen creative thinking, versus 16 percent who expect improvement.
3 min · source-led record↗
An opinionated writing assistant shifted what people believed
A CHI 2023 experiment found a one-sided writing assistant shifted both what 1,506 participants wrote and what they later said they believed.
3 min · source-led record↗
ChatGPT access cut writing time and lifted quality in a trial
A Science-published experiment found ChatGPT access cut professional writing time by 40 percent and raised rated quality by 18 percent among 453 workers.
3 min · source-led record↗
An annual benchmark scored machine translation of web fiction
The WMT24 findings paper reports 10 systems judged mainly by humans on Chinese-English web fiction, a narrower scope than the task page had announced.
3 min · source-led record↗
AI-text detectors flagged most non-native writers' essays
A Stanford study found seven GPT detectors misclassified 61 percent of non-native English essays as AI-written, versus near-perfect accuracy on native essays.
3 min · source-led record↗
Word-frequency analysis found AI edits in conference reviews
A Stanford method estimates 6.5-16.9 percent of reviews at four AI conferences were AI-modified; a related study found up to 17.5 percent in CS papers overall.
3 min · source-led record↗
A readership survey found researchers split on chatbot use
Nature's survey of about 1,600 self-selected researchers found many use generative AI for manuscripts; its methodology says the sample is not representative.
3 min · source-led record↗
A classifier caught AI-written book reviews with high accuracy
A transfer-learning classifier reached 96.86 percent accuracy distinguishing real Amazon book reviews from reviews generated by the Vicuna language model.
3 min · source-led record↗