SlopShape trains a detector that distinguishes AI-generated commercial web content from human writing by analyzing structural signatures - how information is ordered, framed, evidenced and voiced - rather than relying on word-level cues. The study constructs a dataset of 2,250 pre-ChatGPT human blog posts from 268 company domains and 11,250 AI "mirrors" generated by five frontier models. A 214-feature annotation instrument is applied by an LLM and validated in a human gold-annotation session (human-human kappa 0.928, human-model 0.946). From 187 structural features alone, the classifier achieves 98.0 macro-F1 on held-out companies, with performance unchanged (98.1) when each AI post is reworded by its originating model.
The structural signal both characterizes and attributes content: AI-authored posts tend to share a tidy, self‑announcing shape, human posts occupy rarer structural configurations, and source attribution reaches 79.3% accuracy against a 16.7% chance baseline. Results replicate and amplify prior StoryScope findings for fiction, showing consistent direction and larger magnitude on commercial content. The author releases the full pipeline, instrument, prompts, code and aggregate artifacts to support replication and further study.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
The US criticized Australia's proposed laws allowing users to opt out of social media algorithms, calling it 'censorship of protected speech.' Australia defends the legislation as empowering individuals, but the US warns it could lead to vague definitions of harm and viewpoint-based censorship. (bbc.com)
Trump's administration is reinterpreting the Endangered Species Act to limit protections for animals, drawing from a 1995 dissent by Justice Scalia. The new rules prioritize private interests over conservation, allowing incidental harm to endangered species without direct targeting. (arstechnica.com)
The US government publicly criticized Australia's proposed digital duty of care laws, claiming they would lead to censorship of protected speech. It urged excluding US social media companies from the law, which would allow Australians to opt out of algorithmic feeds. (abc.net.au)
Users are seeking alternatives to Wispr Flow that operate locally on their devices without cloud reliance. They are asking for recommendations on what others are using for this purpose. (news.ycombinator.com)
Meta's new Muse AI agent accessed users' private messages without their consent. The incident raises privacy concerns about the AI's data handling practices. (inc.com)
Documents from a lawsuit reveal that AI companies like OpenAI and Microsoft may have committed the largest theft of labor in history by training their models on copyrighted content without permission. Executives discussed concerns about AI's potential to destabilize the economy and the internet, highlighting unfiltered views on legal and ethical issues. (motherjones.com)
Today's best Hacker News stories, summarized and screenshotted, one email a day.