NEW: White House and Anthropic are working to create a formal technical assessment framework that can quantify the severity of the jailbreak in question and create a standardized methodology for evaluating similar incidents in the future.
It’s the clearest sign yet that talks are moving forward and it reflects an understanding that no AI model can be completely immune to hacking.
Aim is to developing a common set of benchmarks that could be used to assess future jailbreaks, including the extent to which safeguards were bypassed, the capabilities exposed, and the practical consequences of the breach.
w/ @cheyennehaslett
https://t.co/d5nJMk7bzh
Researchers analyzed 14 million academic papers published between 2010 and 2024. They tracked every word. They found that ChatGPT is rewriting the English language.
Not metaphorically. Literally.
After ChatGPT launched in November 2022, certain words that had been stable in academic writing for over a decade suddenly exploded in frequency. The researchers at the University of Tübingen and Northwestern University mapped every excess word and categorized them.
The words are ones you already recognize.
"Delve." "Intricate." "Meticulous." "Commendable." "Underscore." "Pivotal." "Nuanced." "Landscape." "Comprehensive." "Multifaceted." "Showcasing." "Groundbreaking." "Innovative." "Invaluable."
329 excess style words appeared in early 2024 that were not there before. The spike is unprecedented in the history of the dataset.
Here is what makes this different from every other vocabulary shift ever recorded. During COVID, excess words also appeared. Up to 188 of them in 2021. But those were content words. "Respiratory." "Remdesivir." "Ventilator." Words that described a new reality.
After ChatGPT, the excess words are not content words. They are style words. Not what people write about. How people write. The subject matter did not change. The voice did.
The researchers estimate that at least 10% of all academic papers published in 2024 were processed with ChatGPT. Not written entirely by AI. Processed. Edited. Polished. Run through the model and published with its fingerprints still on the page.
You have seen these words everywhere. In emails. In LinkedIn posts. In articles. In cover letters. In reports your colleagues sent you. You could not explain why everything started sounding the same. Now you can. The entire internet passed through the same model. And the model left the same fingerprints on everything it touched.
The researchers proved something else. The contamination is not slowing down. The number of excess words grew from 188 during COVID to 329 after ChatGPT. The curve is still climbing.
ChatGPT did not just change what we can do with language. It changed the language itself. One model. One voice. Fourteen million papers. And a vocabulary shift larger than a global pandemic.
JUST IN: The White House is reportedly demanding Anthropic make Fable 5 impossible to jailbreak before rereleasing it, a condition cybersecurity experts say cannot be met by any AI company.