February 1, 2023
More stuff on detecting AI written text vs human written text.
First is from OpenAI, who are probably geniuses. They create a problem and then show up with the solution. Both available for a fee. Like the old Indian snake charmers, first they let the cobras out into the village, then they come by offering to catch any cobras that just might happen to be loose in the village.
New AI classifier for indicating AI-written text
Weâre launching a classifier trained to distinguish between AI-written and human-written text.
And a paper on catching GPT text.
[2301.11305] DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature
The increasing fluency and widespread usage of large language models (LLMs) highlight the desirability of corresponding tools aiding detection of LLM-generated text. In this paper, we identify a property of the structure of an LLM's probability function that is useful for such detection. Specifically, we demonstrate that text sampled from an LLM tends to occupy negative curvature regions of the model's log probability function. Leveraging this observation, we then define a new curvature-based criterion for judging if a passage is generated from a given LLM. This approach, which we call DetectGPT, does not require training a separate classifier, collecting a dataset of real or generated passages, or explicitly watermarking generated text. It uses only log probabilities computed by the model of interest and random perturbations of the passage from another generic pre-trained language model (e.g., T5). We find DetectGPT is more discriminative than existing zero-shot methods for model sample detection, notably improving detection of fake news articles generated by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code, data, and other project information.
-
-
A truly delightful thread!
https://infosec.exchange/@SecurityWriter/109777576538835360We have one client which we manage an Azure tenant for. They require, and have specified, a zero-tolerance for device non-compliance.
In roughly two hours, 1647 devices are about to be locked out of access to organisation resources, wiped, and removed from Intune permanently.
4 meetings, 124 emails, and two phone calls a day for the last 14 days have warned them of this.
We’ve been *very* clear about what is about to happen for the last 13 months. Their internal management have *acknowledged* what is about to happen. But still, time marches on.
Death by middle-management.
-
-
One of the joys of the National Archaeological Museum of Naples is its epigraphic collection, which includes amazing examples of ordinary script painted on dry plaster. This wall starts at top with a rental ad by the enterprising Iulia Felix: "In the property of Iulia Felix, daughter of Spurius, there are for rent a stylish bath for people of rank, workshops with housing above, first-floor apartments, from the first before the Ides of August to the sixth before the (following) Ides of August, for five consecutive years". Below this dignified text are four different electoral graffiti. "Vote for Aulus Vettius Verus for aedile in charge of streets and sacred buildings, a man worthy of public administration and an honest man." "Of Metellus..." "Vote for Lucius Seius Secundus for aedile. Proculus and Cantus request it." "(Vote for) Lucius Caecilius Capella (for duumvir)."" loading="lazy">
-
-
How to modernize C arrays for greater memory safety: a case-study in refactoring the Linux kernel and a look to the future
Bounded Flexible Arrays in C — kees
How to modernize C arrays for greater memory safety: a case-study in refactoring the Linux kernel and a look to the future Kees Cook C i...
-
This JDN explores the fundamentals of ISR.
UK moves from traditional/linear intel to activity-based intel & problem-centric ISR.
