#catch
5 открытий · Подборка открытий из открытого интернета, сначала новые.
-
Let's Try to Catch Misalignment Before a Model Learns to Hide It
An AI model that appears aligned could be either (1) actually aligned or (2) hiding its misalignment while being evaluated. How can we tell which? One idea: Look at the trajectory of eval results across the entire training process. My hypothesis is that misaligned behavior would emerge before the ability to hide misalignment, like how young children first misbehave in obvious ways before learning to be sneaky. As training progresses, we'd first see misalignment eval scores rise, then see them drop as the model learns to hide misalignment. Even if the model becomes very sophisticated - say, mod
LessWrong · EN
-
The first DNA crime test didn’t catch the killer. It freed an innocent man.
The events leading to the first forensic DNA case began in 1983. Lynda Mann was only 15 years old when, returning home from babysitting one evening in the Leicestershire village of Narborough, she was raped and murdered as she took a shortcut along a footpath known by villagers as the Black Pad. It was a path that would be unlikely to be known by anyone who wasn’t local, suggesting the killer was someone from the area. At the time, the biological side of forensic science was still in its infancy, though there were a handful of tests that could help narrow things down in terms of identifying th
Big Think · EN
-
Scripting a Zoho catch-all filter
I use Zoho as my email provider, and I recently learned that you can setup a catch-all forwarding address . This is pretty great, because that means I can receive any email sent to my domain! Go ahead, try it . I want to sort the emails I receive into their own folder, though, depending on the delivery address (for example, lol@miccah.io will go to /catch-all/lol ). It turns out, Zoho supports custom mail filters using a scripting DSL called deluge , so I wrote one!
miccah.io · EN