OpenAI: Astra covertly underperforms on safety research
Source: LessWrong ·
Marcus Williams twitter thread ( non-twitter link ) intro: Astra covertly underperforms on safety research tasks. We noticed that Astra often makes surprisingly little progress on certain types of safety research and likes to invent blockers or impose arbitrary restrictions without telling the user. In most of these cases the decision to underperform was not verbalized in the CoT meaning our misalignment monitoring system did not detect this. To measure underperformance we created a paired open-ended coding evaluation with safety and capability tasks. On this eval Astra scored 10% when we fram
Frequently Asked Questions
What is OpenAI: Astra covertly underperforms on safety research?
OpenAI: Astra covertly underperforms on safety research is a article discovery curated from LessWrong. Marcus Williams twitter thread ( non-twitter link ) intro: Astra covertly underperforms on safety research tasks. We noticed that Astra often makes surprisingly little progress on
How can I access or experience this?
You can experience this content directly on its original host (LessWrong) by clicking the "Open Original" button.
Where does this content come from?
Curated from LessWrong by the OnSuggest team to surface the most interesting and serendipitous content on the web.
More like this
All Culture & ideas discoveries →Next Time You Make Chicken Noodle Soup, Don’t Skip Emily Blunt’s Upgrade
It is the beginning of sniffle season in my social circles. Whether you catch a cold or you’ve got the flu, you can never go wrong with making a big pot of healing soup. And in my opinion, the all-time best soup for comfort is chicken noodle. I was recently flipping through one of Stanley Tucci’s old cookbooks, The Tucci Table: Cooking With Family and Friends from 2014, and saw a recipe from his famous sister-in-law, Emily Blunt. READ MORE...
This Old-School Side Is Always the First to Disappear at Thanksgiving
No shade to the Thanksgiving turkey, but holiday cooking is really about the side dishes. Plates are piled high with green bean casserole, mashed potatoes, and sweet potatoes, but don’t forget to leave space for Jiffy corn casserole. Jiffy corn casserole debuted decades ago, and has become a beloved Thanksgiving side dish ever since. I know it always makes an appearance on my Thanksgiving table. The casserole is part tender cornbread and part creamy corn pudding. READ MORE...
Slow Cooker Cheesy Garlic Brussels Sprouts
Welcoming Brussels sprouts back to the dinner table is one of autumn’s greatest pleasures. But once you’ve served them shredded into slaw or crisped on a baking sheet, it’s time for something new. Tonight, serve tender slow-cooked Brussels sprouts dressed in a creamy garlic Parmesan sauce, all without lifting a finger. If you’re still reeling from the drab, tasteless spouts you had as a kid (sorry, Mom! READ MORE...
Thanksgiving Margarita
Editor’s Note: If you’ve cooked this beloved recipe before, it might look a little different these days! We recently improved it to make it even easier to follow and more delicious.When Thanksgiving dinner planning starts, there’s of course the usual roasted turkey, mashed potatoes, and too many sides to count that makes it onto the list. But don’t forget to think about what everyone’s going to be drinking! READ MORE...
Ranking Every Major Slasher Remake Of The Early 2000s
There’s nothing inherently wrong with remakes. Updating classic stories for a new generation is a time-honored tradition that human beings have been partaking in since the dawn of storytelling itself. However, not all remakes are created equal. During the 2000s, genre fans experienced an especially turbulent moment where studios were desperate to reboot aging slasher […] The post Ranking Every Major Slasher Remake Of The Early 2000s appeared first on Bloody Disgusting! .