cfgt retweeted
We all know ChatGPT can be jailbroken. But what makes these attacks possible?
In a new paper (w/ @nhaghtal @JacobSteinhardt), we delve into how safety training fails, identifying two key failure modes that enable jailbreaks: competing objectives and mismatched generalization
1/
cfgt retweeted
Times ChatGPT hallucinated functions existing in the rust ffmpeg bindings: 152
Times it got the code right: 0
For strange shit, we just aren’t there yet. But if you are doing something non-complex sure you might be able to tab complete your way there ¯\_(ツ)_/¯
cfgt retweeted
Nothing like seeing people complain about something you have in github open for PR's without actually making any effort to fix things.
cfgt retweeted
AWS keynote Sydney. “Data beats intuition”. Really? If data defines everything at Amazon why is the search ranking so crap. #awssummitsydney
I believe I've finally pinpointed the source of my woes in iOS 16 - it looks like if I disconnect from my network while connected to an SMB share, there is a non-zero chance that the entire phone will just hang at a random point doing a network task later on.
cfgt retweeted
Reinforcement Learning by Human Feedback is just parenting for a supernaturally precocious child.
cfgt retweeted
page 37 of the GPT-4 paper: cdn.openai.com/papers/gpt-4.…
cfgt retweeted
This is absurd. To check if I'm a robot, they're using AI generated images and none of them look remotely like basketballs.
Yet *I'm* the one who's wrong.
cfgt retweeted
Cones: Concept Neurons in Diffusion Models for Customized Generation
abs: arxiv.org/abs/2303.05125
cfgt retweeted
A huge credit to the make-up artist. She previously worked on Mad Max where she was usually applying black eyes rather than hiding them.
cfgt retweeted
I've seen a lot of people asking "why does everyone think Twitter is doomed?"
As an SRE and sysadmin with 10+ years of industry experience, I wanted to write up a few scenarios that are real threats to the integrity of the bird site over the coming weeks.
Apple Music has an interesting edge case - there's a song with both Japanese and English versions, are both on the same single/album with the same name and artist, but you can only add one of the versions to your library. The physical version has the names in different languages.
























