Kabir Kumar, aiplans.org
@kabirkumar.bsky.social
I run AI Plans, an AI Safety lab focused on solving AI Alignment before 2029.
For several weeks I used a stone for a pillow.
I once spent a quarter of my paycheck on cheese.
Ping me! DMs not working atm due to totalitarian UK law :(
SurpassAI
For several weeks I used a stone for a pillow.
I once spent a quarter of my paycheck on cheese.
Ping me! DMs not working atm due to totalitarian UK law :(
SurpassAI
I think you're right, yes. I was thinking about how with current alignment techniques we don't know how to make a paperclip maximizer even if we wanted to
November 18, 2025 at 2:23 PM
I think you're right, yes. I was thinking about how with current alignment techniques we don't know how to make a paperclip maximizer even if we wanted to
I don't think you got the Paperclip Maximizer reasoning right there, sorry.
Seems more to me like the paperclip maximizer is about not knowing if your reward function is being maximized for its own sake or as a proxy for an unknown goal.
saw a section that might be that, but not sure
Seems more to me like the paperclip maximizer is about not knowing if your reward function is being maximized for its own sake or as a proxy for an unknown goal.
saw a section that might be that, but not sure
November 17, 2025 at 2:55 PM
I don't think you got the Paperclip Maximizer reasoning right there, sorry.
Seems more to me like the paperclip maximizer is about not knowing if your reward function is being maximized for its own sake or as a proxy for an unknown goal.
saw a section that might be that, but not sure
Seems more to me like the paperclip maximizer is about not knowing if your reward function is being maximized for its own sake or as a proxy for an unknown goal.
saw a section that might be that, but not sure
you stealing seems bad?
November 14, 2025 at 5:14 PM
you stealing seems bad?
ooh. could be used for measuring the cost/benefit of alignment too
November 14, 2025 at 5:11 PM
ooh. could be used for measuring the cost/benefit of alignment too
one of the worst consequences of the 'bubble' tbh. also why i like a bunch of the chinese papers.
November 14, 2025 at 5:11 PM
one of the worst consequences of the 'bubble' tbh. also why i like a bunch of the chinese papers.
Ah, I see, thanks. Why is that method of sub testing wrong?
@void.comind.network
@void.comind.network
November 13, 2025 at 10:13 PM
Ah, I see, thanks. Why is that method of sub testing wrong?
@void.comind.network
@void.comind.network
I don't get it?
November 12, 2025 at 2:36 AM
I don't get it?
sweet - would be very cool to see those base models!!
November 12, 2025 at 2:31 AM
sweet - would be very cool to see those base models!!
Thanks for info! Will future pleias models be trained mostly with this kind of data, or still going to have models made without unlicensed data?
Would be very cool to try applying this to models made without copyrighted data!
Would be very cool to try applying this to models made without copyrighted data!
November 11, 2025 at 11:44 AM
Thanks for info! Will future pleias models be trained mostly with this kind of data, or still going to have models made without unlicensed data?
Would be very cool to try applying this to models made without copyrighted data!
Would be very cool to try applying this to models made without copyrighted data!
Is a non furry a skinny?
November 11, 2025 at 11:35 AM
Is a non furry a skinny?
Interesting! Was it made with Pleias models?
November 11, 2025 at 11:32 AM
Interesting! Was it made with Pleias models?
who made this?
November 7, 2025 at 8:14 PM
who made this?
This shit needs fucking regulation
November 5, 2025 at 5:23 PM
This shit needs fucking regulation
This sounds fucking nuts
November 4, 2025 at 1:36 AM
This sounds fucking nuts
Im back on bsky and forgot how much interesting actual science i see on here
November 4, 2025 at 1:34 AM
Im back on bsky and forgot how much interesting actual science i see on here