roobscoob (@roobscoob1) header
roobscoob (@roobscoob1) profile picture
Manhattan, NY
Joined October 2020
Photo by roobscoob (@roobscoob1)Photo by roobscoob (@roobscoob1)Photo by roobscoob (@roobscoob1)Photo by roobscoob (@roobscoob1)Photo by roobscoob (@roobscoob1)Photo by roobscoob (@roobscoob1)Photo by roobscoob (@roobscoob1)Photo by roobscoob (@roobscoob1)Photo by roobscoob (@roobscoob1)Photo by roobscoob (@roobscoob1)
Replying to @MiniUlisse
Oh man, this is something i'd love to attend if I lived in SF. Loved checking out integralaltrusim.com though!
1
49
Replying to @reckoner165
Why do I even need to start tweeting when I agree so much with all of yours
1
32
Happy Halloween @recursecenter ! Also cool lights @roobscoob1
3
10
368
roobscoob retweeted
True. They are so giddy to comply they are like this adorable cow.
“atheists” when you ask them to criticize any religion that isn’t christianity
4
1
38
1,299
Replying to @_fcjr
don't subtweet me like this on main
3
29
Replying to @_fcjr
vaugepost king 🧲 🧲 🧲 🧲 🧲
1
35
Replying to @_fcjr
my agent's j-space
2
23
Replying to @sam_kececi
The purpose of an apology is to make visible what was previously invisible (your care for someone, your desire to do right by them, etc.) - in this way, every step away from you (incl. autocorrect) serves to obfuscate what you want to make visible. the opposite of an apology.
1
19
The terms intelligence, deception, want are abstractions over predicted human behavior. Applying it to an LLM takes exceptional evidence that these abstractions still apply.
9
1. you understood me, so clearly it does 2. you have still not made it clear why this would change 3. yes. given certain context. i'm arguing we can effectively steer them.
1
13
"Why would future AI operate like current AI?", and not "Assuming future systems will simply scale today’s behavior" do not imply not "future systems will simply follow whatever goal a human gives them"
1
28
Replying to @matthewktromp
> Once they are smarter than us, they will conspire to take control of earth and we will all die. I really think this is nonobvious. It seems models at present follow the goal of the prompt they are given and it's not clear why this will change.
3
2
174
Replying to @chaotictransfem
I'm really interested in your architecture for this! I was planning on working on something similar. Was going to embed a bunch of media by feeding it to a VLM and slicing midway through the model's layers. Then using RLHF (me) finding vectors that correlate with my enjoyment
2
23
Replying to @_ueaj
idk deontology seems to be right
53
damn time for me to lock in
1
13
sent this to my anthropic friend and they framed it
6
154
Replying to @chaotictransfem
This sounds super interesting! I've been wanting to toy around with base models. A few friends of mine have been toying with this idea as well, but nothing has formed
1
83
Replying to @k_grajeda
Is this source-available? would love to toy with it!
1
154