Flock Capital retweeted
you can now sign up for alpha and register your interest on the website! we will roll out access to a few select users as developement goes.
sign up now at reflet.sh, and feel free to join our discord server for updates
sign up now at reflet.sh, and feel free to join our discord server for updates
we're building a fully native disassembler that doesn't suck
0
3
37
2,252
Flock Capital retweeted
we are pleased to present RETROGRADE, a procedurally generated, behaviorally graded benchmark for reverse-engineering agents
there’s no fixed corpus to leak. every target is a program that didn’t exist until the generator drew its seed
results from Round 0 to follow 🧵
there’s no fixed corpus to leak. every target is a program that didn’t exist until the generator drew its seed
results from Round 0 to follow 🧵
5
7
80
7,665
Flock Capital retweeted
fun demo! this runs on the ontologizer work we’ve been doing.
sparse autoencoders can’t make this chart. they give you a bag of 16k latents per person. an ontologizer gives every attendee a discrete code word, which is why you get a typology instead of a feature dump.
sparse autoencoders can’t make this chart. they give you a bag of 16k latents per person. an ontologizer gives every attendee a discrete code word, which is why you get a typology instead of a feature dump.
At @WakaruLabs, we needed to decide who to prioritize talking to at EA Global. I used a truebase's residual stream activations and our library to make a color-coded ontologizer of conference attendees
5
3
32
2,858
Flock Capital retweeted
At @WakaruLabs, we needed to decide who to prioritize talking to at EA Global. I used a truebase's residual stream activations and our library to make a color-coded ontologizer of conference attendees
15
13
73
10,836
Flock Capital retweeted
we made sparse autoencoders compete with tiny softmax classifiers and the classifiers won on bits 🐐
a pure discrete code (no coefficients!) beats a top-k SAE at matched bits on SONAR, and an 800-bit hard code recovers 97.6% of gpt-2's loss
paper out now! (link in replies)
a pure discrete code (no coefficients!) beats a top-k SAE at matched bits on SONAR, and an 800-bit hard code recovers 97.6% of gpt-2's loss
paper out now! (link in replies)
2
3
45
7,389
Our updated research paper about Ontologize, a proposal for MOE-Based interpretability, is now live on our research repository.
Link in Replies
Link in Replies
1
1
3
1,919
the flock capital research department would like to announce that it has been doing interpretability
with Wakaru. paper soon. we also printed the plates ourselves
with Wakaru. paper soon. we also printed the plates ourselves
excited to announce a proposal from @flockcap and Wakaru: ontologize
mixture-of-experts interpretability where the features are the parameters.
instead of an SAE's flat bag of directions, you get stacks of contrastive classifiers you can read from weights and poke directly 🧵
1
1
3
940