@Gabriel2KS

AI safety with @MATSprogram at @redwood_ai / On sabbatical from @NatLabRockies / @pomonacollege Computer Science ’23 / he/him

 Berkeley, California
gabrielks.com/
Joined October 2019
  • Tweets4
  • Following100
  • Followers48
1Media
In the first result of my @MATSprogram fellowship at @redwood_ai, I show that small LLMs can perform no-CoT two-hop reasoning on pairs of synthetic statements individually fine-tuned in, as long as the statements share a real entity "bridge." More at lesswrong.com/posts/CdWbCvKE…
1
7
85
3,039
Proud of this work I recently presented at the @icmlconf Mechanistic Interpretability Workshop with my @LASRlabs teammates!
Nikita Menon@nikxtacoJul 27
Model organisms (MOs) are frequently used as testbeds for mechanistic interpretability methods. We train 54 of them and find that their interpretability scores vary widely depending on how they were trained. Your interpretability benchmark result may be a lottery. 🧵
3
147
Hey @SunCountryAir, left my laptop in a black padded case on a plane from Belize to @mspairport yesterday, SY712. Already filed a lost item report. Any advice?
2
270
for(long t=0,n=0,y=1;;n-=y*y+(y=(int)(t*y+n>>7))*y>>62)System.out.println(n++*4d/t++); //☕😉
57
Load more