Logan Riggs

Sparse Au­toen­coders Find Highly In­ter­pretable Direc­tions in Lan­guage Models

Con­vinc­ing All Ca­pa­bil­ity Researchers

Trauma, Med­i­ta­tion, and a Cool Scar

Make a Movie Show­ing Align­ment Failures

Really Strong Fea­tures Found in Resi­d­ual Stream

Want­ing to Suc­ceed on Every Met­ric Presented

Us­ing GPT-N to Solve In­ter­pretabil­ity of Neu­ral Net­works: A Re­search Agenda

Find­ing Sparse Lin­ear Con­nec­tions be­tween Fea­tures in LLMs

(ten­ta­tively) Found 600+ Monose­man­tic Fea­tures in a Small LM Us­ing Sparse Autoencoders

Con­vinc­ing Peo­ple of Align­ment with Street Epistemology

To­day a Tragedy

Sav­ing the world in 80 days: Epilogue

A sur­vey of tool use and work­flows in al­ign­ment research

• Logan Riggs 25 Oct 2021 18:36 UTC

  44 points

  on: Self-In­tegrity and the Drown­ing Child

  At a pond, my niece was in a child floaty, reached too far and flipped over into the water. I slammed my half-eaten sandwich on my brother’s chest, hoping he would grab it and ran into the water and saved her.

  She was fine and I got to finish my sandwich.

Map­ping Out Alignment

Was Re­leas­ing Claude-3 Net-Nega­tive?

Solve Cor­rigi­bil­ity Week

Kiss­ing Scars

Sparse Au­toen­coders: Fu­ture Work

Lan­guage Model Tools for Align­ment Research