mwatkins

The Tor­toise and the Lan­guage Model (A Fable After Hofs­tadter)

Ex­plor­ing the pe­ter­todd /​ Leilan du­al­ity in GPT-2 and GPT-J

Ex­plor­ing SAE fea­tures in LLMs with defi­ni­tion trees and to­ken lists

Nav­i­gat­ing LLM em­bed­ding spaces us­ing archetype-based directions

What’s up with all the non-Mor­mons? Weirdly spe­cific uni­ver­sal­ities across LLMs

Phal­lo­cen­tric­ity in GPT-J’s bizarre strat­ified ontology

Map­ping the se­man­tic void III: Ex­plor­ing neighbourhoods

Map­ping the se­man­tic void II: Above, be­low and be­tween to­ken em­bed­dings

′ pe­ter­todd’’s last stand: The fi­nal days of open GPT-3 research

Map­ping the se­man­tic void: Strange go­ings-on in GPT em­bed­ding spaces

Lin­ear en­cod­ing of char­ac­ter-level in­for­ma­tion in GPT-J to­ken embeddings

The “spel­ling mir­a­cle”: GPT-3 spel­ling abil­ities and glitch to­kens revisited

The ‘ pe­ter­todd’ phenomenon

AI Safety Info Distil­la­tion Fellowship

SolidGoldMag­ikarp III: Glitch to­ken archaeology

SolidGoldMag­ikarp II: tech­ni­cal de­tails and more re­cent findings

SolidGoldMag­ikarp (plus, prompt gen­er­a­tion)

All AGI Safety ques­tions wel­come (es­pe­cially ba­sic ones) [~monthly thread]