LESSWRONG
LW

845
Carlos Ramón Guevara
5030
Message
Dialogue
Subscribe

Posts

Sorted by New

Wikitag Contributions

Comments

Sorted by
Newest
No posts to display.
No wikitag contributions to display.
Adam Optimizer Causes Privileged Basis in Transformer LM Residual Stream
Carlos Ramón Guevara1y10

Still elementwise, so yeah

Reply
Bing Chat is blatantly, aggressively misaligned
Carlos Ramón Guevara3y30

Apparently a "Sydney" model existed at least as early as 17 Dec 2021.

Reply
SolidGoldMagikarp (plus, prompt generation)
Carlos Ramón Guevara3y40

Why do you think that GPT-3 has untied embeddings?

Reply