Goodfire Discovers Bouba-Kiki Activation Space in LLMs

Goodfire revealed that LLMs like Llama and Gemma contain a "bouba-kiki" activation direction, demonstrating that the models can map phonetic qualities like sharpness and roundness even for meaningless words.

2026-07-28 ~ 2026-07-28 · 2 related posts

• LLM activations expose a bouba-kiki axis learned from nonsense words — burny_tech · 2026-07-28
• Goodfire says LLMs contain a Bouba-Kiki activation space in Llama and Gemma — voooooogel · 2026-07-28