Task vectors are akin to punchcards: you feed them to your LLM and it implements specific tasks, without in-context demonstrations. Liu's new paper examines at what scale, where in the network and when during training do they emerge, and how to encourage their emergence. arxiv.org/pdf/2501.09240 - ThreadSky

dimitrisp.bsky.social • 40 days ago

Task vectors are akin to punchcards: you feed them to your LLM and it implements specific tasks, without in-context demonstrations. Liu's new paper examines at what scale, where in the network and when during training do they emerge, and how to encourage their emergence.

https://arxiv.org/pdf/2501.09240

1 / 4

Comments

Posting Rules

Comments

Posting Rules

Reply