Sabrina Zaki

Message

Cognitive Science student, working with ML safety @ Apart Research

Automated Sandwiching & Quantifying Human-LLM Cooperation: ScaleOversight hackathon results

We ran a hackathon on scalable oversight with Gabriel Recchia as keynote speaker (watch the talk) and Ruiqi Zhong as co-judge. Here, we share the top projects and results. In summary: * We can automate the “sandwiching” paradigm from Cotra [1] by having a smaller model ask structured questions to...

Feb 23, 2023•8

Message

Cognitive Science student, working with ML safety @ Apart Research

13 karma

Member for 3 years

Sabrina Zaki — LessWrong

Sabrina Zaki

Message

Cognitive Science student, working with ML safety @ Apart Research

Sabrina Zaki

Automated Sandwiching & Quantifying Human-LLM Cooperation: ScaleOversight hackathon results

Feb 23, 2023•8

Message

Cognitive Science student, working with ML safety @ Apart Research

13 karma

Member for 3 years

Automated Sandwiching & Quantifying Human-LLM Cooperation: ScaleOversight hackathon results

Esben Kran

Esben Kran, fbarez, Sabrina Zaki, gabrielrecc, rz2383+ 0 more

Esben Kran, fbarez, Sabrina Zaki, gabrielrecc, rz2383

We ran a hackathon on scalable oversight with Gabriel Recchia as keynote speaker (watch the talk) and Ruiqi Zhong as co-judge. Here, we share the top projects and results. In summary:

We can automate the “sandwiching” paradigm from Cotra [1] by having a smaller model ask structured questions to elicit a true answer from a larger model and getting a response accuracy rate as output.
We can understand coordination abilities between humans and large language models quantitatively using asymmetric-information language games such as Codenames.
We can study scaling and prompt specificity phenomena in-depth using a simple framework. In this case, word reversal is investigated to evaluate the emergent abilities of language models.

Watch the project presentations on... (read 1771 more words →)

LESSWRONG
LW

LESSWRONG
LW

Sabrina Zaki

Sabrina Zaki

Sabrina Zaki

Automated Sandwiching & Quantifying Human-LLM Cooperation: ScaleOversight hackathon results

Sabrina Zaki

Sabrina Zaki

Sabrina Zaki

Automated Sandwiching & Quantifying Human-LLM Cooperation: ScaleOversight hackathon results