What's the Dyfference?

Open science infrastructure for long-lived safety assessments.

from dyff.client import Client

client = Client(api_key="XXXXXX")

dataset = client.datasets.create_arrow_dataset(
    "/my/data", account="XXX", name="mydata"
)

client.datasets.upload_arrow_dataset(dataset, "/my/data")

But Why?

Assessment Integrity

A machine learning system built to pass its safety assessments is not assessed for safety. Dyff selectively exposes safety assessment results so developers can't game the test.

Assessment Lifetime

Safety evaluations are typically discarded after their application. Dyff provides the capacity to surface and communicate safety without exposing the underlying data. Dyff assessments are long-lived.

Assessor Viability

Long-lived high integrity safety evaluations are not maintained by researchers. Dyff demonstrates a path for an economically sustainable evaluation ecosystem across the thousands of digital use cases.

Get the Dyff client

python3 -m pip install dyff