"The term "YOLO" was cool once, made so in 2011 by Drake in his hit song, "The Motto." Then it slipped into the domain of the unhip and out-of-touch. Well, it's now back. This time, it's being used by the AI vanguard to describe the state of the industry, which is a tad worrying to those concerned about AI's far-reaching implications for the world."
"In a "yolo run," he said, a researcher "directly implements an ambitious new model without extensively de-risking individual components. The researcher doing the yolo run relies primarily on intuition to set hyperparameter values, decide what parts of the model matter, and anticipate potential problems. These choices are non-obvious to everyone else on the team." This approach contrasts with the traditional research approach to carefully change one thing at a time, he added."
YOLO culture has reappeared within the AI industry, characterizing large investments and fast-paced model development. Some leaders warn that certain players are "YOLO-ing," pulling the risk dial too far and favoring reckless over rigorous methods. Other researchers praise "yolo runs" as intuitive, flow-state experiments that implement ambitious models without extensively de-risking components, relying on intuition for hyperparameters and design choices. That approach stands in contrast to the traditional practice of changing one variable at a time. The resurgence of YOLO-like attitudes raises concerns about safety, governance, and the broader implications of accelerating AI without sufficient safeguards.
Read at Business Insider
Unable to calculate read time
Collection
[
|
...
]