Winnie Yang

Jan 6, 2025 | nymag.com | Kiki Aranita |John deBary |Winnie Yang

As a former restaurant chef, current recipe developer, and onetime dishwasher (I did all the dishes for the food truck I operated for seven years), I am constantly seeking the perfect sponge. At this point in my culinary career, I have tested at least 20 different brands of sponges. I generally keep three sponges of different levels of abrasion by my kitchen sink, rotating between them depending on the task: If I’m cleaning glassware, dishes, pots and pans, my oven, or wiping down the counters.

Dec 5, 2024 | nymag.com | Hilary Reid |Winnie Yang

Illustration: Clara Kirkpatrick Oft associated with designer pears, artisanal cheeses, and high-end chocolates, holiday gift baskets can be a handy, one-size-fits-all present to check multiple people off your list. But what if holiday gift baskets were tailored exactly to your recipient’s taste and interests? That was the question we posed to the Strategist team as we set out to create 76 hyperspecific gift baskets.

Dec 3, 2024 | nymag.com | Winnie Yang

Photo-Illustration: The Strategist; Photos: Retailers Obviously, Strategist staffers are excellent shoppers (it’s why you’re here visiting the site right now). So once a month, we highlight all the things we’ve been buying for ourselves. You’ll notice that it’s both a blessing and a curse to be a Strategist editor; we’re picky, but sometimes we just have to get on with it. (It’s online-shopping expertise meets IRL needs.) Below, what we bought in November.

Aug 27, 2024 | lesswrong.com | Winnie Yang

Content Warning: This blog post contains examples of harmful language generated by LLM. As models grow increasingly sophisticated, they will surpass human expertise. It is a fundamentally difficult challenge to make sure that those models are robustly alignmed (cite scalable oversight). For example, we might hope to reliably know whether a model is being deceptive in order to achieve an instrumental goal.

Aug 22, 2024 | lesswrong.com | Winnie Yang |Jojo Yang |Lennart Buerger

As models grow increasingly sophisticated, they will surpass human expertise. It is a fundamentally difficult challenge to make sure that those models are robustly aligned (Bowman et al., 2022). For example, we might hope to reliably know whether a model is being deceptive in order to achieve an instrumental goal. Importantly, deceptive alignment and robust alignment are behaviorally indistinguishable (Hubinger et al., 2024).