Post

I don’t think there’s a single serious critic who says AI’s not useful? The concerns I see are about content rights, worker impacts, environmental effects, epistemological threats, or media manipulation. But that’s all predicated on the idea that it *does* work, often too well for people to manage…

Kevin Roose@kevinroose.com

The NYT Magazine asked me and @caseynewton.bsky.social to open their AI issue by talking about how we use AI, why it's not going away, and how you can't be a serious critic if you're in denial about how useful it is. Bluesky's gonna love this one! www.nytimes.com/2025/06/16/m...

I do not think it rots people’s brains to say, “I’ve got these ingredients in my fridge, what could I make from them” and get a recipe. Like, there should be safeguards! But I don’t think autocomplete in our IDEs rotted our brains either.

0 reposts23 likes

I agree. And I helped build those features, more than once, on indie food sites. But they don’t have those features now. Because people tend to not seek out individual sites or apps, they’ve been trained to favor generalized interfaces that they’re familiar with.

Or, that specific use case is actually not that useful. If it was, the previous specialized feature would have been used more often. This hits exactly at the reason the supposed utility doesn't stand up to the massive costs / negative externalities for society.

I’m not saying it’s worth the negative externalities. But that’s a separate point. Like, I don’t think sharing pictures of your dog is worth propping up a company that enables a genocide, but it is true that Instagram lets you share photos. It would be stupid to say “no, it can’t share photos”.

Let's use that example. I want to make dinner, I could: a) Perform an indexed search of key ingredients (individually or together) and see what real recipes exist that people have made and found tasty. Read a few, compare, see what other ingredients I'd need. Adjust. Decide. b) Ask Chgpt and have it

give me a probabilistic median of recipe words often used with those ingredients. Zero guarantee that quantities yield the expected result. Backtracking into some other algorithmic output if I say to change anything. Obtain incredibly cringe cooking instructions, fully unproven. Turn off my brain.

Sure when you analyze the effects of a single use. But if everything creative a person produces, be it writing, art, or even coming up with a meal given available ingredients is now offloaded to an LLM, that person's creativity will atrophy.

This is what fundamentally doesn't square. All of the real use cases that get thrown around feel so penny-ante (or actively harmful) against the scale of the costs of LLMs. It's not that they are "useless" it's that the hype machine is so out of whack with the actual use cases.

I think that’s largely true! They’re wildly over-hyped. But people do find utility in the modest use cases for simpler things, and that’s why normal people are using them, often for very mundane reasons. It’s not productive to deny that.

Sure, recipes are fine I guess. But the Roose stuff about how "everybody" is using it is in such bad faith or shows how cloistered they are that it's impossible to take seriously. The denial is more like, these are wildly unsustainable and cannot keep going like this.

autocomplete doesn’t tell you what to do, the decision making is still 100% in your court. you have to think and it saves you typing. and i bet you’ll find people do stop remembering eg the function names if deprived of autocomplete

Handing off cognitive tasks to the computer is the entire point of computers though. Like that's the whole reason productivity has kept increasing the past 40 or so years. It's not a bad thing to outsource tasks to a computer so you can focus on higher-level things