我试用了纽约时报烹饪频道新推出的AI食谱修改功能,还真挺好用的

内容来源:https://www.wired.com/story/nyt-cooking-new-ai-powered-recipe-modifier-is-pretty-good/
内容总结:
《纽约时报》烹饪频道近日在iOS端测试一项AI功能,允许订阅用户用人工智能修改食谱,引发大量用户不满。10月23日相关公告发布后,评论区涌现大量反对声音,用户威胁取消订阅、质疑功能必要性并询问如何关闭。四天内172条评论几乎全是愤怒反馈,随后该文章评论功能被关闭。
《纽约时报》烹饪频道不公开订阅用户数据,但该报执行传播总监乔丹·科恩表示,该网站和应用在2025年访问量达5.14亿次,其YouTube频道订阅用户超过140万。评论者主要担忧AI介入会损害《纽约时报》作为高质量食谱内容制作者的核心价值。该频道拥有60名编辑、撰稿人和视频记者,其食谱以经过严格测试和具有人格化风格著称。
在实际测试中,该AI功能表现好于预期。用户可在食材列表上方找到“定制此食谱”输入框,提出替换食材、调整分量或适配饮食需求等要求,所有修改会以红橙色文字标注,并附有“由AI制作的版本,未经原食谱创作者测试或认可”的提示。测试涵盖鸡肉饺子、谷物碗、华夫饼、意面、酸菜和花生面等多个食谱,多数修改合理,能根据食材特性调整烹饪时间和步骤。但曾出现建议将煮熟的鸡胸肉回锅再煮30分钟导致口感变差的问题。
《纽约时报》烹饪频道总经理卡米拉·贝拉斯克斯表示,AI会依次参考食谱本身、该报食材替换指南、用户评论及AI“专业经验库”给出建议,定位是辅助工具而非主体,并强调该频道“招聘力度前所未有”,“完全没有减少食谱发布量的意图”。她同时表示,用户将可选择在应用中隐藏该AI功能。目前该功能仍仅限iOS端测试,尚未公布其他平台上线时间。
文章作者认为,AI替换功能本质上是减法式修改,很难超越原食谱创作者的精心设计,但其价值在于帮助用户利用现有食材、减少购物次数或适配特殊饮食需求,是技术对产品进行渐进式改良的少见正面案例。作者同时指出,烹饪团队与用户之间的情感联结十分珍贵,用户的强烈反应源于对这一联结可能被削弱的担忧。
中文翻译:
在我最近观看的《纽约时报》烹饪频道的视频中,梅丽莎·克拉克制作了她的柠檬白豆汤配火鸡和绿叶菜。在视频的一个旁白片段里,这位美食记者兼食谱专栏作家邀请观众留言,告诉她他们是怎么处理羽衣甘蓝茎的。这凸显了一个媒体品牌旗下人物与受众之间那种温馨又有趣的关系——一种温暖随性的氛围。
我并不属于烹饪版块评论区大军中的一员,但几天后,我非常惊讶地看到其中一些人在公开抗议。更让我惊讶的是原因:在一个仅限iOS端的测试版中,《纽约时报》正在使用AI让订阅者修改食谱。
在10月23日宣布这一变化的文章下方,评论者们大声表达不满,威胁要取消订阅,质问谁需要这个功能,以及如何关闭它。四天后,在172条几乎全是愤怒的评论之后,该文章的评论功能被关闭了。
《纽约时报》不公开烹饪版的订阅者数量,但NYT执行传播总监乔丹·科恩表示,该网站和应用程序在2025年的访问量达到5.14亿次,其YouTube频道拥有超过140万订阅者。读完该版块的AI计划,再想想我见过多少次将技术生硬塞进厨房里根本不需要技术的地方的失败尝试,我猜我会站在评论者那一边。然后我试用了新的AI功能,发现它出奇地好用。
过去十年我为《连线》撰写的文章一次又一次地告诉我,给厨房产品添加技术充其量是有风险的。我反复目睹公司尝试并失败:所有电器都加蓝牙连接、过度依赖糟糕食谱的应用程序、配有光学传感器或触摸屏却仍然烤不好面包的烤面包机。几乎所有属于“智能厨房”的东西往往根本不怎么智能。
开发可靠、美味的食谱是艰难且昂贵的工作,可能需要一整支团队。然而偷工减料会损害品牌。两家拥有深厚人才储备的知名公司——《纽约时报》烹饪版和美国测试厨房——都是烹饪内容创作者。食谱,也就是美味、高质量、经过测试的食谱,正是他们的看家本领。
既然AI往往是个糟糕的厨师,让它在NYT这座内容金矿的呈现中扮演角色是一个巨大的风险。
今年早些时候,《纽约时报》允许AI协助开发了一个食谱缩放功能(调整食材用量,把“4人份”变成“6人份”),以及将英制计量转换为公制——杯、汤匙和华氏度转换为克、毫升和摄氏度。现在,在仅限iOS的测试版中,食材列表上方出现了一个“定制此食谱”的输入框。
如果你想做香蕉面包但告诉它你没有吐司模具,它会问你有什么样的烤盘,然后调整食谱以适应你的情况。食谱中所有的改动都会以醒目的红橙色文字标注。经过几天的测试,我看到了你可以做相对简单的事情,比如把猪肉末换成牛肉末,或者提出更复杂的要求,比如让它根据你的饮食需求进行调整。它所做的修改在整个食谱中都非常醒目。如果你问它太离谱的东西——比如把埃里克·金的辣椒酱黄油面条变成饼干——它会给你一个“硬性重定向”,指向一个更合适的食谱,在这个例子中是金的辣椒酱焦糖饼干。
“我们投入了大量精力,试图在某个改动不再符合食谱精神时告诉你不要那样做,”NYT烹饪版总经理卡米拉·贝拉斯克斯说。
对于其推荐,大语言模型会依次参考食谱本身、《纽约时报》的食材替代指南、用户评论,然后是她所谓的AI“借鉴我们的专业知识”。贝拉斯克斯说AI应该是一个助手,而不是主角,并指出:“你不应该跟它聊天。”
她解释说,当《纽约时报》构建AI功能时,食谱编辑帮助创建了最初的规则集,然后在最后评估了它的运作效果。
当我问这对60名编辑、撰稿人、视频记者和其他员工——也就是创建人们使用的食谱、策划读者所共鸣的氛围的团队——的工作意味着什么时,贝拉斯克斯强调NYT烹饪版一直在“比以往任何时候都更多地招聘”,并且“完全没有减少食谱发布量的打算”。
她说,到目前为止,人们主要用它来做食材和器具的替换,以及根据饮食需求调整食谱。“有些人用聊天机器人来缩放份量,或者根据现有食材来缩放,”她举例说,有人只有一杯油,而食谱需要更多。
作为一个因AI而职业生涯发生了一些重大转折的美食作家,我不会说我一直在大声呼吁这个功能,但我发现《纽约时报》这种审慎的做法效果相当不错。
我开始测试时,用艾莉森·罗曼的一道五星食谱做了鸡肉饺子,该食谱有超过6000条评分。我让AI工具把带骨带皮鸡腿换成无骨去皮鸡胸。它思考了一会儿——用AI烹饪的行话说就是“炖煮”——然后给出了“鸡胸肉饺子”,标题中的“胸”字以红橙色呈现,表示经过了修改。就在标题下方,它写着这是“约瑟夫用AI制作的版本。未经原食谱创作者测试或认可。”除了这些可见的变化,我还可以在我的版本和原版之间来回切换。
一旦开始烹饪,一切都很顺利。我是照着食谱做的人,不是即兴发挥型,而《纽约时报》的食谱很容易跟随。修改后的版本也是。同样让我印象深刻的是,应用程序标注了鸡胸肉的炖煮时间会缩短,因为它比鸡腿肉熟得更快,而且由于鸡胸肉不像带皮鸡腿那样出那么多油脂,它在面糊中加了一些黄油。这些改动被纳入了食谱中正确的位置。
一切都很合理,除了一点。它建议把撕碎的熟鸡胸肉放回锅中再煮约30分钟,同时酱汁收浓、饺子煮熟。我预见到了这一点,嗯,我可不会那样对待我的晚餐。我按指示放了一小把撕碎的肉进去以便看看会发生什么,其余的留了出来。在那额外半小时里煮着的碎肉本身并不差,但口感明显比我留出来的那些逊色,更远不如鸡腿肉。
这是我在所有测试中遇到的最大问题。如果我是《纽约时报》原食谱的创作者,我不确定自己会如何看待用鸡胸替代鸡腿的选项——这并非闻所未闻,但在经典鸡肉饺子食谱中并不常见。也许这是可以微调的,或者干脆不作为选项提供。
从那里开始,一切就比较顺利了。我让它把弗朗西斯·林的烤蘑菇西兰花谷物碗变成烤蘑菇、花椰菜和南瓜谷物碗。十字花科的替换是一对一的简单操作,但我还告诉它加一磅蜜糖南瓜。它让我把南瓜切成半月形,这样所有食材能同时熟透,然后增加了橄榄油的总用量。林的诗意语言在两个版本的食谱中都得以保留,告诉你要把橄榄油加热“直到它像水一样在锅中滑动”,然后加入打散的鸡蛋做成节庆菜肴顶部的蛋丝。我的朋友戴夫过来参加 cribbage 纸牌之夜,这顿令人愉快的健康赛前晚餐,之后他又两次把我打得落花流水。
我一直很好奇它会如何处理面糊类食谱,它成功地将耶万德·科莫拉费的烤芝麻葱花华夫饼改造成了酵母版本,与原版不同的是,这个版本需要在冰箱里过夜,让活性干酵母起泡并增添诱人的酸味。
一个朋友推荐了科卢·亨利的意面鹰嘴豆炖菜。我把食品柜里的罐装番茄丁换成了整颗番茄,并要求它“多加蔬菜”。它给出了与食谱相符的选项——胡萝卜、芹菜、洋葱、西葫芦、蘑菇和额外的绿叶菜——然后把我选择的食材放入了整个操作流程中正确的位置,使所有食材都在正确的时间加入。做出来的晚餐健康又丰盛。
最近去了波特兰发酵节,想要一些关于在我的酸菜中加入几种额外食材的指导,尤其是大蒜,我用AI修改了莱迪·霍伊克的原版食谱,说“加胡萝卜、大蒜和羽衣甘蓝”,它分别建议擦丝、切薄片和切碎,用量比例感觉都对。它还增加了盐量,因为我做的酸菜更多了。我打字时那大罐酸菜就放在我旁边,虽然离做好还有几天,但闻一下就知道它正朝着好批次的方向发展。
最后,我把苏·李的花生面配辣豆腐改成了花生乌冬面配豆腐、白鱼和黄瓜。我一开始要求它把原版做得不那么辣、加蔬菜,并“保留豆腐”,但加入我在超市找到的一些好看的石斑鱼。令人印象深刻的是,由于鱼是额外添加的,它让我先把鱼煮熟,然后盛出并擦干净锅(!)再煮豆腐。由于我把原版的粗管面换成了乌冬面,它也定制了煮面的指导。
烹饪版在推出这个功能时遇到一些阻力似乎很自然。《纽约时报》雇用了业内一些最好的食谱创作者。用AI帮助我们做食材替换是有用的,应用程序将这些改动整合到食谱中的方式也令人印象深刻,但这种对原版的修补本质上是一个减法过程。我们中有多少人做替换后能做出比梅丽莎·克拉克、耶万德·科莫拉费或弗朗西斯·林当初设计时更好的东西呢?
相反,它帮我们省了一趟去商店的路,或帮我们用手头现有的东西,或让食谱变成无麸质的,作为一个如何用技术让产品逐步改进的正面且罕见的例子。
听起来评论者们的担忧也被听到了。贝拉斯克斯在我们交谈时马上告诉我,如果人们愿意,可以隐藏应用程序中的AI功能。《纽约时报》目前也把整个功能保持在仅限iOS的测试版中,且没有透露何时会在其他平台上推出。
你能看到烹饪团队与其用户之间的那种纽带是一件非常美好的事。内心深处,我猜那些不满的烹饪版评论者担心他们会失去一部分人与人之间的连接。我希望这种连接永远不会消失。
评论
返回顶部
英文来源:
In an aside in the New York Times Cooking video I watched recently, where Melissa Clark made her lemony white bean soup with turkey and greens, the food reporter and recipe columnist invited viewers to leave a comment telling her what they do with the kale stems. It highlighted this sweet, funny relationship between a media brand’s personalities and its audience, a warm-casual vibe.
I am not among the multitudes in the Cooking section's commenter army, but a few days later I was very surprised to see some of them in open revolt. I was even more surprised to see why: In an iOS-only beta, the Times is now using AI to allow subscribers to modify recipes.
Below the article announcing the change on October 23, commenters loudly aired their grievances, threatening to cancel subscriptions, asking who wanted this, and how to turn it off.
After four days and 172 almost exclusively angry comments, commenting for the story was turned off.
The Times doesn't share Cooking's subscriber numbers, but NYT executive communications director Jordan Cohen said the site and app were visited 514 million times in 2025, and its YouTube channel has over 1.4 million subscribers. Reading over the section's AI plans and thinking about how many bungled attempts I'd seen to mush technology into places where it doesn't need to be in the kitchen, I guessed that I would side with the commenters. Then I tried the new AI feature, and found that it worked surprisingly well.
My writing for WIRED over the past decade has shown me again and again that adding tech to a kitchen product is risky at best. I've repeatedly watched companies try and fail: Bluetooth connections for all the appliances, apps that lean too heavily on crummy recipes, toasters with optical sensors or touchscreens that still couldn't toast particularly well. Just about anything that was part of the "smart kitchen" tended not to be very smart at all.
Developing dependable, delicious recipes is hard, expensive work that can require a whole staff. Yet skimping can hurt the brand. Two notable companies with a deep bench—The New York Times Cooking and America's Test Kitchen—are cooking content makers. Recipes, which is to say delicious, high-quality, tested recipes, are what they do.
Since AI is often a lousy chef, giving it a role in the presentation of NYT's gold mine of content is a big risk.
Earlier this year, the Times allowed AI to help with a recipe-scaling feature (modifying ingredient quantities so "serves 4" turns into "serves 6") and converting from imperial measurements to metric—cups, tablespoons, and Fahrenheit to grams, milliliters, and Celsius. Now, in the iOS-only beta, a "customize this recipe" field appears just above the ingredient list.
If you want to make banana bread but tell it you don't have a loaf pan, it'll ask you what kind of pans you do have, then modify the recipe to suit. All the changes throughout the recipe are noted in a visible red-orange text. Over several days of testing, I saw how you can do relatively simple things like swap ground pork for ground beef, or make more complicated demands like asking it to tailor things to your dietary needs. The modifications it makes are very visible throughout the recipe. If you ask it for something too out there—like turning Eric Kim's gochujang buttered noodles into a cookie—it gives you a "hard redirect" and points you to a better-suited recipe, which in this case is Kim's gochujang caramel cookies.
"We put a lot of effort into trying to tell you when not to do something if it's no longer in the spirit of the recipe," says Camilla Velasquez, the general manager of NYT Cooking.
For its recommendations, the LLM looks at the recipe itself, the Times’ ingredient substitution guide, user comments, and then what it calls AI "drawing on our expertise," in that order. Velasquez says the AI should be a helper, not the main act, noting, "You shouldn't be chatting with it."
She explains that when the Times built the AI functionality, recipe editors helped create the original rule set, then evaluated how it worked at the end.
When I ask what this means for the jobs of 60 editors, writers, video journalists, and other employees—the team that creates the recipes people use and curates the vibe readers relate to—Velasquez stresses that NYT Cooking has been "hiring more than ever" and that it has "zero intention of publishing fewer recipes."
So far, she says people are mainly using it for ingredient and equipment substitutions, along with adjusting recipes for dietary needs. "Some people use the chatbot to scale, or they'll scale to match an ingredient," giving the example of someone who had only one cup of oil for a recipe that needed more.
As a food writer whose career has taken some substantial turns due to AI recently, I wouldn't say I've been clamoring for this, but I find the Times’ measured approach works pretty well.
I started testing it out by making chicken and dumplings using a five-star Alison Roman recipe with more than 6,000 ratings. I asked the AI tool to swap out bone-in, skin-on chicken thighs for boneless, skinless breasts. It thought about it for a moment—"simmering" in AI Cooking patois—and then offered up "Chicken Breast and Dumplings," with the "breast" in the title rendered in the red-orange color that indicates a modification. Just below the title, it said it was "Joseph's version made with AI. Not tested or endorsed by the original recipe's creators." Along with those visible changes, I could also toggle back and forth between my version and the original.
Once I started cooking, things went smoothly. I'm a recipe follower, not an improviser, and the Times’ recipes are easy to follow. The modifications were too. I was also impressed by how the app flagged that the chicken breast simmering time would be reduced because it cooks faster than thighs, and since they don't render as much fat as skin-on thighs do, it added some butter to the roux. These changes were incorporated into the right places in the recipe.
It all made sense except for one bit. It suggested the shredded, cooked breast go back into the pot for about another 30 minutes while the sauce thickens and the dumplings cook. I saw this coming and, well, I wasn't going to do that to my dinner. I put a handful of shredded meat in as directed so I could see what happened and reserved the rest. The shredded bits that were in for that extra half hour weren’t bad per se, but their texture was notably less succulent than what I'd reserved and way less than thighs would be.
This was the largest issue I encountered in all of my testing. If I had been the creator of the Times' original recipe, I'm not sure how I'd feel about the option to use a breast instead of thighs, which is not unheard of, but not very common in classic chicken and dumpling recipes. Perhaps it's something that could be fine-tuned or simply not offered as an option.
From there, it was fairly smooth sailing. I had it turn a Francis Lam roasted mushroom and broccoli grain bowl into a roasted mushroom, cauliflower, and squash grain bowl. The crucifer crisscross was an easy one-to-one, but I also told it to add a pound of delicata squash. It had me slice the squash into half moons so everything would finish at the same time, then upped the total amount of olive oil. Lam's poetic voice was preserved in both versions of the recipe, telling you to heat the olive oil "until it slips around the pan like water" before adding beaten eggs for festive dish-topping egg ribbons. My buddy Dave had come by for cribbage night, and this made for an enjoyable, healthy, pregame dinner before he went on to skunk me twice.
I'd been curious how it would handle batters, and it capably transformed Yewande Komolafe’s Toasted Sesame and Scallion Waffles into a yeasted version, which, unlike the original, gets an overnight in the fridge, letting the active dry yeast get bubbly and add an appealing tang.
A friend suggested Colu Henry's pasta and chickpea stew. I swapped canned diced tomatoes I had in my pantry for whole, and asked it to "add more vegetables." It gave options in line with the recipe—carrots, celery, onion, zucchini, mushrooms, and extra greens—then put what I chose into the right spots throughout the procedure so everything went in at the right time. It made for a healthy, hearty dinner.
Having been to the Portland Fermentation Festival recently and wanting guidance on incorporating a few extra ingredients, particularly garlic, into my sauerkraut, I used the AI to modify Lidey Heuck's original, saying "add carrots, garlic, and kale," which it suggested grating, thin slicing, and mincing, respectively, all in the right-feeling proportions. It also added more salt, since I was making more kraut. The big jar is sitting next to me as I type, and while it's several days from being ready, a sniff test tells me it's on its way to being a good batch.
Finally, I turned Sue Li's peanut noodles with spicy tofu into peanut udon with tofu, whitefish, and cucumber. I started by asking it to make the original less spicy, add veggies, and "keep the tofu," but add some good-looking rockfish I found at the grocery store. Impressively, since the fish was an add-on, it had me cook it, then set it aside and wipe out the pan (!) before cooking the tofu. Since I swapped udon for the original bucatini, it customized the noodle-cooking guidance, too.
It seems pretty natural that the Cooking section ran into some resistance when launching this feature. The NYT employs some of the best recipe creators in the business. Using AI to help us make substitutions is helpful, and the way the app integrates those changes into the recipes is impressive, but this tinkering with the original is essentially a subtractive process. How many of us are going to make substitutions and land on something better than Melissa Clark, Yewande Komolafe, or Francis Lam set out to make by design?
Instead, it's saving us a trip to the store, or helping us use something we have on hand, or making it gluten-free, serving as a good and rare example of how you can use technology to make a product incrementally better.
It also sounds like the commenters’ concerns have been heard. Velasquez let me know right away when we spoke that people will be able to hide the AI feature in the app if they want. The Times is also keeping the whole thing in iOS-only beta for now, and not saying when it would be available on other platforms.
That bond that you can see between the Cooking team and its users is a pretty sweet thing. Deep down, I'm guessing the upset Cooking commenters felt worried they were going to lose some of that human connection. I hope it never goes away.
Comments
Back to top
文章标题:我试用了纽约时报烹饪频道新推出的AI食谱修改功能,还真挺好用的
文章链接:https://news.qimuai.cn/?post=5508
本站文章均为原创,未经授权请勿用于任何商业用途