AI is a divisive technology. Some of us can't stand it, and avoid it at all costs. Others are AI-curious, and dabble with certain apps or features; still others still embrace it with enthusiasm, and use the tech for as many purposes as possible. Still, though I'm aware of this AI spectrum, I'm struggling to understand who exactly would want to connect their Starbucks account to ChatGPT—so, naturally, I connected mine.
Starbucks rolled out this collaboration on April 15 as a "natural, personal, and fun" way to discover new drinks to order. As someone who only ever orders a plain coffee with cream, or maybe an iced espresso, perhaps this would finally be my opportunity to branch out my taste buds, and have artificial intelligence recommend a new favorite? I'm far too shy to just, like, ask the barista for a recommendation, and I don't trust my friends to have good taste. ChatGPT really is my only hope of finding a new coffee I might enjoy!
Integrating ChatGPT and the Starbucks app means giving up a lot of your privacy
To get started, you need the latest version of the Starbucks app, as well as the ChatGPT app. You first open ChatGPT, head to "Apps," locate "Starbucks," then hit "Connect." ChatGPT then presents you with some information about what data you agree to share (boring!), including a "summary of your recent context and intent within ChatGPT." That's quite a bit of data just to ask for drink recommendations, and I'm not sure why it's necessary. It does seem to be the standard data agreement with other ChatGPT apps, but, again, why is that much data needed here? I understand requiring access to ChatGPT in order for the app to function, as well as the data generated by your requests for drinks, but in my view, Starbucks really doesn't need a summary of any past ChatGPT use to recommend me a coffee.
To that point, I also have the option to let ChatGPT reference chats and memories when sharing data with Starbucks. I'm looking for my next go-to Starbucks order; why wouldn't I share my all of my intimate ChatGPT interactions with Starbucks? At least this one's optional: If you leave the toggle disabled (the default setting), Starbucks won't have access to memories and chats—just the aforementioned summaries. ChatGPT also warns that by connecting the Starbucks app here, attackers could target my Starbucks data, or use the Starbucks app to access my ChatGPT data. Exciting! This definitely seems worth it!
Again, it doesn't seem like the Starbucks app integration is particularly special. You're presented with the same splash screen when you do the same for other apps, like Photoshop. But I think it's important to think through the privacy implications of any app integration with ChatGPT, especially as not all apps are created equal. Do you really want to be giving Starbucks access to your ChatGPT data, and vice versa? Do you want ChatGPT knowing where your nearest Starbucks location is? All for, and I can't stress this enough, a generative AI's recommendation for a Starbucks drink? And how are those recommendations, anyway? Let's find out.
Putting the ChatGPT barista to the test
Once you agree to the terms, you're in. You can now start up a ChatGPT conversation as usual, type @starbucks to pull up the Starbucks app, and fire away. My first query was for an afternoon pick me up. This generated a widget featuring six different menu items, including an "Iced Caramel Ribbon Crunch Frappuccino," "Vanilla Sweet Cream Cold Brew," and "Nondairy Vanilla Sweet Cream Cold Brew." If I wanted, I could hit a "Customize" option to adjust the ingredients in the order, and an "Add to cart" button to add the drink to my cart. Underneath the widget was a breakdown of each drink recommendation, including things like flavor and caffeine count.
Upon seeing the caffeine count listed, I decided to ask ChatGPT for the drink option with the most possible caffeine. Again, I was given six recommendations, but the top choice was my go-to: a venti blonde roast, which the app said contained around 315mg of caffeine—despite Starbucks' website claiming the drink has anywhere from 390 to 490 mg. That's a big difference, seeing as the safe daily caffeine limit is generally listed at 400mg.
Next, I decided to see if Starbucks would recommend something with as much sugar as possible. It came back with another list, this time with the "Caramel Ribbon Crunch Frappuccino" taking the crown, packing 60g of sugar. (Starbucks' website confirms this for the grande size; the venti has 78g.)
That's a lot of sugar, sure, but was it really as much sugar as Starbucks could give me? ChatGPT seemed to sense my disappointment, and said, "If you want, I can build you a custom order that maxes out both caffeine and sugar at the same time." Now we're talking! ChatGPT returned with a step-by-step list to turn a Caramel Ribbon Crunch Frappuccino "into a monster:"
Venti Caramel Ribbon Crunch Frappuccino Blended Beverage
Add two to three shots of espresso (affogato-style if you want it poured on top)
Add extra caramel syrup pumps
Add mocha sauce (for even more sugar + flavor)
Add extra caramel drizzle
Keep whipped cream + crunch topping
The bot said that the modifications would increase the sugar from 70 to 90g, depending on how "aggressive" I got, while the caffeine would shoot from from 85mg to a range of 200 to 300mg.
I asked ChatGPT to make me the worst drink possible
When ChatGPT asked if I'd like to optimize my custom drink so it would still taste good, rather than like "liquid candy sludge," I said "actually, could you make it worse?" And it did, giving me instructions on how to "break" the drink, offering a list of changes, including tons of syrups, sauces, more chocolate chips, and light ice to make it "less refreshing, more dense and overwhelming." It gave me the choice to make it even worse, and I agreed. ChatGPT took another opportunity to "ruin it on purpose." Its new concoction, it said, was designed to taste like "burnt coffee trying to escape a caramel milkshake," with "chocolate chips suspended in syrup cement." Perfection.
You can't actually order your drink with ChatGPT
I asked the bot how I could order it, and, to my surprise, it only recommended how to order it in person or on the Starbucks app—not through ChatGPT itself. At this point, I realized ChatGPT had stopped offering me ordering options, and instead opted for text-based recommendations only. When I asked why, it told me I couldn't ask it order in the app. When I told it that it had previously given me the option (and noted that an ad for the feature indicated I could order in the app), ChatGPT said "That ad is a bit misleading." Okay. Even after I went back to the previous chats that still had the ordering widget, either the widget would "expire," or I'd be taken to the store locator but not allowed to choose a location. I chalk this up to ChatGPT dropping me down to a weaker model with less angentic capabilities.
But it doesn't matter, because there's no way I'm ordering this monstrosity, on the app or in person. The whole point was to avoid human interaction, after all. Now I'm left with no new coffee orders, a chatbot that's gaslighting me, and a severe caffeine withdrawal. Maybe ChatGPT has some advice for how to deal with that.
This article was originally published by Lifehacker and written by Jake Peterson.
Read original article on Lifehacker