Microsoft is back in action another round of ads showing people talking to Copilot AI on their computers. This time it’s a Christmas theme and includes a cameo from the huge man in red. The 30-second TV spot asks if you’re “ready for the holidays” and features the cast in various holiday home settings, asking the co-pilot to lend a hand with everything from holiday lighting and cooking to huge outdoor decorations.
Just like I tested Copilot Vision and Voice Mode last time, I’ve made a list of all the prompts in the Microsoft ad and I’m double-checking them. Or as many times as I can stand before I hit my leg.
IN new adCopilot offers to lend a hand a homeowner make their intelligent home more festive when prompted: “Show me how to sync the Christmas lights to my music.” The user clicks on cloud-connected intelligent lighting controls on a website called Refresh the cloud as Co-pilot says, “Let’s get through this together.” The ad cuts to the pulsating house lights set to the classic holiday song “A-Punk” by Vampire Weekend.
Relecloud is not as well-known a intelligent home company as Philips Hue or Govee. In fact, it’s not a real business at all. This is one of the fictitious companies that Microsoft uses published case studies (see also: Contoso). I’m inclined to believe that its operate in this case indicates a simulation of all of Copilot’s advertised actions, but the Microsoft representative insists that this is not the case. This is Nicci Trovinger, General Manager of Windows Marketing Edge“All Copilot responses are actual responses Copilot provided to the scenarios shown and questions asked at the specific point in time. Responses have been shortened to fit the length of the creative spot, consistent with standard advertising practices.”
I tried this test with Copilot in two ways: one using a still image of the lighting interface from the ad, and the other using the Philips Hue Sync app. Copilot made its best guess at where I should click on the ad image by highlighting the “Sync Mode” drop-down menu with its own screen cursor, but it couldn’t get me much further. He would often tell me that he was highlighting something when it wasn’t, and hallucinating a green “Apply” button that was actually a set color for green lighting.
Copilot’s responses were equally confusing when presented with a fully configured app for my Philips Hue lights. It identified the Hue Sync app and initially correctly told me to click the Music tab and the “Start Light Sync” button. But then it hallucinated buttons that weren’t there, pointed me to Entertainment Zones that I had already set up, and kept saying that it highlighted things on the screen when it didn’t. The cursor highlight feature in Copilot is useful, but it usually only does it when you ask it to, and the response is painfully ponderous. It often stays on the screen long after Copilot moves on to another tip.
Several other prompts in the ad remain unanswered, including “Help me develop these instructions” and “Convert this recipe on screen to read 12.”
“14!” another actor, probably their partner, interjects.
In these cases we can’t see what’s on the screen, so I tried to show the Ikea Copilot as part of the instructions Kallax shelf 2×4 — a classic with online assembly instructions. The co-pilot continued to identify the studs as screws or nails. It often detected page numbers as step numbers, making any attempts to follow along even more disordered.
Scaling up the recipe for a mass gathering is a common holiday practice in my Italian-American family. I tried showing Copilot recipe for stuffed mushrooms with Sip and feast. She admitted that going from six to 14 servings would require multiplying each ingredient about 2.3 times, but she usually only did a few calculations before expecting me to do the rest or try to move on to another topic by asking me a question. When she noticed that there were options on the site to augment the size of the recipe, she confused the “2x” and “3x” buttons with pluses and minuses, which allowed me to select exactly 14 servings, and insisted that that was what those buttons were for. They are not. Then, at the last minute, I asked him to simply calculate each ingredient and spell it out in the document. The co-pilot told me yes and then did nothing.
In a recent ad example, a homeowner with intelligent lighting asks, “Can you read the HOA guidelines and make sure I’m not breaking any rules?” The screen shows a made-up document titled “HOA Guidelines” and a photo of a giant inflatable reindeer in front of the house, encroaching on the neighboring property. The image also appears to be AI-generated; Microsoft’s Trovinger confirmed that “Both the reindeer yard decoration photo and the HOA guidelines document were created for advertising purposes and based on testimonials.”
The co-pilot tells the actor in the ad to “position the inflatable reindeer so it doesn’t go into the neighbor’s yard.” When I showed Copilot a screenshot of the document and image from the ad, it detected the line that “large inflatables cannot cross your property line.” But the co-pilot only gave me evasive answers about whether the reindeer had flown too far or not. Reindeer seemed to be speculated power I push boundaries, but I often defer to my judgment.
There is one more incentive in the announcement. It’s jolly elderly St. himself. Nick, who asks the co-pilot why toy production is sinking. In the ad, Co-Pilot claims it’s because the elves drank too much heated cocoa. Perhaps it’s because management is pushing to push AI into their workflows.
I have to hand it to Microsoft’s marketing team for including what seems like an admission that the entire ad campaign is selling a fantasy. Believing that Copilot can do what Microsoft claims — or that any of these AI assistants can do — is like believing in Santa Claus.