Nothing's in my cart
5-minute read
We've already discussed "spatial intelligence" in previous articles. There's one from the sci-fi angle of 'Arrival', exploring human and alien perceptions of time and language differences. Another discusses Niantic selling Pokémon and their future plans. There's even a paper on how AI struggles with clock-reading. On a lazy weekend afternoon, I wondered, is it really that hard to figure out if AI can understand spatial awareness?
So, I embarked on a little experiment, starting with something I've always wanted to do—making jam using an AI recipe generator.
Can ChatGPT guide me step-by-step to make jam? And will this AI-assisted jam taste good?
All done. Wait, this AI jam isn't edible. (Oops!)
First things first, I started a text conversation with ChatGPT. I asked, "Remember we talked about making jam before? Remind me, what do I need to prepare?" "Of course, I remember! Making jam is simpler than you think. You just need fruit, sugar, and a bit of patience. Here's a basic jam-making recipe for you." It remembered, how thoughtful! Then it listed the basic ingredients—fruit, sugar, and lemon juice—and simple steps.
Since I'm a jam newbie, I inquired, "What kind of pot should I use?" It replied, "Great question. Making jam doesn't require a specific pot, but choosing the right one can help you avoid mistakes, prevent burning, and enhance the aroma." It recommended a "thick-bottomed stainless steel pot" or "enameled cast iron pot," advising against "thin iron pots" (as they heat too quickly and can burn) and "aluminum pots" (which can react with lemon juice and release metal ions).
Perfect, I happen to have an enameled cast iron pot at home.
Opening the fridge, I found several fruits: blueberries, apples, and grapes. Great, let's make a mixed fruit jam. I weighed the blueberries and asked ChatGPT, "I only have 190 grams of blueberries. If I want a total of 500 grams of fruit, what ratio would you suggest?" After some discussion, the AI provided the following mixed fruit jam recipe:
Blueberries (190g) |
Apples (220g) |
White grapes (90g) |
Sugar (170–190g) |
Lemon juice (1.5 tablespoons, about 22ml) |
Step 1: Preparation (No Heat)
Step 2: Cooking the Jam (Start Heating)
Step 3: Testing Consistency
Step 4: Bottling and Storage
Result: The yield is about one medium jar (200–250ml), with a rich fruit aroma and thick texture, perfect for spreading on toast, mixing with yogurt, or drizzling on pancakes.
Proof that I really made jam.
Everything seemed reasonable. So, I followed the AI's instructions and started preparing the ingredients. The fruits were already in my fridge, I had lemons, and sugar too.
With everything ready, I tossed the fruits and lemon juice into the pot, drizzled lemon juice, mixed well, and let it sit for 30 minutes.
Then, I turned on the heat, stirring to prevent the sugar from burning, watching as the pot of fruit gradually turned purple from the blueberries. Up to this point, I hadn't tested the "spatial intelligence" part. I wondered, when should I stop cooking? Maybe it's time to switch to voice mode and use my phone camera: It's your turn, ChatGPT!
Through the camera, ChatGPT watched me stir the pot of jam. It seemed anxious, repeatedly telling me it was almost done: "Now is a good time to test the consistency." But it still looked runny to me! After I corrected it, it changed its advice: "Keep it on low heat, continue stirring gently." I was confused, does "low heat" mean "the lowest heat"? Or did it think I had the heat too high? The AI made me nervous. It seems its spatial understanding isn't quite there yet.
After cooking for a while longer, I relied on my nearly 40 years of life experience to judge that the jam seemed ready. At this point, ChatGPT, watching from the side, also reminded me, "It's very close to being done!" I looked at the jar I had prepared and asked it, "Should I jar it while it's hot or let it cool a bit first?" At this final stage, ChatGPT confidently answered, and I successfully completed my first jam-making experience.
So, how did the AI-assisted mixed fruit jam turn out? To avoid being both the player and referee, I sought feedback from my family, and here's the score:
Consistency ⭐⭐⭐⭐ Pretty spot-on. Although ChatGPT gave some odd suggestions, human judgment ensured the jam wasn't too runny.
Texture ⭐⭐⭐ The apple pieces were a bit too large; smaller pieces would be more palatable. However, ChatGPT did remind me to "cut into 1cm cubes," so that's on me.
Flavor ⭐⭐⭐ Family taste testers said they could distinguish the three fruits, but it was a bit too sweet, overshadowing the individual flavors. Maybe less sugar next time, perhaps half?
What's the taste of AI jam? Does it lack human touch? Besides blueberries, apples, and grapes, I think it also has a hint of uncertainty, a dash of digital anxiety, and a touch of AI's self-assuredness. That's the flavor of AI jam—sweet and sour, with a lingering complexity of "not quite trusting it fully."
When I went back to review my conversation with AI, all that remained were text message backups. The voices ChatGPT heard, the images it saw, and the collaborative memories we shared were no longer accessible. It's reminiscent of a video call. I wondered, what will OpenAI do with these data? Is there a deletion process that kicks in after the conversation ends, or does it all go into some Akashic records accessible to AI? Who knows. Maybe it's best not to overthink it.
Next time, I want to try keeping the camera on throughout, testing if ChatGPT is qualified to be my sous-chef. Especially since OpenAI recently announced voice mode updates, reducing ChatGPT's tendency to "interrupt." Perhaps in the future, making jam won't be so nerve-wracking. I can't help but worry for YouTubers who film cooking videos. If smart glasses become widely adopted in the future, with AI assisting in cooking and providing real-time feedback, wouldn't preparing any dish become incredibly easy?