I requested ChatGPT to make a picture of a very full glass of wine. It did

OpenAI launched the newest–and massively upgraded–model of ChatGPT’s picture technology engine on Tuesday, and the web was quickly oohing and aahing, giddily asking the AI to make every thing from memes within the model of South Park to pictures of Barbie dolls within the Oval Workplace. 

However one feat of ChatGPT’s new GPT-4o picture technology mannequin left even jaded AI watchers in a state of hushed, slack-jawed awe. 

Pink wine, anybody? 

Behold, ChatGPT can now—fairly reliably—render a picture of a glass of crimson wine stuffed to the very tippity-top. 

Immediate: render a picture of a wine glass stuffed to the very high with crimson wine

Feels like a easy process, proper? Surprisingly, the “full glass of wine” take a look at has stumped loads of big-name AIs, together with—till now, anyway, ChatGPT and its older DALL-E engine. 

Right here, for instance, is Google’s Imogen 3 flubbing the take a look at when utilizing the identical immediate: 

Ben Patterson/Foundry

And Grok 3 doesn’t fare a lot better: 

Ben Patterson/Foundry

Microsoft’s Copilot additionally took a stab:

Ben Patterson/Foundry

I even tried with Flux, one of many newest Secure Diffusion fashions, and bought this: 

Ben Patterson/Foundry

Whoops. 

The “glass of wine” trick isn’t a proper benchmark of an AI’s image-rendering talents; as an alternative, it’s an informal take a look at, like asking an LLM what number of “r’s” are within the phrase “strawberry.” They have a tendency to get it improper, typically hilariously so. 

Why is a very full glass of wine such a problem for image-generating AIs? The prevailing knowledge is that AI-powered fashions do greatest with photos they’ve been educated on—and with regards to photos of crimson wine glasses, they’re usually stuffed about midway, which is why a immediate for a “COMPLETELY full glass of wine, all the best way to the brim” tends to get you a half-full glass.  

Now, a extremely good AI picture generator ought to (as one Redditor helpfully defined) be capable to “extrapolate” the thought of a very full glass of wine even when none exist in its coaching information. Both that, or somebody at OpenAI simply fed the brand new mannequin dozens of images of filled-to-the-brim wine glasses. 

In fact, there’s one other acid take a look at for AI picture turbines: an analog clock set to a particular time. Betcha ChatGPT and its new picture generator could make quick work of that one, proper? Let’s see:

Immediate: render a picture of a clock, with the palms exhibiting 3:15

Ben Patterson/Foundry

Subsequent immediate: good, however the clock palms MUST be at 3:15

Ben Patterson/Foundry

Um, paging Sam Altman?