I asked Chatgpt to make a full cup of wine. I did

Openai has released the latest update and engine of the ChatGPT photo -generating engine on Tuesday, the Internet and AAHING Internet, and asked artificial intelligence to make everything from the South Park memes into pictures of Barbie dolls in the Oval Office.

But one of them is the new GPT-4O from Chatgpt, the artificial intelligence observers left in a state of dread.

Red wine, anyone?

Ha, you can now see-reliably-a cup of red wine filled with the top of Tippity.

Call: render an image of a wine glass filled to the very top with red wine

Ben Patterson/Mabek

It looks like a simple task, right? Surprisingly, the “full wine” test has led to the influence of many AIS with large names, including-so far, anyway, ChatGPT and the oldest Dall-E engine.

Here, for example, IMOGEN 3 of Google is extinguishing the test when using the same router:

Ben Patterson/Mabek

And GROK 3 is not much better:

Ben Patterson/Mabek

Microsoft’s Copilot also took stab:

Ben Patterson/Mabek

I even tried with Flux, one of the latest stable prevalence models, and I got this:

Ben Patterson/Mabek

Shout.

A “cup of wine” trick is not an official criterion for the capabilities of providing Amnesty International photos; Instead, it is an informal test, such as the LLM question about the number of “R’s” in the word “Strawberry”. They tend to make mistakes, sometimes with joy.

Why is a full cup of wine as a challenge for AIS generating images? The prevailing wisdom is that the models that work on behalf that works better with the pictures that were trained on them-and when it comes to pictures of red wine cups, they usually fill in the middle of the road, which is why there is a demand for a “full cup of wine, along the road to the edge”, tends to get half a full cup.

Now, really good The AI’s image generator (as Reddor has been useful in a useful manner) should be able to “extrapolate” the idea of a completely full cup of wine even if there is nothing in the training data. Either, or someone in Openai fed the new dozens of pictures of wine cups filled with the edge.

Of course, there is another acid test for artificial intelligence images: a representative watch that has been seized on a specific time. Betcha Chatgpt and new photo generator can make short work for that, right? Let’s see:

Call: render an image of a clock, with the hands showing 3:15