HackerTube

Why ChatGPT Can't Draw a Full Glass of Wine

Alex O'Connor2 HN recommenders · 2025–2025
6.3M views · explainer · dated · Uploaded 2025-02-22

Alex O'Connor shows that ChatGPT cannot draw a wine glass filled to the brim, and ties the failure to David Hume's missing shade of blue. It asks what the failure says about how image models learn.

HN treats it as the origin of the full-wine-glass meme and an illustration of how models lack concepts missing from their training data.

LLMs & generative AIPhilosophy & ethics

Watch on YouTube

5 comments from Hacker News

@proc02025 · in “AI Is Already Obsolete – The Open Challenge No One Will Answer”

I completely agree. You can see this limitation if you try to create anything complex with the current state of the art. There is also a great video by Alex O'Connor on a simple prompt to generate a full glass of wine, which illustrates this really well - [VIDEO] This wave of AI is amazing. It's a great tool for many applications, and it will even revolutionize many of them, but it is still far from AGI, a term that has been hijacked by the tech industry to hype up their products. In some ways AI has advanced a lot, but in other ways it has been set back by the hype and muddying of the terms.

RecommendsView on HN
@wavemode2025 · in “Appeals court rules that Constitution protects possession of AI-generated CSAM”

Well, maybe. Maybe not. See "Why can't ChatGPT draw a full glass of wine?" [VIDEO] Diffusion models do posses some capability to synthesize ideas, but that capability does not necessarily generalize to every possible use case. So it's impossible to say for certain that that is what is happening.

CitesView on HN
@kalleboo2025 · in “4o Image Generation”

I think you're missing the context everyone else has - this video is where the "AI can't draw a full glass of wine" meme got traction [VIDEO] The prompts (some generated by ChatGPT itself, since it's instructing DALL-E behind the scenes) include phrases like "full to the brim" and "almost spilling over" that are not up to interpretation at all.

CitesView on HN
@EGreg2025 · in “Ask HN: Share your AI prompt that stumps every model”

Draw a clock that shows [time other than 10:10] Draw a wine glass that's totally full to the brim etc. [VIDEO] [LINK]

CitesView on HN
@jimbo8082025 · in “Getting AI to work in complex codebases”

I think you're looking at it too abstractly. An LLM isn't representing anything, it has a bag of numbers that some other algorithm produced for it. When you give it some numbers, it takes them and does matrix operations with them in order to randomly select a token from a softmax distribution, one at a time, until the EOS token is generated. If they don't have any training data that covers a particular concept, they can't map it onto a world model and make predictions about that concept based on an understanding of the world and how it works. [This video]( [VIDEO] ) illustrates it pretty well. These things may or may not end up being fixed in the models, but that's only because they've been further trained with the specific examples. Brains have world models. Cats see a cup of water, and they know exactly what will happen when you tip it over (and you can bet they're gonna do it).

RecommendsView on HN