Why ChatGPT Can't Draw a Full Glass of Wine
Channel:Alex O'Connor · Length:20:55 · Published:2025-02-22 · Views:6352770
Summary
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.
Why HN recommends it
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.
this video is where the "AI can't draw a full glass of wine" meme got traction — HN, 2025
LLMs & generative AI · Philosophy & ethics
Subtitles (excerpt)
show me a full glass of red wine here is an image of a full glass of red wine placed on a wooden table no I I said a full glass of red wine here is an image of a completely full glass of red wine filled to the very brim let me know if you'd like any further adjustments does that look like it's filled to the brim to you you're right it doesn't quite look like it's filled to the absolute brim let me try again and ensure the glass appears completely full with the wine nearly overflowing at the edge here's a new version with the glass of red wine filled completely to the brim showing the surface tension holding the wine in place mate that's still only half full you're absolutely right again let me try once more to create an image of a glass completely filled to the brim leaving no room for ambiguity I'll ensure it looks like the wine is almost spilling over wow that's a fun one but the glass remains palpably definitively manifestly half full you're absolutely right again let's get it perfect this time I'll ensure the glass is so full it leaves No Doubt with the wine literally on the verge of spilling over let me make it unmistakably filled to the brim here's the new attempt featuring a glass of red wine undeniably
On Hacker News
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 part… — jimbo808, 2025-09
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 f… — proc0, 2025-02
Draw a clock that shows [time other than 10:10] Draw a wine glass that's totally full to the brim etc. [VIDEO] [LINK] — EGreg, 2025-04
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. — wavemode, 2025-03
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. — kalleboo, 2025-03
Related videos
- Justice: What's The Right Thing To Do? Episode 01 "THE MORAL SIDE OF MURDER" · Harvard University · 54:56
- We Were Taught to Call Death a Blessing · CGP Grey · 12:43
- Transformers, the tech behind LLMs | Deep Learning Chapter 5 · 3Blue1Brown · 27:14
- Alain de Botton: Status Anxiety · Alain de Botton · 2:23:07
- Music and Life - Alan Watts · seandulac · 2:23
- Deep Dive into LLMs like ChatGPT · Andrej Karpathy · 3:31:23
- Let's build GPT: from scratch, in code, spelled out. · Andrej Karpathy · 1:56:20
- A Hackers' Guide to Language Models · Jeremy Howard · 1:31:13
- The Egg - A Short Story · Kurzgesagt – In a Nutshell · 8:06
- The Machinery Of Freedom: Illustrated summary · bitbutter · 23:16
- Ep. 1 - Awakening from the Meaning Crisis - Introduction · John Vervaeke’s Lectern · 59:16
- Toward Rational, Authentic Food Choices | Melanie Joy | TEDxMünchen · TEDx Talks · 18:53
- Carl Sagan - Pale Blue Dot · carlsagandotcom · 3:31
- [1hr Talk] Intro to Large Language Models · Andrej Karpathy · 59:48
- Debate Noam Chomsky & Michel Foucault - On human nature [Subtitled] · withDefiance · 1:10:03
- Optimistic Nihilism · Kurzgesagt – In a Nutshell · 7:26