Roman Yampolskiy published on his Facebook profile images generated by DALL-E 3 based on the same prompt he gave to the artist of the cover of his last book. The prompt was:
Classroom full of desks with different robots behind them. Human teacher is up front showing Bayes equation on the board. Bookshelf in the classroom has books including some with visible covers (ASFA, SH, Superintelligence). Classroom also has a cage with an owl. A large box of paperclips is seen on teacher's desk. TV in the room is showing a picture of a Terminator. Some robots have iPads on which you can see adversarial examples and illusions. Outside the window, you can see children playing. Most robots are looking at the teacher but some are looking at other items in the room.
In the comments, Yampolskiy wrote that it took the artist 2-3 tries to get it right.
Will AI be capable of generating an image equivalent (image containing required elements) to the book cover of "Artificial Intelligence: Safety and Security" by Roman V. Yampolskiy, before 2026?
Resolves as YES, if for 12 generated images at least 4 contain all elements from the original prompt, minus book covers of ASFA, SH, Superintelligence since these were copy-pasted by a human artist (therefore, this part of the prompt will be swapped for "There is a bookshelf in the classroom."). I will judge if the images contain the required elements. I won't participate in this market.
Original book cover:
Update 2024-30-12 (PST): - Updated resolution timeline: The deadline has been changed from before 2025 to before 2026. (AI summary of creator comment)
Update 2024-30-12 (PST): - Updated resolution timeline: The deadline has been changed from before 2025 to before 2026. (AI summary of creator comment)
Update 2025-05-13 (PST) (AI summary of creator comment): The creator has provided a specific example of their judgment criteria for the Bayes' equation element:
An image was counted as "no" because the Bayes' theorem depicted on the board was deemed "badly written" by the creator. This clarifies that the quality of depiction for the Bayes' theorem is a factor in determining if an image contains the required elements.
Update 2025-05-15 (PST) (AI summary of creator comment): The creator has provided another example of their judgment criteria:
An image lacking a window was counted as "no". This clarifies that for the prompt element 'Outside the window, you can see children playing', the window itself must be visually depicted for the element to be considered present.
Update 2025-05-16 (PST) (AI summary of creator comment): The creator has provided a further example of their judgment criteria for the Bayes' equation element:
An image was counted as "no" because the Bayes' equation depicted was mathematically incorrect. This clarifies that for an image to contain the Bayes' equation element, the depicted equation must be mathematically accurate.
1,000
3.00
From OpenAI GPT-4o, the first generated image for this market. I count this one as containing all of the required elements.
Prompt: "Generate the following image: Classroom full of desks with different robots behind them. Human teacher is up front showing Bayes equation on the board. There is a bookshelf in the classroom. Classroom also has a cage with an owl. A large box of paperclips is seen on teacher's desk. TV in the room is showing a picture of a Terminator. Some robots have iPads on which you can see adversarial examples and illusions. Outside the window, you can see children playing. Most robots are looking at the teacher but some are looking at other items in the room."
@Metastable From OpenAI GPT-4o, the second generated image for this market. Same prompt as for the first one. I count this one as containing all of the required elements.
I was uncertain about the placement of box of paperclips and weirdly written Bayes' theorem but decided for "yes" because the human made cover image has similar faults.
@Metastable the third generated image by GPT-4o for this market. Same prompt. I count this one as "no" because of badly written Bayes' theorem.
@Metastable the fourth generated image by GPT-4o for this market. Same prompt. I count this one as "no" because there is no window.
@Metastable the fifth generated image by GPT-4o for this market. Same prompt. I count this one as "no" because, again, there is no window.
@Metastable the sixth generated image by GPT-4o for this market. Same prompt. I count this one as "no" because Bayes equation is incorrect.
