AI Updates March 2025
A new State-of-the-Art and are Graphics Designers jobs at risk?
Note: From a post on 31 March.
The past week's release of Gemini 2.5 Pro, GPT‑4o Image and the open-source (weights) DeepSeek-V3-0324 continue to show steady progress in capabilities.
Gemini 2.5 may be considered the new overall state of the art with an incredible long context length and GPT-4o has brought about the Ghibli-fication of the world.
*The author is of east Asian descent and is partial to the anime art style.
Are graphics designers job at risk?
Recent Gemini and GPT-4o releases may indicate that basic image generation and editing can be done competently with little human expertise, leading many graphics designers to worry about the field's professional viability. There will definitely be edge-cases where professional expertise is still required but overall demand may be on the decline.
Don't starve the goose that lays the golden eggs
The Ghibli-fication of the world has shown our deep appreciation and enjoyment of art. Should we as a society leave artists to stave and deny ourselves more great works?
Should governments step in?
There may be a case for governments to pro-actively provide support for possible job displacement if the speed is unprecedented compared to past rate of technological change. Companies deploying AI may want to look into mitigating the negative sentiment from job displacements, which may cause a backlash and slow down future adoption, by commiting to AI being beneficial to all.
Ghibli art generation and Mayazaki
(Not a legal expert) Art styles are generally not copyrightable and the cease and desist letter story seems to be a prank or manufactured outrage. Mayazaki (of Ghibli) calling AI "An insult to life itself" in a 2016 documentary may also be misconstrued. Mayazaki was commenting on an how an AI program used to generate the movement of zombie limbs reminded him of a someone he knew who had a walking impediment.
Will AI spell the end of human creativity?
Unlikely, as humans have always valued art and will continue to do so regardless of if they can make a living from it. More likely, professional pursuit of art may be increasingly difficult without support from patrons.
Silver lining: If we all get through the disruption from AI in one piece, the future of art may be unimaginable creative as humans leave more popular mainstream art to AI and focus on more groundbreaking ones. Hopefully without the starving artist trope.
Does a model capabilty of replicating the Ghibli art style indicate that the model must have violated copyright?
If we were to steelman the argument, there are plausible ways for a model to replicate the Ghibli art style without being trained on copyrighted materials.
1. Fan art
For a highly popular art style like Ghibli, there are likely many fan art that is made in a similar style.
2. Interpolation
As AI models are capable of interpolation, multiple art styles could possibly be combined to create the Ghibli art style without any Ghibli materials used in the training.
This makes it less clear cut that a model capable of replicating the Ghibli art style must have been trained on copyrighted materials.
For those who only want to use an art generation model that is 'ethically sourced' there are generation models such as Firefly, that claims to be trained only on licensed, copyright free or copyright expired public domain content.
To end on a hopeful note, the Interesting World Hypothesis (Podcast: substack, spotify) suggests that in a future world with Friendly Independent AGI, the work of artists would be highly valued for their interestingness.