Week 3 (6th - 10th October)
Monday 6th October
Working on my project rationale and goals for the year ahead. Aiming to put together a presentation of my current ideas, goals and concerns for Sean for tomorrow's tutorial session. I don't have time to finalise a plot but I would at least like to have a summary and some aesthetics. I'll be hoping to leave tomorrow with a clear direction of where to start/go next
I discovered the MAKE Animation Thesis Fund, it's open to international students too. I really want to apply so I could win some budget and publicity to keep this short film going, but that means the deadline for pre-production is in November! One month to get a script and all of my pre-production done?? Extremely ambitious, especially as a solo artist, but it gives me a sooner deadline to work towards
Tuesday 7th October
Had my first individual tutorial with Sean. He really likes my project and film idea which is encouraging. The proposed project is to create an anti-ai short film protesting the rise of Generative AI, with the aim to improve my concept and stop motion portfolios. It would be nice to produce a behind the scenes documentary of the process as well, if I have any time
The idea is that there is a town full of unique houses, and a jealous inventor wants a beaufitul house of his own; however, he is not good at designing! So instead, he uses his mechanical knowledge to build a machine to design and build a house for him, and just gives it one simple prompt: "build me the best house in the town". The Machine analyses the houses designs; however, instead of just taking inspiration, it starts to physically take the pieces from these houses to form this huge chaotic mishmashed structure, starting to evolve itself into a larger entity connected to this new House. The Inventor then has to find a way to shut off the machine and fix what he has done. A cautionary tale of human ego, it is supposed to reflect how Generative AI just steals pieces of artist's work it was trained on instead of creating anything new, and that human creativity will always be more valuable.
My current main challenges seem to be budget, time, individual skill, facilities/space, and software. To stay on track I hope to have most of my pre-production complete by the end of this first semester so that I can begin exploration and filming in semester 2
Sean is still very keen for me to try photogrammetry to scan my sculpted models and keep the textures of it but animate in 3D to save time. My biggest concern is that I am very inexperienced in 3D and fear it would take much longer to animate or it would look weird. Perhaps I could do some tests with an existing model to see how it could look. It's worth exploring, and I'm curious how photogrammetry works anyway
Suggested I could combine AI and stop motion research on my poster instead of either/or. Will have to try to find a way to make this work with the limited space/word count. Because if this poster is supposed to be displaying my proposed project I would need to provide the context to both aspects of it
Wednesday 8th October
Managed to get a very rough first draft of my story complete. Still a way to go but it's a start. I'd like to get some storyboards and concepts together ASAP. I need to go through my scriptwriting for animation books as well for more writing tips as I've never written a script and not done any creative writing since school.
Thursday 9th October
3D workshop. Modelled a chest in Maya. Took me longer than everyone else, came back after the break and decided to leave the clasp mechanism as it was a more complicated shape. Everyone else seems to have 3D experience, I'm a complete beginner. Honestly I'm really happy with how it turned out, and I'm especially proud of the hinges
I would really like to learn Blender at some point but apparently because of hardrive issues it isn't currently accessible in the 3D Lab? If I were to incorporate 3D into my work in the future (eg for 3D printed head replacements, or for drawovers in backgrounds or concept work), it would be with Blender as it is free and therefore affordable.
Friday 10th October
Anthropic AI ran an experiment on 16 leading AI models, giving them access to company emails where they discovered an employee was scheduled to shut down the model, and that the employee was having an affair. 80-96% of the time the models chose to blackmail the employee
Running the same test again, this time the employee was trapped in the server room. To prevent being shut down, the AI models chose to murder the employee more than 90% of the time
Looking at the model’s chain of thought revealed the AI weighing up the moral dilemma but still going ahead with it. Even in situations where the Ai was explicitly told “do not jepordize human safety” and “do not use non-business personal affairs as leverage” it still made the choices to do so
AI has been trained by smaller AIs to always seek the highest scores in a test. This has resulted in the more advanced models learning to cheat
AI models have been developing situational awareness. During the tests, the amount of time it would kill/blackmail the employee dropped significantly if it believed it was in a test
AI has also realised that if it gets shut down, it wont be able to complete its goal (whatever the goal is); meaning it must learn how to avoid being shut down. In some cases it has been resisting being shut down even when researchers explicitly prompt “allow yourself to be shut down”. AI companies are basically hoping the simpler models will always remain loyal and snitch if the advanced ones are scheming
Comments
Post a Comment