Monthly Archives: August 2023

Week 2 Blogpost

Here is my group’s media sketch of the week:

https://docs.google.com/document/d/1QCMD8XnfzsrHwsaM-O5Ti9MecfXdOf1LJNf6HPY-Lyc/edit?usp=sharing

Please use your RMIT account to access the document

Week #2 Vlog Reflection:

Week 3 Blogpost

Here is the access link to our media sketch:

https://docs.google.com/document/d/1Jlv0gk9_BcMvUyvkUE2_nK9qGrUAmfpYDLrZL6VqEfc/edit?usp=sharing

Please use your RMIT account to access the document

Here is my written reflection:

Week #3: 

A Reddit user posts his Ai-generated recreation of the famous Mona Lisa in the modern day in the AI-art subreddit and received critical responses from other users. Here is the link to the post:

https://www.reddit.com/r/aiArt/comments/11e3elz/modern_day_mona_lisa/

I think it is an interesting case study to remind us that even the most updated language models are fragmented with prejudices and biases, especially in this example, unrealistic visual depictions of women’s bodies. Also, I find out it is Stable Diffusion, the notorious text-to-image, diffusion learning model. To put it simply, training images will be processed through image encoders to represent the information as vectors (Zhao 2023). The diffusion model combines all this information to find relative associations and utilizes them to create final products (Zhao 2023). However, during my research, I discover that the model was primarily trained on images with English descriptions. This explanation explains why the image of the AI modern Mona Lisa is rifled with irrational visuals, largely due to the unhealthy environment of data input. 

With our group’s media sketch this week, we plan to create a melodrama, music theatre script of the scandalous scene of Chris Rock slapping Will Smith. What does not work well is that the Chat GPT model does not update the event in its system. Therefore, it is unable to make a logical analysis of tension and it leads to the issue that the tension is resolved in an illogical sense. In the end, we are gladly happy with the results and our collaborations. Everyone helps spark the idea together so that all the scenes are linear and meaningful. For future media works of a similar kind, it is recommended that we could teach Chat GPT about what should/should not include. Although we frame boundaries to make sure that the story is not astray, we are thrilled with mesmerizing ideas that Chat GPT can spark.

This week’s reading creates an interesting comparison between large language models like ChatGPT and lossy compression algorithms, in order to identify the limitations and challenges of LLMs (Chiang 2023). There is a key quote that really stood out for me “If students never have to write essays that we have all read before, they will never gain the skills needed to write something that we have never read” (Chiang 2023). He asks about the effectiveness of the AI model, whether that “blurry copy of unoriginal work’ helps improve our writing or not. Everyone has to start by mimicking other writers’ styles before crossing to their original work. Therefore, AI might not foster writing but rather disrupt the necessary process in the quest to become a good writer. 

 

Reference List:

Chiang T (9 February 2023) ‘ChatGPT Is a Blurry JPEG of the Web’, The New Yorker, accessed 6 August 2023. https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web

Zhao G (2023) How Stable Diffusion works, explained for non-technical people, Medium website, accessed 6 August 2023. https://bootcamp.uxdesign.cc/how-stable-diffusion-works-explained-for-non-technical-people-be6aa674fa1d