Pressure project1-Geometric Starfield

Posted: October 2, 2026 | Author: Wenxi Zhu | Filed under: Uncategorized | Leave a comment »

Brief Description

I used simple geometric shapes such as circles, rectangles, and triangles in TouchDesigner and combined them with movement, feedback, noise, and other visual effects. Once the program starts, the shapes continue to move and change without requiring any interaction. I also added audio to the project so that the movement and visual effects of the shapes develop along with the sound, creating an audiovisual piece that can run and loop on its own.

Process Reflection

My original idea was to create something similar to the game Snake, where a shape would continuously move around the screen and create new changes after touching or “eating” other shapes, such as changing its size, movement, or visual effects. However, once I started working on the project, I realized that creating a complete game system with too many interactions would be difficult to finish and polish within five hours, especially based on what I currently know about TouchDesigner. Because of this, I changed my direction and started with simple geometric shapes, focusing more on their movement and visual relationships with each other.

I used LFOs to create continuous movement and connected different values to Transform parameters to control the position, size, and movement of the shapes. One of the visual elements I created was a small white ball with a trailing effect that moves along an equilateral triangular path. I then duplicated it and reversed the direction of the path. When the two triangular paths overlapped, they created the appearance of a six-pointed star. I also experimented with Feedback, Noise, Level, and Composite to create more visual changes. At first, I considered adding more shapes, interactions, and colors, but I realized that too many elements would not only be difficult to complete within five hours, but could also make the visuals feel too busy. My first version used mostly rainbow colors. Although it looked colorful, the overall result felt messy and even a little cheap. I eventually changed to a simpler color palette so that the movement of the shapes could become the main focus. I also added particle effects in the background to create an atmosphere that feels like a starry sky.

While learning TouchDesigner and working on this project, I was surprised by how much a small change in a parameter could affect the final result. For example, changing the speed of an LFO or adjusting the position, size, and movement range of a shape could completely change the rhythm and feeling of the composition. Combining different nodes also gave me a lot of flexibility to create different effects. Compared with After Effects, which I have used before, I found that I prefer TouchDesigner for creating this kind of looping animation. Its real-time workflow allows me to adjust parameters and immediately see the results, which makes the experimental process more direct.

During the in-class presentation, I felt that most of my classmates watched at least one full cycle of my piece. Although I did not achieve the “make someone laugh” achievement, I still received some positive feedback. The presentation also gave me a chance to see how other people responded to the pacing and visual changes in my work. Overall, this project helped me become more familiar with TouchDesigner and gave me my first experience creating a visual system that can continuously run on its own. In the future, I would like to experiment with more complex systems and audio controls to make the work more dynamic and engaging.

Wenxi_Pressure project1.toe

---

Pressure Project 1 – Music Visualization

Posted: October 2, 2026 | Author: Yiyi Liu | Filed under: Uncategorized | Leave a comment »

Before starting this project, I spent a long time agonizing over what exactly to do. From my perspective—as someone who works very slowly—five hours is a very tight timeframe. So, after spending nearly an hour thinking about what to do, I decided I needed to pick up the pace. I’ve always been fascinated by music visualization, and I wanted to see if I could recreate it in Touch Designer.

My work primarily consists of several interconnected visual and audio systems. To begin with, I created small, constantly moving particles on the screen and used various motion parameters to make them rotate, move, and change size within the space. Rather than having all particles move in exactly the same way, I incorporated continuously varying values so that their movements would not repeat exactly. Next, I adjusted effects such as blur, displacement, and color, allowing the originally simple geometric shapes to gradually evolve into the organic forms seen today.

A key element of this project is the particle trail effect. I used a Feedback system so that the previous frame does not disappear immediately but instead re-enters the next frame. Then, using nodes such as Level, Blur, and Composite, I gradually reduced the intensity of the old image and recombined it with the new particle image. This way, when a particle moves, the position it previously occupied is briefly retained, creating a continuous trail. By adjusting the feedback strength and blur intensity, I can control how long the trail persists. If the feedback is too strong, the entire image gradually builds up; if it’s too weak, the trail disappears too quickly, so I spent quite a bit of time finding the right balance between the two.

The integration of sound and visuals is one of the most important aspects of this work. I first imported the music into TouchDesigner, then used nodes such as Noise and Math to analyze the sound, converting the original music into data that could control visual parameters. As a result, sounds of different frequencies and intensities are no longer simply played back; instead, they directly influence changes in the visuals. I remapped and constrained this audio data so that it could be used to control visual effects, rather than using the raw audio data, which fluctuates too wildly.

One of the most obvious visual responses is the variation in the number of particles. The intensity of the music is analyzed and converted into a numerical value, which is then used to control the number of particles generated on screen. When the music becomes louder, or when the rhythm or certain frequencies become more pronounced, the system generates more small particles; when the sound fades, the number of particles decreases. Consequently, particles do not appear in a fixed quantity but constantly fluctuate in response to the music. Since each newly generated particle enters the feedback loop, when the sound is louder, not only does the number of particles increase, but the motion trails and visual density in the image also become more pronounced.

My approach to creating this project is similar to many generative art projects: I start by building a simple system and then gradually increase its complexity. At first, I focused solely on making basic shapes move; then I added musical responses, followed by variations in particle count, feedback trails, blur, displacement, and color changes. During the production process, I experimented with many different connection methods; some feedback settings caused the visuals to disappear too quickly, while others made them too chaotic. Through constant adjustments, I finally found an effect that preserves the particles’ motion trails without completely obscuring the original shapes.

This project gave me a deeper understanding of the relationship between sound and visuals in TouchDesigner. In the past, I tended to view music and visuals as two separate elements, but this time I tried turning the sound itself into data that controls the visual system. The music determines when the number of particles increases or decreases, while the particles’ movement leaves trails through Feedback, and the colors and other automatic changes allow the system to continuously generate new visuals even without manual intervention. This also transforms the entire piece into a generative visual system that runs on its own while constantly evolving in response to the music.

Link to original work: https://github.com/HexisHere/ACCAD.git

---

Pressure Project 1: Tidal Memory: An Audio-Reactive Particle Sea

Posted: October 2, 2026 | Author: Leyi Wang | Filed under: Uncategorized | Leave a comment »

Brief description: Tidal Memory is a self-generating TouchDesigner patch that turns music into shifting white particle waves. The points rise, curl, leave glowing trails, and appear to reflect in moving water.

Process reflection

I approached the five-hour limit by committing to one visual idea: an ocean made from moving points and lines. I used the Bilibili particle wave TouchDesigner tutorial as a starting point, then concentrated on building a version I could run and document. I first established a long wave shaped particle source, then worked on motion and trails. I kept the last part of the session for connecting audio, shaping the reflection, and recording the result.

The patch scatters points over a stretched spherical form and disturbs it with noise. A feedback loop carries the points forward, while animated noise changes their paths. Trail POP records where the moving points have been, drawing fine strands that collect into a crest. Audio analysis turns changes in the music into control signals, so the result follows a rhythm without repeating exactly the same shape each time.

I chose white particles, cool blue highlights and a black background. This limited palette keeps the delicate lines visible. The moving reflection below the wave gives the abstract structure a sense of place. I left out extra text, scenery, and camera effects: within five hours, those additions would have distracted from the central motion and taken time away from testing the patch.

The biggest surprise was how much the trails changed the image. Individual points looked sparse, but their recent paths made the wave feel dense and almost fabric like. The reflection also made the image read more clearly as water. I had to balance energy against legibility: too little movement felt static, while too much obscured the wave’s form.

My main achievement was completing an audio-reactive system and recording a final output that continues to change while it runs. I cannot claim a formal course badge without checking my course record. With more time, I would add a keyboard control to reset or change the wave and test how it responds to several songs.

How the patch works

How are the particles generated? Sphere POP provides the stretched form; Noise POP disturbs it; Sprinkle POP places about 3,000 points on it.

Are the particles “alive”? They keep moving through Feedback POP, but the patch does not specify an individual birth-and-death cycle.

How is the water reflection made? The lake effect network mirrors the image and distorts it with noise to create ripples.

How are the tails made? Trail POP stores recent point positions; Line Metrics and Lookup Attribute POPs shape the visible strands.

Process

Techniques and Acknowledgments

Particle-wave tutorial: “粒子海浪—TouchDesigner教程｜音乐可视化 Vol.2.01” by 布鲁斯Qing on Bilibili

Beat-detection tutorial: “A More Accurate Beat Detection—TouchDesigner Tutorial” by bileam tschepe (elekktronaut)

Music: “珠玉” by Shan Yichun (单依纯)

Techniques used: Sphere, Noise, Sprinkle, Feedback, and Trail POPs; audio analysis with CHOPs; mirrored water reflection

---

TrajectoryPressure Project 1 – Automatically Generated Trajectories

Posted: October 1, 2026 | Author: Shihua Wang | Filed under: Uncategorized | Leave a comment »

“Trajectory” is a sound and video work created using TouchDesigner. A group of black birds fly around a red bird; their numbers increase and decrease, until they finally disappear. The music influences their movement patterns, while the fading traces on the screen show both past and present positions simultaneously.

Reflection on Creation

When creating “Trajectory”, I didn’t follow a pre-designed visual plan. Instead, I tried different approaches to improve the work step by step. I used a bird model from my own resources, adjusted its structure and flying behavior, and then tried to create repeating animation effects in TouchDesigner. As I worked on this, the first question that arose was: What would happen when these birds gather together?

It wasn’t enough just to have the birds appear on the screen. When they moved together, their movements seemed mechanical. So I tried different speeds of movement, paths, and wing animations to make their behavior more varied. I wanted them to belong to the same system, but also maintain some independence. For me, this approach was more interesting than simply adding more birds.

Over time, an atmosphere began to form. A smaller number of birds creates a sense of openness, but when there are more birds, the same movements become crowded and tense. The fading traces make this effect even more noticeable. Each bird leaves behind traces, and these traces accumulate, making it difficult to distinguish individual birds. That’s why I named this work “Trajectory” – I’m interested not only in the direction of the birds’ movement, but also in the traces they leave behind.

I chose white as the background color because I wanted the sense of tension to come from the birds themselves. White highlights the outlines of the birds and the fading traces. This reminds me of a blank page gradually filled with various ideas. When the birds disappear, the background becomes empty again, but the feeling of emptiness after being crowded is completely different. The red bird in the center becomes the visual focus of these changes. I want it to stand out, but I don’t want the audience to have to interpret what it represents. But at the same time, a bold and fitting image came to mind: the scene from “Odyssey” where the bowing test takes place. It covers all the main themes: Odysseus returns to his hometown to prove himself and kills all the intruders. So I incorporated the sound effects from that scene into the control system as one of the chaotic elements.

During the five hours of creation, I focused on how the birds move, their relationship with music, and the contrast between accumulation and disappearance. I chose to keep the environment simple, rather than creating complex scenes. I also used pre-made flying actions instead of developing more complex animation systems. This allowed me to focus on the overall experience.

Many useful discoveries came from mistakes. The birds flew in the wrong direction, their bodies turned over, and sometimes the screen became dark for a moment. To solve these problems, I had to observe more carefully. I kept thinking “let’s try again”, so I almost forgot to check the time.

Feedback from class helped me explain the process from random attempts to deliberate choices. Before having clear results, I wouldn’t claim to have achieved any specific outcome. I can describe how this project changed my attention: initially, I just checked if it worked properly, but later I started paying attention to the experience during use.

Materials and Thanks
– Software: TouchDesigner.
– Bird models: From my personal resources; skeletal binding and flying actions were created by me.
– Music: “The Trail”, taken from the music in “Odyssey”, obtained from Bilibili.

---

GGEZ: An Audio-Reactive Visualizer

Posted: October 1, 2026 | Author: Isabella Lo | Filed under: Uncategorized | Leave a comment »

Description

For this project, I made a self-generating audio-reactive visualizer in TouchDesigner using the song “GGEZ” by Kaien Nuen, SUGAR (CNCA), and M.Sasuke. The visual includes 3D “GGEZ” text, and organic blob, particles, and different effects that react to the audio. Once the patch is started, it runs on its own and the audio continuously changes different visual parameters.

Process Reflection

When I first started this project, I wanted to make a digital organism that woudl grow, change, and eventually die. Since I was getting comfortable with TouchDesigner, I started looking a tutorials to figure out how I could actually build something like that. I first followed Pao Olea’s Organic Creature tutorial, which helped me learn more about SOPs, CHOPs, instancing, geometry, 3D space, materials, feedback, and bloom. I ended up making a tentacle-like spiral while experimenting with these techniques, and this made me much more comfortable working with 3D geometry.

After that, I become more interested in audio reactivity, so I followed Nicholas P.J.M’s tutorials on audio-reactive particles and 3D text. I learned how to use POPs, Audio Analysis, Math, Transform, Noise, and Random, and then started combining those techniques with what I had learned from the first tutorial.

The five-hour limit definitely affected my process. I oiginally had a pretty specific plan, but ended up exploring two connected directions instead of spending the entire time on one organism. I think this actually helped me learn more becuase I was able to experiment with different parts of TouchDesigner and then bring them together in the final piece.

I originally planned to use “American Girl” by Tiffany Day, but I changed the song to “GGEZ” becuase I thought its recognizable and intentionally cringe quality would get more of a reaction from the class. I also added the 3D “GGEZ” text becuase I thoguht it would be funny to make the song itself part of the visual. Unfortunately, after listening to it for five hours, it definitely got stuck in my head.

One thing that suprised me was how much the different parts of TouchDesigner build on each other. TOPs, SOPs, CHOPs, and POPs initally felt very separate, but it started to understand how they can all work together. I also realized how much a small parameter change can completely hange the final visual.

For the presentation, the patch ran on its own and held the class’s attention for at least a minute. I also got a noticeable reaction from the class. If I had more time, I would turn the “GGEZ” text into a full lyric visualizer and return to my original organism idea to create a clearer growth adn death cycle.

Documentation

Answering Questions

What is causing the audio reactivity, and what is the input?

The main input is the song itself. I use an Audio File In CHOP to bring the song into TouchDesigner, which then goes into Audio Analysis. From there, I use Math and Lag to adjust and smooth the values before sending them into different parameters. Different parts of the audio control different parts of the visual, including the galaxy shape and the “GGEZ” text. This is what allows the visual to keep changing as the song plays.

How are the particles moving?

I used POP operators to create and control the particles. I used Math and Transform nodes along with Noise and Random to change their movement and make it less predictable. Originally, the particles looked more like little boxes, so I used a Trail to make their movement more continuous. This made them look more like particles flowing through the space instead of separate objects moving around.

How did you layer everything?

I started with the different visual elements separately and then combined them through different TOPs and a Geometry component. The galaxy image was converted into a 3D object, which gave me the main organic shape in the piece. I also created the “GGEZ” text as a 3D element and rendered it back into a TOP so I could keep manipulating it. From there, I layered things like Noise, Slope, Displace, Feedback, color effects, and Bloom to create the final look. The feedback helped create trails and build on previous frames, which made the visual feel more constantly in motion.

What is the organic shape, and how did I arrive at it?

The organic-looking shape is actually an image of a galaxy that I converted into a 3D object. I originally started the project wanting to make a digital organism that would grow and eventually die. I started by following Pao Olea’s Organic Creature tutorial, which taught me more about geometry, instancing, SOPs, CHOPs, and 3D space. I then experimented with a tentacle-like spiral before moving into the audio-reactive particle tutorial. While experimenting, I realized I could take an image and turn it into a 3D form, which is how I ended up with the galaxy shape in the final piece.

Where did I start, and how did I get here?

I definitely did not start with the final idea. I started with the organism concept, then went through a tentacle/geometry exploration, and eventually became more interested in audio-reactive visuals. I used a few tutorials to learn specific techniques and then combined parts of them into my own patch. The five-hour limit meant I had to experiment while I was learning, so the final piece became a combination of several things I learned rather than one idea that I had completely planned from the beginning.

I also originally planned to use “American Girl” by Tiffany Day, but switched to “GGEZ” because I thought the song would get more of a reaction from the class. I wanted the giant “GGEZ” text to make the piece a little funny instead of just being a serious audio visualizer.

What would I do with more time?

If I had more time, I would turn the “GGEZ” text into a full lyric visualizer instead of having the same word throughout the piece. I would also like to return to my original organism idea and make the galaxy/organic shape actually grow, change, and eventually break apart or die. I think that would make the piece feel more like the self-generating organism I originally imagined.

---

AI Expert Audit – Evaluating AI-Generated Multimodal Material

Posted: September 10, 2026 | Author: Leyi Wang | Filed under: Uncategorized | Leave a comment »

A. Source Material

For this assignment, I used my essay, “Early Chinese Dumplings (Wonton) from Qimin Yaoshu,” which I wrote last semester for the final project in a food history general education course. The project required us to recreate a historical food using a recipe from historical sources. I made a simplified reconstruction of early Chinese wontons and reflected on the cooking process and the differences between historical and modern ingredients and tools.

Because I researched the topic, prepared the food, and wrote the essay myself, I am familiar with both its content and the experience it describes. This allows me to evaluate whether the AI accurately represents my work. The project is meaningful to me because it connected a familiar food with its history through hands-on experience.

B. The AI-Generated Materials

AI-Generated Podcast

https://file.garden/ap8TDC2-D925c-lY/The_1500_year_old_survival_dumpling.mp3

AI-Generated Mind

AI-Generated Infographic

C. My Expert Audit

Accuracy Check

Overall, I did not notice any obvious factual errors in the AI generated materials. The podcast explained the basic information about wontons clearly. The mind map was easy to follow and I did not notice anything important missing. The infographic was also well organized and covered the basic content of my essay.The main issue was the podcast’s emphasis. It repeated simple information and spent more time explaining ordinary details than I thought necessary. One specific example is the infographic’s instruction to let the dough “rest for 20 minutes.” This matches what I did in my cooking project, but my essay does not establish that this exact timing came from the historical recipe. By placing this instruction under the title “The 6th-Century Wonton,” the infographic could lead a reader to mistake my modern reconstruction choices for documented historical instructions. The detail itself is accurate to my experience, but its presentation makes the historical connection seem more certain than my essay supports.

My essay discusses a familiar food through my own cooking experience, but the podcast sometimes made the topic sound more complicated or remarkable than it felt to me. This did not necessarily change the facts, but it changed how the material came across. Someone unfamiliar with my essay might not notice that difference in tone.

Usefulness for Learning

If I were learning about this topic for the first time, I think these resources would help. The podcast explained the basics clearly, which could benefit someone unfamiliar with wontons. However, its thoroughness also made it feel unnecessarily   long. At times, it sounded like two people reading an essay aloud rather than having a natural conversation. They repeated straightforward points that I would not expect a typical podcast to spend so much time discussing.

The mind map worked well because its structure was clear and the content felt complete. It made the information easy to follow without requiring a long explanation. The infographic also presented the basic content in an orderly way and was useful for a quick overview.

Based on my experience, the mind map was the most effective format because it was clear and direct. The podcast was the least effective for me because of its repetition, although its detailed explanations might be more useful to a beginner. However, the structure of my essay is fairly simple, so it may not be representative of more complex texts. However, in my previous experience, AI has been especially helpful for breaking down texts with complex arguments and making their logic clear at a glance.

The Aesthetic of AI

The most noticeable “AI weirdness” was the podcast hosts’ exaggerated emotions such as surprise or questions. My original essay is about a fairly normal cooking experience. However, the hosts sometimes reacted as though the information was surprising or extraordinary. These reactions seemed intended to imitate human enthusiasm, but they felt out of place. Their small talk and exchanges were also slightly stiff.

The podcast had an enthusiastic, explanatory voice, but the enthusiasm did not always match the subject. It seemed to prioritize keeping the conversation lively and explaining everything thoroughly. As the author, I would have preferred less repetition and more attention to the personal experience behind the essay.

The infographic looked organized, but its uniform font styling gave it a somewhat formulaic appearance that I associate with AI generated posters. This is a personal impression rather than proof of AI authorship, since human designers can also use a consistent font.

Trust & Limitations

My experience with these materials connects to how my use of AI has changed. I started using AI to help write essays in high school because English is my second language. At that time, organizing my ideas into a complete English essay felt extremely difficult. As my English improved, I began noticing recurring structures and expressions in AI-generated writing. The writing could be fluent, but it often failed to express my feelings accurately.

Now, I mainly use AI to suggest a broad outline or help identify the main points of an assignment. I write the important ideas and emotional content myself. This gives me support while keeping the writing closer to what I actually want to say.

---

AI Expert Audit: LeBron vs Michael Jordan

Posted: September 9, 2026 | Author: Benjamin FitzGerald | Filed under: Uncategorized | Leave a comment »

A. Your Source Material

• What document(s) did you use?
• Why are you an expert in this content?
• What makes this material important or meaningful to you?

I utilized two documents that I felt would give the AI a well-rounded and robust overview of the topic. I first gave the AI an article that simply showcases side by side statistical comparisons. This covered the objective and mathematical portion of the debate. I then gave AI a link to the most popular reddit thread on this topic, so it can have access to the more subjective side where oddly specific details were discussed to a greater extent. Every possible argument people made, the AI had access to now. What makes me an expert in this content is the fact that I’ve been talking about it and posing the question for years now even writing a paper on it in high school which I eventually switched stances from. I’ve done the research and due diligence. This material is meaningful to me because the game of basketball happens to be my favorite sport and I believe it is human nature to want to recognize the absolute pinnacle of any sporting event or endeavor. At least it’s in my nature. 

B. The AI-Generated Materials

• Embed or link to the podcast, mind map, and infographic NotebookLM created
• (You can screenshot the mind map and infographic, or describe them if embedding isn’t possible)
• MindMap
  • 
• Podcast  
• Report 
• Claude Infographic

C. Your Expert Audit

Respond to ALL of the following:

1. Accuracy Check

• What did the AI get right?
  • The AI got a lot right in this analysis. It correctly noted that not all opponents are created equal, and that a ring isn’t just a ring, since MJ never had to face a team like the 73-9 Warriors that LeBron did. It also recognized that the modern NBA is a fundamentally different and harder game, with greater talent depth and ball knowledge across the league. The AI did a solid job tracing LeBron’s different team phases as he moved around, compared to Jordan’s primarily Bulls-centric career, and it accurately pointed out that MJ never lost a Finals series when he was favored, unlike LeBron’s losses in critical series like the one against the Mavericks. It also picked up on how offensive and defensive strategies have evolved, along with physical factors like court conditions. One of its stronger points was recognizing that you can’t just teleport a player into a different era without accounting for the transfer of era-specific knowledge and resources, showing genuinely good era recognition. It also acknowledged that differing definitions of greatness mean this debate will likely never be fully settled, and overall it compared accolades between the two players well. 
• What did it get wrong, oversimplify, or miss entirely?
  • On the other hand, there were some notable shortcomings. The AI oversimplified LeBron’s longevity by framing it as an unanswerable marathon-versus-sprint comparison, when in reality his extra seasons should have been treated as a real, substantive point of comparison rather than dismissed. It also spent too much time diving into the players’ business footprints and finances, effectively defining their value too heavily around how good a financial asset each of them is rather than their on-court impact. It missed their physical attributes entirely, failing to mention that LeBron is the bigger player while MJ had the higher vertical. And it leaned too much on points per game without giving enough attention to shooting percentages, which left an important efficiency dimension of the comparison underexplored. 
• Were there any subtle distortions or misrepresentations that a non-expert might not catch?
  • Not that I noted, what the AI spoke on seemed to be accurate. 

2. Usefulness for Learning

• If you were encountering this material for the first time, would these AI-generated resources help you understand it?
• What do the podcast, mind map, and infographic each do well (or poorly) as learning tools?
• Which format was most/least effective? Why?

If I were encountering this material for the first time, the AI-generated resources would definitely help me understand it. The podcast did really well at exploring specific arguments like era-specific differences, but spent a lot of time on a few when I think spending less time on more arguments would’ve been more all encompassing. The mind map laid out all of the data in a really neat and organized way only coming short when it came to oversimplifications. For example in the cumulative totals nodes, only three stat lines are presented. The report was an all around great document of information that dove the deepest into statistical comparisons. 

3. The Aesthetic of AI

• What “AI weirdness” did you notice? (Strange phrasings, generic language, odd emphases, uncanny tone, oversimplifications, etc.)
• Did the AI have a particular “voice” or style? How would you describe it?
• What does the AI prioritize or emphasize that a human expert might not?

Some AI weirdness I noticed was at times in the podcast it almost seemed to crash and freeze for a few seconds. Leaving us with a creepy long stutter. The AI didn’t really have a particular voice or style because the three different sources it generated all came at the debate from different angles. I don’t think in the context of the topic I chose the AI emphasized a specific point that I couldn’t see a human expert emphasizing. Although it was partially based on a reddit thread so it’s almost echoing human thought.

4. Trust & Limitations

• Based on this experiment, what would you warn someone about trusting AI-generated educational materials?
• Where does AI seem most reliable? Where is it most likely to mislead?

I would say to use AI as a general overview offering the comfort to trust the salient points. It seems the most reliable in covering the big arguments and doing so in a fully cohesive way. Where I would caution someone is about the finer details. It seems AI can oversimply or overlook them in an effort for a more efficient conclusion/explanation. In my experience though AI didn’t necessarily state any incorrect information. 

---

#2 AI Expert Audit: Attack On Titan

Posted: September 9, 2026 | Author: Brody Hall | Filed under: Uncategorized | Leave a comment »

Briefing Doc: Here
Character Mind Map: Here
Podcast: Here

For my investigation on NotebookLM, I decided to do an overview of both the Attack on Titan the manga, and the anime. I imputed the sources of both the wikis for the different mediums, and a few other articles on bigger episodes of the series. I chose Attack on Titan because it is a piece of media that I have obsessed over for awhile now, due to its intense world building and focus on deep storytelling between characters, and an overarching story. Now after inputting these sources, I found that AI had a good grasp on what the story is, but isn’t able to know finer details without me letting it learn other sources. I want to go over some of the parts that AI got right, and the parts that AI got wrong.

The AI had a good grasp on what the general idea of the story was, along with the characters and what they represent. But, it had mistakes on certain things like pronunciation, character connections, and small, yet key parts of the story. It oversimplifies a lot of the parts of the story, leaving out what really makes the show intense and intricate. For example, I generated the mind map to create connections between characters, yet it just divided them into sub-group, diminishing what the depth of character connection there is in the story. 

I believe that if I didn’t know much about the story, it would be a good way for me to get a basic understanding of it. I found the podcast to be the most useful, and entertaining way of learning the material because it feels more engaging and goes further in depth. The mind map didn’t really do much, and the briefing doc was informative, but not engaging. The idea of it being AI made me feel weird in itself, especially with how more human AI is sounding, with expressions and banter. I told the AI to make the voices engaging and funny, which made the podcast feel more enjoyable. With these human emotions though, it wasn’t felt with the words they said always, a lot of it sounded scripted, along with occasional voice glitches that took me out of it.

After going through this research, I found that it is a new possible way to learn about more specific topics that there might not be many resources for. Even then, I wouldn’t use it as something to rely on. You have to make sure your sources you’re uploading first are good sources, and even after that, it can focus on the wrong information if you don’t prompt it correctly. If they were able to find a way for it to take up less resources, along with more accuracy, I could find myself using something like this. The medium that led me to feel this way is the podcast option. It was very engaging when it wasn’t glitching out, and it dove deep into the storytelling and world building that I asked it to, and the information was mostly accurate. I think it has the possibility of being a big learning medium compared to things like ChatGPT that just spew information back out at you.

---