Latest / Mini Scroll / MINI TECH SCROLL: YouTube AI-enhancement drama, AI environmental footprint revealed + Ironmouse named as top creator

Transcript

MINI TECH SCROLL: YouTube AI-enhancement drama, AI environmental footprint revealed + Ironmouse named as top creator · Mini Scroll

• 0:00
  This episode of Mini Tech Scroll is brought to you by Blackbird,
• 0:03
  a venture capital firm investing in Aussie and Kiwi startups.
• 0:06
  From the very beginning, Blackbird invests in all kind of
• 0:10
  technology. So if you've got an idea and
• 0:12
  want to take it to the next level, check out Blackbird's
• 0:14
  Giants program. It's a free eight week program
• 0:17
  offering one-on-one mentoring with top founders, operators and
• 0:20
  investors. Applications are currently open
• 0:23
  and close September 7th. So if you're looking for support
• 0:26
  and mentoring, check out the link in our show notes and get
• 0:29
  your application in. Hey everyone, and welcome to
• 0:31
  Mini Scroll, the daily Internet culture podcast bring you the
• 0:34
  biggest stories from social media, the creator economy and
• 0:37
  the digital space every Monday through Thursday and sometimes
• 0:40
  Friday. My name is Annabelle and I'm
• 0:42
  your Wednesday mini tech scroll host and also the deputy editor
• 0:46
  of the youth digital media brand Centennial World.
• 0:49
  Welcome back to mini tech scroll.
• 0:51
  I hope you guys have had a good week so far and are enjoying the
• 0:54
  new filter on the mic. If you're watching, it's a cute
• 0:57
  little pop of color. I don't really have too much to
• 1:00
  say at the top of this week's episode other than Please
• 1:02
  remember to rate, review and subscribe and feel free to check
• 1:05
  out the Buy me a coffee link in the description if you're so
• 1:08
  inclined. But yeah, let's just jump right
• 1:10
  into the stories because I feel like today is going to be a bit
• 1:13
  of a long one. The first story we need to talk
• 1:15
  about is how YouTube has reportedly made enhancements to
• 1:19
  uploaded videos without telling creators.
• 1:22
  According to reporting from BBC and The Atlantic, YouTube users
• 1:25
  have noticed a couple of changes in various videos.
• 1:29
  They have been subtle, but definitely noticeable.
• 1:32
  As Alex Reisner explained for The Atlantic quote, viewers have
• 1:35
  noticed extra punchy shadows, weirdly sharp edges, and a
• 1:38
  smooth that look to footage that makes it look like plastic.
• 1:42
  Many people have come to the same conclusion.
• 1:44
  YouTube is using AI to tweak videos on its platform without
• 1:48
  creator's knowledge. So a guitarist and creator named
• 1:51
  Rhett Shawl also posted a video about this phenomenon around 2
• 1:55
  weeks ago and it's currently sitting at almost 800 K views at
• 1:58
  the time of recording, so it is clearly piqued everyone's
• 2:01
  interest in the description of the video he wrote.
• 2:04
  Quote. I recently noticed my videos
• 2:06
  look strange and smeary on YouTube compared to Instagram,
• 2:09
  almost like a cheap deep fake. In this video I investigate
• 2:12
  what's going on and why I believe it to be a massive
• 2:15
  problem for everyone on this platform.
• 2:17
  And that's exactly what he did. There were a couple sections in
• 2:20
  the YouTube videos, but in one of them he spoke to his wife who
• 2:24
  also happens to be a professional photographer, and
• 2:26
  she agreed that many of his YouTube short videos looked
• 2:30
  weirdly smoothed out and kind of had the qualities of an AI
• 2:33
  generated image or video. He also pointed to a Reddit
• 2:37
  thread that was posted two months ago that had the title
• 2:40
  quote, YouTube shorts are almost certainly being AI upscaled.
• 2:44
  And he ended up concluding the video expressing how concerned
• 2:48
  he is about how his audience would receive these videos
• 2:51
  thinking that they are AI generated.
• 2:53
  He said quote. I think it's going to lead some
• 2:55
  people to think that I'm using AI to create my videos or that
• 2:59
  it's being deepfaked or that I'm cutting corners somehow.
• 3:01
  It will inevitably erode viewers trust in my content.
• 3:04
  With all this speculation going around.
• 3:06
  YouTube has since gone on to issue a statement on X on August
• 3:09
  21st, YouTube's creator liaison Renee Ritchie clarified that
• 3:13
  quote. No upscaling was happening.
• 3:16
  As he wrote, we're running an experiment on select YouTube
• 3:18
  Shorts that uses traditional machine learning technology to
• 3:22
  unblur, denoise, and improve clarity in videos during
• 3:25
  processing, similar to what a modern smartphone does when you
• 3:28
  record a video. So even though Renee Ritchie
• 3:31
  said that AI wasn't being used, he did cut some backlash from EX
• 3:36
  users asking how YouTube could say that it wasn't using AI
• 3:40
  while simultaneously admitting to using machine learning to
• 3:44
  improve videos. In a follow up tweet, Renee
• 3:46
  Ritchie went on to add some additional context to the
• 3:50
  terminology he used. As he wrote quote, Gen.
• 3:53
  AI typically refers to technologies like Transformers
• 3:56
  and large language models, which are relatively new.
• 3:59
  Upscaling typically refers to taking one resolution and making
• 4:02
  it look good at a higher resolution.
• 4:04
  This isn't using Gen. AI or doing any upscaling.
• 4:08
  So I thought this back and forth was quite interesting, not
• 4:11
  necessarily because of what YouTube is doing, but because
• 4:14
  it's prompted a conversation around whether users actually
• 4:17
  know what is and what isn't AI. Because AI at the end of the day
• 4:21
  has become such a catch all term online.
• 4:23
  In this case, it seems like there's been some confusion
• 4:26
  between AI and machine learning. For context, AI is obviously a
• 4:31
  very broad time, but it refers to using technologies to build
• 4:34
  machines and computers that kind of mimic human intelligence,
• 4:38
  whereas machine learning is an application of AI that allows a
• 4:42
  machine or system to learn and improve from experience.
• 4:46
  Obviously the two are really closely related, so the
• 4:49
  confusion is definitely understandable and valid, but it
• 4:52
  doesn't help that so many companies are adopting AI to
• 4:55
  feed into the buzz, which makes it even harder to differentiate
• 4:59
  the two for people who aren't necessarily familiar with this
• 5:01
  kind of stuff. And you could argue that this is
• 5:04
  all a matter of semantics, but I think Rhetchol's point about AI
• 5:09
  eroding trust in viewers definitely holds weight here.
• 5:12
  Reaction to AI on social media are so often pretty intense, and
• 5:17
  when people start to assume that everything is AI generated or is
• 5:20
  using AI, it undermines the experience, especially when that
• 5:24
  isn't the case. So I definitely think it is
• 5:26
  important for journalists, experts, influencers to make an
• 5:31
  effort to explain to their audiences about the differences
• 5:35
  between AI and machine learning and all stuff like that.
• 5:38
  OK, the next story I wanted to talk about is honestly quite an
• 5:42
  extensive 1 and it's all about how Google recently released a
• 5:45
  study analyzing the the environmental impact of AI.
• 5:49
  So today we are becoming women in STEM for real.
• 5:53
  To preface though, I don't want to pretend that I am some kind
• 5:56
  of expert in this field or kind of analysis, but I definitely
• 6:00
  have done my very best to provide a high level breakdown
• 6:03
  for you guys, especially because I think the environmental
• 6:06
  impacts of AI is definitely something we should all be very
• 6:09
  conscious of when we choose to use AI or when we don't.
• 6:13
  On August 22nd, the Google Cloud blog posted a piece with the
• 6:17
  headline How much energy does Google's AI use?
• 6:20
  We did the math where they reportedly measured the
• 6:23
  environmental impact of AI inference.
• 6:26
  Inference in this context though, is basically running an
• 6:29
  already trained model to generate an output, so that
• 6:32
  doesn't necessarily include training, and that is definitely
• 6:35
  something to note. Anyway.
• 6:36
  In the blog they link to a technical paper that explains
• 6:39
  how they measured the energy emissions and water impact of
• 6:43
  Gemini prompts. One of the key findings that has
• 6:46
  start up quite a lot of conversation online was that a
• 6:49
  typical Gemini text query uses 0.24 Watt hours of energy, emits
• 6:55
  0.03 grams of carbon dioxide equivalent and consumes 0.26
• 7:00
  milliliters or around 5 drops of water.
• 7:04
  They also estimated that 12 months ago the per query energy
• 7:08
  use was around 33 times higher than it is now.
• 7:12
  For this episode, I wanted to hone in on the water aspect
• 7:15
  here, given that there have definitely been concerns for a
• 7:18
  long time about a is quote UN quote excessive water
• 7:21
  consumption. For example, a piece that I
• 7:24
  found that was published in The Conversation last year noted how
• 7:27
  AI uses large volumes of water related to hardware production,
• 7:31
  where this usually involves mining for raw materials that
• 7:34
  ultimately contribute to water pollution.
• 7:37
  The authors also added quote initial research shows that AI
• 7:41
  has a significant water footprint.
• 7:43
  It uses water both for cooling the servers that power its
• 7:46
  computations and for producing the energy it consumes.
• 7:49
  On top of this, there's also been research that has been
• 7:52
  published that indicate that a lot of water is being used in
• 7:56
  data centers and the training phases of AI models.
• 8:00
  One study I found while researching for this episode
• 8:02
  that was actually published back in 2023 found that quote the
• 8:06
  water footprint of AI models can vary widely depending on the
• 8:09
  complexity of the model and the size of the data center required
• 8:12
  to run it. For example, GPC Three, an AI
• 8:15
  model developed by Open AI, reportedly consumed
• 8:18
  approximately 700,000 litres of water during its training phase.
• 8:22
  This is a staggering amount of water, particularly when one
• 8:25
  considers that GPT 3 is only one AI model among many.
• 8:29
  So that's just a little bit of context about what is being said
• 8:32
  and what is believed about a eyes water usage.
• 8:35
  Obviously it's fair to say that the amount of water used depends
• 8:38
  on the model and the company, and most of these studies have
• 8:41
  used ChatGPT as an example as opposed to Google Gemini.
• 8:45
  But the pre-existing public narrative and assumptions stand
• 8:48
  in stark contrast to what Google has published in this recent
• 8:51
  report slash study. Obviously this could be a result
• 8:54
  of technological improvements and efficiency as AI becomes
• 8:58
  more advanced, but experts are not convinced that this is
• 9:02
  entirely the case, so I wanted to break that down for you.
• 9:05
  According to reporting for The Verge, journalist Justine Calmer
• 9:09
  notes that the study quote omits some key data.
• 9:12
  Speaking to experts, Comma notes that the study's decision to
• 9:15
  exclude quote indirect water usage is a problem here.
• 9:19
  So from my understanding, water usage should include direct
• 9:22
  consumption, which is water that could be used for cooling the
• 9:25
  systems, as well as indirect consumption, which would be
• 9:28
  water used in electricity production, say, as she writes
• 9:32
  quote. A big issue experts flag is that
• 9:35
  Google emits indirect water use in its estimates.
• 9:38
  Its study included water that data centers use in cooling
• 9:41
  systems to keep servers from overheating.
• 9:44
  Now, attention is shifting to how much more electricity data
• 9:48
  centers might need to accommodate new AI models.
• 9:51
  In fact, a majority of the water a data center consumes stems
• 9:55
  from its electricity use, which Google overlooks in this study.
• 9:59
  Another article I read from the MIT Technology Review noted that
• 10:03
  the Google report was also strictly limited to text
• 10:06
  prompts, so it doesn't necessarily represent the amount
• 10:09
  of electricity or water needed to generate an image or video,
• 10:13
  which would presumably be higher.
• 10:15
  But both of the articles from The Verge and the MIT Technology
• 10:19
  Review note other factors about the report that should
• 10:21
  definitely be noted when you're reading it.
• 10:24
  But we don't have time to get into all of this now, especially
• 10:28
  if we want this podcast to stay under 15 minutes long.
• 10:31
  But I would definitely recommend giving those articles a read if
• 10:34
  you're interested in this topic. OK, so with all of that, let's
• 10:38
  take a moment to think of this from a bigger picture
• 10:41
  perspective. There are a lot of positives, of
• 10:44
  course, to Google releasing this report.
• 10:46
  This is definitely the most transparent that a company has
• 10:48
  been about the environmental impacts of their AI, even as
• 10:53
  third party organizations and researchers try to examine this
• 10:57
  themselves. There is definitely a veil of
• 10:59
  secrecy that kind of exists around a company's AI.
• 11:03
  So it's useful for data to be coming from the company
• 11:05
  themselves. But I will say I did see one
• 11:08
  post on Blue Sky that was skeptical of this and I thought
• 11:12
  it was a valid point. So as they wrote quote, letting
• 11:15
  Google set the narrative about environmental costs of AI is
• 11:18
  like letting tobacco companies set the narrative about the
• 11:21
  effects of second hand smoke by measuring the health effects of
• 11:24
  sitting through one neighbor's smoking cigarette.
• 11:26
  I definitely think there is some truth to this, and I also think
• 11:30
  it is healthy to approach these kind of things with some level
• 11:33
  of skepticism. Also, the fact that the report
• 11:35
  was quite transparent with its methodology despite some of the
• 11:38
  issues that experts have laid out helps contextualize how AI
• 11:42
  factors into our personal carbon footprint.
• 11:44
  I saw an interesting article from the Sustainability by the
• 11:48
  Numbers sub stack about all of this.
• 11:50
  As the author Hannah Ritchie explains quote, individual usage
• 11:54
  of ChatGPT and other LLMS for most people is a small part of
• 11:58
  their carbon and energy footprint.
• 12:00
  Although the fact that AI chat bots are a small part of most
• 12:03
  individual footprints does not mean I don't think AI and data
• 12:06
  centers as a whole are not a problem for energy use and
• 12:09
  carbon emissions. I would also recommend giving
• 12:12
  that post a read as well. It was really interesting.
• 12:15
  She even drew comparisons of the energy used to generate one text
• 12:19
  post would be like compared to watching TV or microwaving food.
• 12:23
  So I just thought it was interesting to put it into
• 12:25
  perspective because it can be quite difficult to conceptualize
• 12:29
  all this for sure. All in all though, I definitely
• 12:31
  think this kind of reporting is something we should be expecting
• 12:34
  and demanding from big tech. It's something that I think uses
• 12:38
  are owed since we are the ones using various AI products,
• 12:43
  especially if you are environmentally conscious or
• 12:45
  this is something you care about.
• 12:47
  We deserve to have some answers, I think, and in an era where
• 12:50
  everything is changing so rapidly, it might just bring a
• 12:53
  little bit of comfort. OK, my last story for today is a
• 12:56
  little bit of a quick 1. I feel like the first two
• 12:58
  stories today were extensive, or at least they took me a while to
• 13:01
  research, so it felt extensive. But I wanted to quickly mention
• 13:05
  Rolling Stones 25 Most influential creators of 2025 and
• 13:09
  how AV Tuber was actually in the list.
• 13:11
  Lauren will actually speak more about the drama that has ensued
• 13:15
  since this article came out tomorrow, so definitely stay
• 13:18
  tuned for that. But if you aren't familiar with
• 13:20
  what a V tuber or virtual YouTuber is, I will explain
• 13:24
  quickly. So these are essentially virtual
• 13:26
  content creators who are using avatars to produce content and
• 13:31
  entertain their audiences. They have been around for a
• 13:33
  while now and are popular in certain niches online, but I
• 13:37
  think with this we are seeing them slowly enter the mainstream
• 13:41
  or at least Internet culture mainstream.
• 13:43
  Iron Mouse, who is a Puerto Rican Twitch creator, was listed
• 13:46
  as #17 in Rolling Stones list Quote.
• 13:49
  Puerto Rican Twitch creator Iron Mouse may remain anonymous, but
• 13:53
  she's become iconic for her pink and purple head avatar and high
• 13:56
  pitched squeaking voice alone. The vlogger and gamer is a
• 13:59
  mainstay in V Tubing and the most subscribed female Twitch
• 14:02
  streamer of all time. This isn't actually the first
• 14:05
  time that Iron Mouse has entered the mainstream recently.
• 14:09
  Around a month ago, Iron Mouse was making headlines after
• 14:12
  releasing a video alleging that V tuber talent agency V Shojo
• 14:16
  was withholding money in funds that were meant to go to
• 14:19
  charity. In that video, she also
• 14:20
  announced that she was leaving V Shojo.
• 14:23
  This was massive news for the V Tubing community at the time and
• 14:27
  also probably helped this community go a little bit more
• 14:30
  mainstream in terms of Internet culture because that's kind of
• 14:33
  what drama does. Everyone loves to know what is
• 14:35
  going on in various niches, even if you aren't deeply invested in
• 14:39
  it. Or at least that might just be
• 14:40
  me. But it's definitely interesting
• 14:43
  to see this niche become increasingly well known across
• 14:46
  Internet culture. I mean, I was listening to the
• 14:48
  There Are No Girls on the Internet podcast the other day,
• 14:51
  which is an iheart podcast, and they were talking about V Tubers
• 14:54
  selling out concerts. And I haven't really seen V
• 14:57
  tuber discourse in so many more traditional publications or
• 15:01
  mainstream publications. But I'm curious, for any of you
• 15:04
  who have long watched V Tubers, are you feeling this shift
• 15:08
  towards the mainstream at all? And if you are, I'm curious what
• 15:13
  you think about it. Is this something that you think
• 15:15
  will last? Is this something that you're
• 15:17
  enjoying? You're happy to see your
• 15:18
  favorite creators validated? I'm very curious.
• 15:21
  OK, That was everything from mini tech scroll today.
• 15:25
  I feel like we had quite a variety of stories.
• 15:27
  Obviously, a lot of it was AI because that is everything we
• 15:30
  seem to talk about these days. But I will be back with another
• 15:34
  episode next week and Lauren will speak to you for a regular
• 15:38
  mini scroll tomorrow. Thanks for listening.
• 15:40
  Bye.