AI Week Thanksgiving Edition: AI Video
Get your smiler ready, because there are a lot of comics in this week's AI week. But first, meet Tilly Norwood.
Happy (Canadian) Thanksgiving! Get your smiler ready, because there are a lot of comics in this week's AI week. But first, meet Tilly Norwood.
Meet Tilly Norwood
Tilly Norwood, a London actress, was seeking representation earlier this month. Tilly has a website with headshots and a comedy sketch. Tilly Norwood is pissing SAG-AFTRA off because she's not real.
SAG-AFTRA actors union condemns AI avatar Tilly Norwood : NPR
Tilly Norwood, an AI-generated avatar, is being compared by its creators to A-list human actors. SAG-AFTRA and others are pushing back.
Tilly Norwood's AI-generated "comedy" sketch, which you can watch on YouTube is desperately unfunny, verging on outright creepy. Tellingly, it's cobbled together out of sub-10-second AI-generated clips of non-Tilly characters, interspersed with brief clips of the Tilly character facing the camera. At no point does the Tilly character interact with anyone else. Both the brief clips and the Tilly-only-scenes, which may be due to current limitations of AI video generation, make it hard to take "Tilly's" bid for representation seriously. The character now appears to be "represented" by an inhouse "agency" at the European AI production studio that created her.
Penny Arcade on Tilly Norwood
Source: https://www.penny-arcade.com/comic/2025/10/03/enervator
Keep reading for more comics in the comics section below. But first, a bit more about AI video generation... and copyright, AI safety, and the environment.
Sora 2 AI text-to-video model
OpenAI recently launched the Sora 2 text-to-video model with an iOS social app, simultaneously empowering millions of AI-generated video clips and creating massive headaches in the copyright, AI safety, and environmental departments.
OpenAI vs. Copyright (Again)
Sora 2 can generate brief clips and shows big improvements over the original Sora, including background noise and sound effects, multiple visual styles, speech, and much-more-realistic physics, which OpenAI calls "physics understanding". You can also create "cameos", clips featuring yourself or your friends.
Hilariously, OpenAI launched Sora 2 with an "opt-out" for IP violations, as in "Dear IP holders, we will let users create videos featuring your intellectual property unless you proactively opt out."
Obviously, that isn't how copyright works. But that's not the only can of worms the launch opened.
OpenAI vs. Reality
Sam Altman doesn't seem to have opted himself out, because along with AI-generated South Park, Family Guy, Pikachus galore, and other assorted copyright infringement, the app quickly filled up with Sam Altman deepfakes.
OpenAI's new social app is filled with terrifying Sam Altman deepfakes | TechCrunch
OpenAI's Sora app makes it too easy for people to create misleading AI content.
As the article above notes, the app makes it incredibly easy to generate fake videos of people saying things they never said.
A few days and thousands of IP violations after launch, OpenAI walked the copyright infringement back by blocking prompts tied to well-known copyrighted material, such as "South Park". But while you can't generate a South Park episode any more, the AI safety concern remains. For example, I doubt that the app blocks prompts tied to every politician in every jurisdiction. We're well into the era where you can't believe video evidence.
OpenAI vs. the Environment
So, if you have any relatives who seem to be falling for AI slop videos, you might consider creating a demonstration by downloading the Sora app and creating a cameo or two featuring them... but maybe not too many, because even though OpenAI is offering video generation for free, the energy needed to generate a video is orders of magnitude higher than that needed to generate images or text.
OpenAI’s newly launched Sora 2 makes AI’s environmental impact impossible to ignore
As AI shifts from text to video, its appetite for power and water soars, presenting a climate-policy issue.
Generating an image uses the electricity of a microwave running for five seconds, while making a five-second video clip takes up as much as a microwave running for over an hour.
The Sora iOS app soared to #1 in the app store after its launch. Just how much energy did we spend, collectively, generating Pokémon x Family Guy mashups?
Comics Section
I promised you a generous comics section, and here it is!
Next up after the comics section: Refunds for workslop, and an excellent non-AI video about AI slop and the death of the Internet.
1. SMBC
Source: https://www.smbc-comics.com/comic/why-8
2. First Dog on the Moon
Couple of panels from recent FDOTM. First, a plea:
and from a recent reader submission strip:
And a bonus FDOTM about Tilly Norwood: https://www.theguardian.com/commentisfree/picture/2025/oct/01/hollywood-is-in-a-flap-ai-actors-are-trained-on-real-actors-who-will-never-be-paid-for-it. Preview:
3. Alarmingly Bad
Source: https://www.reddit.com/media?url=https%3A%2F%2Fi.redd.it%2Fb45o7ob4f3sf1.jpeg
Refunds for workslop
Top-tier consulting firm Deloitte was caught passing off AI outputs as top-dollar consultant work in a report to the Australian government. The smoking gun? Made-up citations to nonexistent work, as usual. What makes this newsworthy is that Deloitte will issue a partial refund for the partially-correct report.
Deloitte will refund Australian government for AI hallucination-filled report - Ars Technica
Consulting firm quietly admitted to GPT-4o use after fake citations were found in August.
I hope this will set a workslop-discouraging precedent. That would be timely, as Microsoft has just launched "vibe working" by including Copilot in Excel and Word.
Kurzgesagt on AI Slop
One of my favourite YouTube video channels, Kurzgesagt – In a Nutshell, just released a video about the impact of AI slop on the internet. 10 minutes, lovingly hand-animated, well worth your time.
I've noticed the quality of my search results plummeting in the last month. Now I'm seeing results where every link in the first two pages is obvious AI slop. It's getting to the point where I don't even waste my time on anything unless I know the website, or the article is dated 2023 or earlier.
AI Safety: Toxins
The destruction of the open web, annoying though it will be, pales against the threat of AI-designed biological(ish) toxins. To summarize: You can't just order DNA that encodes a toxin or a virus! We've got systems! But, if that toxin is an AI-generated "designer toxin", it could drive right through a blind spot in those systems.
A biological 0-day? Threat-screening tools may miss AI-designed proteins. - Ars Technica
Ordering DNA for AI-designed toxins doesn’t always raise red flags.
That's it for this week's AI week! Thanks for reading. Leave comments on the website below.