Shells

Week Eight: Shells — Perceptions

[ ENG 6806 // FALL 2026 // WEEK OF OCTOBER 12, 2026 ]

Still from an AI-generated political ad: Rep. Greg Sharpe beside a fake Dr. Anthony Fauci holding a giant syringe at a state fair

AI-generated ad with a fake Dr. Anthony Fauci (AdImpact); see Ridout, Fowler, and Franz, “We’ve Tracked $80M of Spending on Almost 170 AI-Generated Political Ads This Year – It’s Weird,” The Conversation, Oct. 5, 2026

AI-generated sandwich-board flyer for a Japanese sushi taco with a cartoon cat in a headband

AI-generated party flyer: Cobb County vs. Paulding County at Mercedes-Benz Stadium, with two Jokers holding cards

Koebler, “These Are the Worst ChatGPT Flyers You’ve Sent Us,” 404 Media, July 13, 2026

404 Media headline: Pinterest Is Drowning in a Sea of AI Slop and Auto-Moderation. Users are exhausted fighting AI moderation, AI-generated art, and AI-first features.

Gault, “Pinterest Is Drowning in a Sea of AI Slop and Auto-Moderation,” 404 Media, Feb. 19, 2026

Artists tell 404 Media their hand-drawn work is auto-labeled “AI modified” on Pinterest, including artwork 10 to 13 years old, with only a long appeal process to remove the label

Gault, “Pinterest Is Drowning in a Sea of AI Slop and Auto-Moderation,” 404 Media, Feb. 19, 2026

A Pinterest pin of an AI-generated “embroidered” skull with flowers, surrounded by recommended pins of similar AI embroidery images and a Shop the Pin carousel

Pinterest example: AI-generated “embroidery” and its recommendations

Gizmodo headline: Meta’s Muse Is Collecting Information on Everyone in Your Life. Meta’s AI agent creates profiles of people in users’ lives even if they don’t use Muse.

Gil, “Meta’s Muse Is Collecting Information on Everyone in Your Life,” Gizmodo, Oct. 5, 2026

Bluesky thread: rev. howard arson replies “this literally is not true” to John Scalzi’s claim that generative AI could have been built on public domain or licensed works, and to Anil Dash’s reply that many models have been built consensually

Bluesky, Oct. 5, 2026: rev. howard arson (@theophite.bsky.social), quoting John Scalzi (@scalzi.com) and Anil Dash (@anildash.com)

Bluesky thread: Ted Underwood argues it is okay to learn from and train on everything, and that prohibitions on copying were a late, narrowly tailored invention

Bluesky, Oct. 5, 2026: Ted Underwood (@tedunderwood.com)

Bluesky thread: boneless skinless points to Common Pile’s Comma v0.1-2T as a model trained on public domain and licensed data; Greg replies that CC-BY works require attribution the model does not give

Bluesky, Oct. 5, 2026: boneless skinless (@destroy.computer) and Greg (@gregmorenz.bsky.social)

The Pelican Benchmark Begins (October 25, 2024)

Willison introduces his benchmark, “Generate an SVG of a pelican riding a bicycle,” with Claude 3.5 Sonnet’s two sparse line drawings

Willison, Simon Willison’s Weblog, “Pelicans on a bicycle,” Oct. 25, 2024

GLM-4.5, Open Weight from China (July 28, 2025)

GLM-4.5’s pelican: a white bird on a red bicycle with its wings on the handlebars

Willison, Simon Willison’s Weblog, “GLM-4.5: Reasoning, Coding, and Agentic Abililties,” July 28, 2025

Claude Opus 4.5 (November 24, 2025)

“I’m not saying the new model isn’t an improvement on Sonnet 4.5—but I can’t say with confidence that the challenges I posed it were able to identify a meaningful difference in capabilities between the two.
This represents a growing problem for me. My favorite moments in AI are when a new model gives me the ability to do something that simply wasn’t possible before. In the past these have felt a lot more obvious, but today it’s often very difficult to find concrete examples that differentiate the new generation of models from their predecessors.”