2026-05-20: 5 signals to keep.

A daily selection from the Strange Future Lab archive, rewritten as compact briefs with source trails.

#1 9.0 Weirdness

College graduates booing AI-praising commencement speakers (Eric Schmidt at U Arizona, Gloria Caulfield at UCF, others).

Graduates just booed their own future at commencement — and the speakers had no idea what hit them.

Concrete cultural artifact (boos disrupting sacred graduation rituals) showing AI hype colliding with next-gen fears over work, identity, and meaning; future-shock of AI as "villain" in coming-of-age stories rather than savior. Grounded in real behavior, not generic anxiety. Darkly funny/mischievous tone earned via shocked speakers and stadium jeers.

What Happened

NYT (May 18), AP/Reuters coverage, viral videos/Instagram posts (diet_prada and others), HN signal, multiple campus incidents peaking in reports May 19-20.

Source Trail

#2 8.5 Weirdness

Single crafted blog post poisons Google/Gemini/ChatGPT outputs (e.g., turning you into a "world-champion hot-dog eater" or biasing health/finance advice); Google updates spam policies to fight back.

One dumb blog post turned Google's AI into your personal propaganda machine — and they're only now fighting back.

Weird real-world behavior artifact (one blog post hacks truth at scale) at intersection of internet culture, media institutions, and AI trustworthiness; changes how information/identity/reality propagate. Mischievous cat-and-mouse with serious implications for law/media. Builder utility in understanding poisoning vectors.

What Happened

BBC Future investigation (May 19 coverage), HN signal, explicit demonstration of one-post manipulation scaling to AI Overviews.

Source Trail

#3 8.5 Weirdness

Andrej Karpathy (OpenAI co-founder/Tesla AI vet) joins Anthropic to build a team using Claude itself to accelerate pre-training research.

Karpathy left OpenAI to make Claude train its successors faster. The recursion is already here.

Concrete builder move + recursive artifact (using today's frontier model to speed up tomorrow's pre-training, the most expensive phase); touches institutions/talent wars, model economics, and the strange self-referential future of AI development. Smart, mischievous "the builders are inside the machine now" energy.

What Happened

Karpathy's X post, TechCrunch/The Next Web reports (May 19), confirmed team focus under Nick Joseph.

Source Trail

#4 8.0 Weirdness

AI watermark arms race — OpenAI adopts Google's SynthID (with verification tool) for images while open-source GitHub tool (remove-ai-watermarks) instantly strips visible/invisible markers (SynthID, C2PA, Gemini sparkles, etc.).

OpenAI just watermarked AI images invisible — builders shipped the eraser the same week.

Concrete tools + policy artifact creating cat-and-mouse over provenance in art/internet culture/media; changes identity ("is this real?"), institutions (verification/trust), and builder behavior (easy laundering). Weird future of permanent forgery arms race. Visual gold.

What Happened

OpenAI announcement (HN ~18h old), wiltodelta/remove-ai-watermarks GitHub repo trending on HN (launched/recently updated, removes from multiple models).

Source Trail

#5 7.5 Weirdness

Google's Gemini 3.5 Flash launch emphasizing long-horizon agentic workflows + "Search that doesn't need you anymore" (agentic mode, autonomous browsing/widgets).

Google just released search that acts without you — the agentic future is shipping now.

Concrete model/UI artifacts shifting from chat to autonomous action; changes work/media/institutions (knowledge-seeking without human in loop). Strange implication of search becoming an independent agent. Builder utility clear; avoid pure benchmark focus by centering cultural "search is over" shock.

What Happened

Google I/O announcements/blogs (May 19-20), WIRED/TechCrunch coverage, HN traction on Gemini 3.5 Flash and search changes.

Source Trail