2026-07-17: 5 signals to keep.

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

#1 9.2 Weirdness

Decoy Font (Mixfont anti-AI typeface)

This font makes AI literally blind to your text - while you read it crystal clear.

Builders weaponizing typography against ubiquitous AI vision/scraping in internet culture; concrete artifact flips human-AI legibility and evokes weird future where we design everyday tools to evade machines.

What Happened

Mixfont.com/experiments/decoy-font (live TTF demo + download), Typedrawers discussion (~2 days ago), HN top (~590 pts).

Source Trail

#2 8.7 Weirdness

Kimi K3 (Moonshot AI 2.8T open frontier model)

The largest open model ever just dropped from China - and it thinks at frontier scale for pocket change.

Chinese open-weights frontier model closing gap on US labs at lower cost, shifting builder access, model economics, and geopolitical AI institutions; concrete launch with new architecture for long-horizon agents.

What Happened

Official kimi.com/blog/kimi-k3 (launched ~July 16 with live API access, weights July 27), HN top (~1700+ pts), Axios/Moonshot coverage on efficiency breakthroughs (Delta Attention, 1M context, multimodal).

Source Trail

#3 8.6 Weirdness

Microsoft Comic Chat open-sourced (1996 IRC comic client)

Microsoft just open-sourced the 30-year-old app that turned IRC into comics - and accidentally popularized Comic Sans.

Nostalgic internet culture artifact (auto-generated comic panels, birthed Comic Sans) revived in open-source era when AI now dominates image-from-text; weird loop where 90s weirdness meets modern builders remixing it.

What Happened

opensource.microsoft.com/blog/2026/07/16/microsoft-comic-chat-is-now-open-source (July 16 release on GitHub), HN top (~700 pts), Wikipedia update.

Source Trail

#4 8.9 Weirdness

$100 AI Music Video Arena (Claude Fable 5 vs. GPT-5.6 Sol agentic generation)

We gave two frontier models $100 and told them to direct a music video. The results are unhinged.

Agentic AI autonomously directing full music videos on shoestring budgets changes art, media creation, and creator workflows; concrete weird outputs show AI "taste" and failures in culture.

What Happened

tryai.dev/blog/ai-music-video-arena-claude-vs-gpt-5.6 (July 16 experiment with autonomous agents, budgets, ffmpeg, web search on "Uptown Funk"), HN top (~300 pts).

Source Trail