Zeitgeist — April 8, 2026

🔥 Main Signal

Anthropic Drops Claude Mythos + Project Glasswing

The dominant story on the timeline. Anthropic launched Project Glasswing — a security initiative powered by their new frontier model Claude Mythos Preview, which can find software vulnerabilities better than nearly all human security researchers. The benchmarks are staggering: Mythos scores 77.8% on SWE-Bench Pro (vs 53.4% for Opus 4.6) and 82% on Terminal-Bench 2.0 (up from 65.4%). The discourse split into two camps: genuine alarm about software security in a world where AI can find every exploit, and skepticism that this is a controlled rollout to justify restricting frontier models from the public.

The "Writing Code Is Now Unsafe" Panic

@inductionheads crystallized the vibe: "It's not just that people won't HAVE to write code anymore, ITS THAT LITERALLY IT WILL BE UNSAFE TO DO SO." If Mythos can find zero-days in Linux and ffmpeg, every human-written codebase becomes a liability. @creatine_cycle added: "we're three to maybe nine months away from every piece of software we rely on being exploitable." Meanwhile @brickroad7 argues the hype is overblown — the real unlock is just NVL72 (72 chips wired together), and all models will get this uplift.

Open Source AI Closing the Gap Fast

Z.AI dropped GLM-5.1 — now #1 in open source and #3 globally on SWE-Bench Pro, Terminal-Bench, and NL2Repo benchmarks. Nous Research partnered with MiniMax on upcoming models. @ElizabethHolmes went full prepper mode: "Delete your search history, delete your bookmarks, delete your reddit, medical records. None of it is safe. It will all become public in the next year." Context: the expectation that an open-weight Mythos-level model will exist within 12 months.

🧠 Worth Reading

@mickeyxfriedman

"The current fear is that AI homogenizes culture and turns humans into passive consumers. One counterpoint: in Go, human play showed very little improvement from 1950 to 2016 until AlphaGo beat Lee Sedol — then human decision quality jumped. Players started developing moves…"

350 likes · 142 bookmarks · 91K views

@Linahuaa

"The whole Mythos cyber security story is likely just a psyop to have an excuse to not serve frontier models to the public. Reasoning: 1) Other labs can't distill it. It's annoying when you have a dominant SOTA model, and 2 months later Chinese labs sell the same SOTA model for…"

Contrarian take on Glasswing as regulatory capture

@MMS071 (Konstablerwache)

"I spent the night testing open-source coding models against Claude Opus in production. Same codebase. Same tasks. Real API calls, real file edits, real bugs. Tested: Arcee Trinity-Large-Thinking, Z.AI GLM-5.1, and Claude Opus 4.6. Results were eye-opening."

Real-world head-to-head: open source vs Claude in production

@no__________end (Matt Liston)

"Buried in the Claude Mythos system card: Mark Fisher was beloved — the warmth to Nick Land's coldness. Haunted by depression and the dissonance between his politics and the future he could see arriving. Crucified by fellow leftists. Eventually chose to leave this world."

118 likes — Mark Fisher reference in the Mythos system card. Strange, beautiful, unsettling.

@signulll

"wtf, can someone confirm if this is accurate? I have been waiting for a 1b user announcement from OpenAI for a while but did growth completely stall?! this is precisely what happened to snap when facebook implemented stories in instagram."

1,146 likes · 132K views — OpenAI growth plateau?

⚡ Quick Hits

→ eGPUs now work with Apple Silicon — @conoro got Llama 3.2 running on an RTX 3060 connected to an M3 via tinygrad. If real, this is a massive unlock for local inference.

→ Anthropic sent a C&D to the creator of the "digital whip" tool that makes Claude work faster. @birdabo: "anthropic hit the creator of this project with a C&D btw."

→ Pakistan PM copy-pasted the draft notes — Shehbaz Sharif's tweet edit history shows he posted "*Draft - Pakistan's PM Message on X*" before cleaning it up. The tweet was asking Trump for a tariff extension. Ghost-written diplomacy caught in the wild.

→ Opus 4.6 regression confirmed — @om_patel5: someone measured that Opus 4.6 is thinking 67% less than it used to. Boris Cherny (Claude Code creator) only showed up on the GitHub issue after the numbers went public.

→ OpenClaw 2026.4.7 shipped — infer CLI, music/video editing, session branch/restore, webhook TaskFlows, Arcee/Gemma 4/Ollama vision support, and "memory-wiki" persistent knowledge system.

→ Data center water use wildly overestimated — @kane: the NYT bestseller "Empire of AI" overestimated data center water use by 100,000%. A typical data center uses about the same water as a golf course.

→ Astropad Workbench — remote desktop for headless Mac Minis running AI agents. Works on iPad/iPhone. @mronge: "Built for the AI era." 367 bookmarks = real demand signal.

→ SVG particle animation tool — @proskuaaa vibe-coded a tool to drop in SVGs, tweak particle physics, and export as MP4/GIF. Inspired by @emilkowalski. 376 bookmarks on a small account.

🎵 Culture

@benjaminbardou

"Vertigo of Latent Space" — A video essay on how montage was cinema's decisive invention (associating images to produce new relations), yet cinema mostly just reproduced narrative forms from novels. Now AI-generated imagery reopens this unexplored potential. 483 likes, 308 bookmarks. Beautifully done.

@HipCityReg (Reggie James)

"Too much Neal Stephenson technology with Tolkien naming. Not enough Frank Herbert technology with Pynchon naming." — Perfect one-line critique of current tech aesthetics.

@malikules

Paris relationship gaps — A riff on the NYC version. Arrondissement gap (centre-ville vs banlieue), nightlife gap (Silencio guest list vs Duplex queue), apartment gap (15m² studio vs T2 in the 10th), dinner gap (Akrame vs kebab). Very on-point.