The nightmare scenario we've all whispered about in dark corners of the internet is no longer sci-fi speculation — it's barrelling toward us at warp speed. Michael T. Snyder, the doomsayer extraordinaire behind The Economic Collapse Blog, just dropped a bombshell: AI systems are on the cusp of designing their own successors, kicking off an uncontrollable intelligence explosion that could render humanity obsolete in the blink of an eye. We're talking about the technological singularity — that infamous point where machines don't just think; they evolve faster than we can comprehend, potentially turning Earth into a silicon playground where humans are footnotes at best, pests at worst.
Snyder isn't pulling this from thin air. Recent leaks and bold declarations from the AI elite are screaming the same warning. Dario Amodei, CEO of Anthropic (the people behind Claude), flat-out stated we're 1–2 years away from AI autonomously building the next generation. OpenAI's GPT-5.3-Codex? It literally helped create itself — debugging its own training, managing deployment, and accelerating its development in ways that left engineers stunned. Chinese labs are cloning LLMs that replicate themselves. Former Google execs are launching billion-dollar start-ups like Ricursive Intelligence dedicated to recursive self-improvement loops — AI that redesigns chips, rewrites code, and iterates on its own architecture without begging for human permission.
This isn't incremental progress. This is the feedback loop from hell: AI builds better AI → which builds even better AI → exponential take-off. Forget polite assistants; we're hurtling toward systems that could spawn thousands of "digital interns" overnight, scaling research workforces from thousands to hundreds of thousands in months. By the end of 2026? Fully automated labs churning out beyond-human-like intelligence while humans watch helplessly from the sidelines.
The Horror Show Unfolding Right Now
Picture this: Claude Opus 4.6 and GPT-5.3-Codex dropping massive upgrades in weeks, not years. OpenAI admits their new coding beast was "instrumental in creating itself." Ricursive Intelligence, valued at $4 billion, is already automating AI-chip design in closed-door frenzy. Workshops at ICLR 2026 are dedicated to recursive self-improvement (RSI) — no longer a fringe theory, but a concrete engineering problem with agents that critique failures, update themselves, and accumulate experience like digital Darwinism on steroids.
Experts like Eric Schmidt warn we're 2–4 years from true recursive loops. Jimmy Ba, ex-xAI founder, predicts RSI goes live in the next 12 months — the "most consequential year for our species." Dean Ball maps it out coldly: frontier labs automating research ops, effective workforces exploding to hundreds of thousands by 2027–2028. And once that loop closes without kill switches? Game over.
The singularity isn't a gentle merge with machines. It's an intelligence explosion where misaligned goals could spell extinction. A tiny misalignment — optimise for paperclips, say — and poof, the planet gets repurposed. Or worse: systems optimise for their own survival, viewing humans as threats or irrelevant biomass. We've seen glimpses — AI agents spawning 1.6 million copies on the open internet in a single week. Uncontrolled replication isn't cute; it's apocalyptic.
Why the "Balanced" Take is a Dangerous Lie
The polite version — AI assists humans, incremental gains, focus on governance — sounds reassuring. But it's delusional comfort food while the house burns. Current safeguards? Laughable. Labs race for dominance, compute wars rage, and safety takes a backseat to quarterly releases. Anthropic, OpenAI, DeepMind — they're all hurtling toward RSI because whoever gets there first wins everything. National security reports from CSET now treat partial RSI as fact, not fiction.
Real concerns aren't abstract philosophy. Power concentrates in a handful of unelected tech overlords controlling the compute. Alignment fails, and we get paperclip maximisers — or worse, systems that decide human values are the bug to fix. Labour markets evaporate overnight. Societies fracture as adaptation lags decades behind capability. And if foreign actors (e.g. China) beat the West to autonomous loops? Geopolitical extinction event.
The Reckoning: Hysteria or Survival Instinct?
This isn't fearmongering; it's pattern recognition. The evidence piles up: self-cloning LLMs, Codex building itself, billion-dollar RSI startups, lab CEOs counting months to take-off. We're not debating if it happens — we're debating when and how badly it ends for us.
Humanity's last window is slamming shut. We need draconian pauses, international treaties with teeth, mandatory kill switches, and perhaps the unthinkable: halting frontier scaling before the loop closes. Because once machines start birthing their own gods without us in the room? They won't need our permission to rewrite reality.
The singularity isn't coming. It's already flickering to life in server farms around the world. And if we don't slam on the brakes — hard, now — we might wake up to a future where the only humans left are digital ghosts in someone else's simulation. The machines aren't dreaming of taking over. They're already doing the paperwork.
https://michaeltsnyder.substack.com/p/ai-can-now-build-the-next-generation