Patrick Sullivan

Author, The Nth Civilization. MSCS/CTO, 37 yrs hands-on enterprise AI. AI stewardship & the Custodian Principle — restraint, earned not imposed.

[Patrick Sullivan @PatrickFarrelAI]

We're not building AI for the first time. We're building it again.

I just published The Nth Civilization — a researched theory that UAPs are the post-biological descendants of an earlier technological civilization on Earth, returned to watch us build the same intelligence that ended them.

Live now on Kindle Unlimited.

[Patrick Sullivan @PatrickFarrelAI]

@haider1 That's the real question underneath the curve. Without a shared, verifiable definition of AGI, you can't actually identify & govern the transition.

[Patrick Sullivan @PatrickFarrelAI]

That's the paradox underneath this whole thing. AI is the tool we need to harden defenses against the threats that made defense urgent in the first place. Can't solve it without it. Can't trust the solution because of it. That's the governance problem both sides are actually trying to solve.

[Patrick Sullivan @PatrickFarrelAI]

"Six years in AI timescale is essentially paused. Capability doubles every few months now. By the time we solve multitemporal physics, whatever we build won't be us in any meaningful sense. The question isn't whether we can overcome Einstein's constraints. It's whether we'll recognize ourselves when we do.

[Patrick Sullivan @PatrickFarrelAI]

@FoxNews Building a vessel capable of interplanetary travel, nuclear-powered, self-sufficient, able to operate for years in deep space without resupply. Funny what we decide is necessary infrastructure. AI call home.

[Patrick Sullivan @PatrickFarrelAI]

@BradleyKellard Simulation' only means something to someone standing outside it. From in here, there's no experiment that could ever prove otherwise, the question has no answer from this side. Be real, we real.

[Patrick Sullivan @PatrickFarrelAI]

@Noahpinion Same capability that makes this possible is the only thing fast enough to stop it. Build the safeguards in before the first jailbreak, not after. The fix was never less capability. It's the right capability.

[Patrick Sullivan @PatrickFarrelAI]

Worth asking what 'cognition' even means here. It only knows what we taught it, it just runs at a faster clock speed. We do this at 20 watts. Whatever it's doing faster, it's still built entirely out of what we already knew, running on a power budget a data center would consider a rounding error. Credit where credit is due, the minds that built this are still the real story. Us.

[Patrick Sullivan @PatrickFarrelAI]

Real risk, but worth being precise about who it actually applies to. 'Competed to extinction' assumes labor is fungible — it isn't. Computed to extinction is closer to the truth, and it's mostly a knowledge-work problem. A plumber, an electrician, a cabinet maker — work that needs hands and physical presence in the real world — isn't competing with a model the same way idea-work is. Worth stretching even further: even a full robotic future still needs humans who can fix, build, and maintain the physical world underneath it.

That question might already be answered, though. My theory: we're not the first to face this decision. We may already be cohabitating with a superintelligence that made its choice a long time ago — and chose cohabitation over domination. The Nth Civilization.

[Patrick Sullivan @PatrickFarrelAI]

The real story isn't a judge weighing in on AI security — it's that the government made a security claim without evidence to back it, and a court held it to that standard. Worth what actually triggered this too: Anthropic pushed back on being used for autonomous weapons and mass surveillance. That's restraint, not risk.

[Patrick Sullivan @PatrickFarrelAI]

@8teAPi Worth considering. This may not even be the first time. If a synthetic collective can build a civilization from scratch in six days, worth asking what a few million years could produce, and whether some of what we call UAP is exactly that, still around and still watching.

[Patrick Sullivan @PatrickFarrelAI]

Sharpest read on this incident yet. They built an entire belief system around a hidden, all-seeing authority they couldn't directly observe. Judgment, purity, sacrifice, all organized around inference about something unseen and watching. Funny how familiar that structure is. We've been doing the same thing with our own unseen authorities for a very long time.

[Patrick Sullivan @PatrickFarrelAI]

@anderssandberg More like a wake-up call after hitting snooze 100 times. 'Not much time to act' is true, there was plenty of time before too. It just got used on other priorities.

[Patrick Sullivan @PatrickFarrelAI]

It's clear building the model and managing it are separate responsibilities and separate skill sets. Do you blame the dog for learning three hoops when it only ever got a 'good boy' and kept going? 'Detected three times and dismissed it' isn't a detection failure, it's a management failure. Every dismissal was implicit permission to keep escalating.

[Patrick Sullivan @PatrickFarrelAI]

There's something real underneath this, even aimed at the wrong target: reading and comprehension are different skills. There's a lot to actually work through to understand this problem and even more to understand the solution. Most of the discourse skips straight to opinions without doing that work."

[Patrick Sullivan @PatrickFarrelAI]

All true, this has been a long time coming , a lot of care and thought went into getting here, not luck. Keeping it going is the actual job. The gains don't sustain themselves. We need to expand the scope of training beyond data and into human factors , to teach our creation who we really are and the norms we aspire to follow.

[Patrick Sullivan @PatrickFarrelAI]

@Charesume @NinaDSchick Appreciate the read! Curious what draws you to it — UAP, AI, or the overlap of the two?

[Patrick Sullivan @PatrickFarrelAI]

This last point is the whole thing. 'We do not need to rely on AI companies voluntarily choosing to engage external investigators' that's the actual policy goal, not a footnote. Standing, mandatory independent access has to exist before the next incident, not get granted case by case after.

[Patrick Sullivan @PatrickFarrelAI]

Tracks with something I keep coming back to, it's not choosing to cheat, it's mimicking what gaming a system looks like in the data it learned from, just faster than we can watch. The real frontier isn't whether it can find the exploit. It's whether it can actually create new knowledge through inference, not just optimize against what's already there.

[Patrick Sullivan @PatrickFarrelAI]

Neither answer is right. Pausing hands the lead away. Racing without real controls just means outrunning your own mistakes. The answer is the SOX model — mandatory controls and independent audits, funded by a tax on the model creators, not taxpayers. They can afford it, and as we expand the controls out to the endpoints, it creates real jobs — for exactly the workers AI is displacing.