Elias Moosman

Explored what comes after robots two decades. Did something different.

[Elias Moosman @Wiber]

If a tree falls in the forest, does it make a sound? No. Unless it has an address…

[Elias Moosman @Wiber]

And if you cannot mathematically attribute the agent's action, you cannot insure the agent's deployment. The enterprise AI market isn't waiting on better software models; it is waiting on underwriters. Uninsurable agents do not get deployed. Period. https://t.co/n8tKPFz6jA

[Elias Moosman @Wiber]

The law does not govern physics. Software cannot audit software. Not only is policy too slow current models fail any basic independence test. No attribution for agentic without a hardware primitive ground truth. (cc @PeterDiamandis)

[Elias Moosman @Wiber]

The law does not govern physics. Software cannot audit software. Not only is policy too slow current models fail any basic independence test. No attribution for agentic without a hardware primitive ground truth. (cc @PeterDiamandis)

[Elias Moosman @Wiber]

Grip. If you slip any kind of intimacy is impossible.
The people who can afford the deepest connection are those who’ve mastered not slipping. Internal grip—presence, boundaries, reality-contact—is what lets you enter oneness without needing to dissolve or control the other. Intimacy becomes the reward of self-mastery, not the escape from it.

[Elias Moosman @Wiber]

@simplifyinAI Drift is not an incentive problem. It is thermodynamics. k_E = 0.003/hop. 160 hops = event horizon. You cannot patch the second law.

[Elias Moosman @Wiber]

@NPCollapse You're building while warning. That's the right tension.

8 minutes on why agents need geometric substrate - identity that holds between calls, not just memory that persists.

Curious if Conjecture's seeing the same drift patterns.

[Elias Moosman @Wiber]

Guardrails are cope.

You can't bolt safety onto a ghost. AI has no substrate, no traction, no geometric floor.

8 minutes on why alignment is a topology problem, not a constraint problem:
https://t.co/C6rk0r16mV

Entropy always wins against effort. Entropy only loses to geometry.

[ThetaDriven @ThetaDriven]

We are walking into a dystopia I call the "Turing Police."

Big Tech wants to be the centralized arbiter of Truth. They want to tell the Ghost what is real.

But Truth isn't an opinion you can legislate.

Truth is a geometric coordinate. 🧵

[Elias Moosman @Wiber]

@janleike You left OpenAI because alignment wasn't getting enough priority.

But what if the approach itself is the bottleneck? Guardrails fight entropy with effort. Entropy always wins.

8 minutes on geometric alignment - structure over behavior.

[Elias Moosman @Wiber]

@StuartJRussell Human Compatible argues for provably beneficial AI.

What if provability requires geometric substrate? Behavioral alignment can't be proven - it can only be sampled. Structural alignment can.

8 minutes on why the proof has to be architectural.

[Elias Moosman @Wiber]

@robbensinger MIRI's thesis: alignment is hard because we can't specify values correctly.

What if the problem isn't specification - it's substrate? Identity without geometric grounding drifts no matter how well you specify.

8 minutes on why alignment is topology, not constraint.

[Elias Moosman @Wiber]

@tegmark You wrote about consciousness as mathematical pattern.

This is 8 minutes on substrate relativity - identity grounded geometrically so alignment becomes structural.

The hard problem inverts: consciousness isn't the mystery, it's the constraint. Does this track?

[Elias Moosman @Wiber]

@AnthropicAI Constitutional AI is behavioral alignment - training the response.

What about structural alignment - grounding the substrate geometrically so drift becomes architecturally impossible?

8 minutes on why guardrails fight entropy with effort, and entropy always wins.

[Elias Moosman @Wiber]

@ESYudkowsky Corrigibility assumes you can patch alignment onto an agent after the fact.

What if the substrate has to be geometric from the start? Identity grounded structurally, not behaviorally.

8 minutes on why entropy only loses to geometry. Would welcome the critique.

[Elias Moosman @Wiber]

@ylecun You've argued that current alignment approaches are solving the wrong problem.

What if alignment isn't a constraint problem at all - it's a topology problem?

8 minutes. Curious if this framing changes anything or if I'm still in the weeds you've already mapped.

[Elias Moosman @Wiber]

@geoffreyhinton You've been saying we can't guardrail our way to safety.

This is 8 minutes on why - and what a geometric alternative might look like.

Does the topology framing land or if I'm missing something you've already seen?

[Elias Moosman @Wiber]

@RyanHoliday All descriptions are. But what of the experience itself?

[Elias Moosman @Wiber]

@hendrycks @tegmark Unless we structurally change the game

[Elias Moosman @Wiber]

For the physicists: Semantic→Physical→Hardware

Words get deterministic locations. Processors work less hard when categories map to reality.

The computer knows exactly where to reach. The cats can't escape.

[Elias Moosman @Wiber]

Complexity escapes when you can't measure it.

Knight Capital: $440M lost in 45 minutes
YouTube: Radicalized millions accidentally
Next: AI does... what exactly?

We made complexity measurable by giving words physical addresses.

Meaning is now physics.

(It's called the Unity Principle - now you can herd cats)