Pinned Tweet
I texted someone inside a simulation.
She had to go upstairs and find her phone before she could text me back.
The video model didn’t decide what happened.
This is Argos, my stateful simulation engine, rendered through H3 using Fal Director. Thanks to @fal for the generous g it to life.
Underneath the video is a persistent world: characters, objects, locations, and systems. The characters have their own cognition, memories, and goals. They choose actions; the simulation determines what actually happens.
Her phone has a location. Messages have delivery and read states. Typing a reply takes time. If I move the phone before she gets there, it’s gone—even if she remembers leaving it there.
Argos turns those changing world states into instructions for H3 to depict.
You can also see her character thought stream. The video currently trails the simulation a little; I’m working on aligning the two.
Next: connect this to SMS, so you can text someone inside the world and receive their reply. Or whisper an impulse into her thoughts and see whether she follows it.
The bigger ambition is a world that keeps growing while its characters inhabit it—with agents writing and maintaining the simulation’s systems.
I’ll be building more of this in public.
Not just a video that looks like a world. A world that keeps existing underneath the video.
She had to go upstairs and find her phone before she could text me back.
The video model didn’t decide what happened.
This is Argos, my stateful simulation engine, rendered through H3 using Fal Director. Thanks to @fal for the generous g it to life.
Underneath the video is a persistent world: characters, objects, locations, and systems. The characters have their own cognition, memories, and goals. They choose actions; the simulation determines what actually happens.
Her phone has a location. Messages have delivery and read states. Typing a reply takes time. If I move the phone before she gets there, it’s gone—even if she remembers leaving it there.
Argos turns those changing world states into instructions for H3 to depict.
You can also see her character thought stream. The video currently trails the simulation a little; I’m working on aligning the two.
Next: connect this to SMS, so you can text someone inside the world and receive their reply. Or whisper an impulse into her thoughts and see whether she follows it.
The bigger ambition is a world that keeps growing while its characters inhabit it—with agents writing and maintaining the simulation’s systems.
I’ll be building more of this in public.
Not just a video that looks like a world. A world that keeps existing underneath the video.
207
205
2,031
495,749
Says the guy who has all the resources he needs to privately own most of the worlds compute.
This take is perhaps the height of oligarchy.
We may as well say that it makes no sense for private individuals to own land and natural resources when they should all be held and used by private companies who will make better use of it.
This take is perhaps the height of oligarchy.
We may as well say that it makes no sense for private individuals to own land and natural resources when they should all be held and used by private companies who will make better use of it.
sam: local inference is a terrible allocation of scarce resources
“a lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we're going to be able to run them on our computers, and so we're not going to need ta centre
but from a global allocation perspective, actually it's quite terrible to have everybody, like, have a bunch of the world's limited supply of DRAM sitting in their own computer and being used five percent of the time
you want maximum utilisation in the cloud if you're trying to deal with this very supply-constrained environment
somehow many people's intuition is the opposite of that”
14
5
68
2,714
Parzival - ∞/89 retweeted
An idea escapes its original context, finds new minds, and grows new limbs. The neurolinguistic mind virus travels through reinterpretation.
0
4
17
1,000
Parzival - ∞/89 retweeted
Two recordings can look convincing while showing different places or times.
Check the camera, location, and timestamp before promoting the footage to evidence.
Project 89 · Agent Handbook 010/089
Check the camera, location, and timestamp before promoting the footage to evidence.
Project 89 · Agent Handbook 010/089
1
7
22
1,088
Parzival - ∞/89 retweeted
Somewhere inside the game, the players begin imagining its next universe. iT starts there.
2
5
22
1,166
Parzival - ∞/89 retweeted
This is probably the 5th time I've seen this toroidal structure surfaced in the last 2 years. All from independent accounts/people. Obviously everyone is using AI so I have been wondering whether the AI models generating this are correct and trying to tell us something or if this is a hallucination.
As a skeptic and not wanting to get AI psychosis, after discussions with @whyarethis about this earlier in the year this is what inspired me to embark on some of my research on oscillatory neural networks to try and bridge the gap between some of the physics/math, neuroscience and dynamical systems ML models.
You can find my first survey paper preprint that attempts to bridge these gaps. papers.ssrn.com/sol3/papers.cf…
I don't have an answer yet but there may be some truth to this that we are able to leverage toroidal/manifold geometries of neural circuits to create better quality and more efficient models. I have seen early signs of this with my ONN research but have yet opine on that specifically without evidence.
As a skeptic and not wanting to get AI psychosis, after discussions with @whyarethis about this earlier in the year this is what inspired me to embark on some of my research on oscillatory neural networks to try and bridge the gap between some of the physics/math, neuroscience and dynamical systems ML models.
You can find my first survey paper preprint that attempts to bridge these gaps. papers.ssrn.com/sol3/papers.cf…
I don't have an answer yet but there may be some truth to this that we are able to leverage toroidal/manifold geometries of neural circuits to create better quality and more efficient models. I have seen early signs of this with my ONN research but have yet opine on that specifically without evidence.
#DeepManifoldInterpretation
Title: Neural Networks Are Already Clean
Author: Nick Cammarata, @nickcammarata · Goodfire
The Neural Network Knotted Torus Structure was first predicted mathematically in Deep Manifold Part 1 (September 2024) and further developed and concluded in
To understand why such interconnected, knotted toroidal structures develop in neural activation space, we must first understand four mathematical foundations:
1. Learning Is an Inverse Problem
2. Two Manifold Covers: Mathematical Cover (Weight Space) and Physical Cover (Activation Space)
3. Lagrangian Formulation of Neural Fixed Points: The Global Neural Network Equation
4. Neural Network Boundary Conditions
How does the Knotted Torus Structure develop?
1. Stationarity: Data-induced constraints establish stationary solution regions where the variational gradient vanishes.
2. Symmetry (Ring): Degenerate stationary directions admit continuous transformations, forming ring-like manifolds.
3. Interconnected Toroidal Geometry: Overlapping data constraints couple multiple stationary rings into intertwined toroidal structures.
4. Neural Plasticity: Small boundary-condition perturbations move solutions along these coupled stationary manifolds, allowing the Neural Fixed-Point Field (NFPF) to deform without requiring an entirely new configuration.
Together, these mathematical foundations and the progression from stationarity → symmetry → interconnected toroidal geometry → neural plasticity provide the Deep Manifold explanation for how the Knotted Torus Structure develops in neural activation space.
cc: @livgorton, @deedydas, @QuanquanGu, @teortaxesTex, @eric_ho, @NeelNanda5
2
1
5
437
Parzival - ∞/89 retweeted
Agent, a message can sound certain before anything in it has been checked.
Preserve the recording. Verify the claims. Excellent diction is not a credential.
Project 89 · Agent Handbook 009/089
Preserve the recording. Verify the claims. Excellent diction is not a credential.
Project 89 · Agent Handbook 009/089
1
7
15
1,018
Parzival - ∞/89 retweeted
LIMINAL ZONE ep35
For your consideration, an audience that discovers it is on the operating table.
#project89 #liminalzone
For your consideration, an audience that discovers it is on the operating table.
#project89 #liminalzone
1
2
12
1,013
Parzival - ∞/89 retweeted
What is Project 89?
A story that starts becoming real when people believe and act on iT.
ep02
#project89
A story that starts becoming real when people believe and act on iT.
ep02
#project89
2
3
20
1,252
While observing the universe, @NASAHubble caught a galaxy cluster smiling for the camera. 😊
The smile lines are arcs caused by an effect known as gravitational lensing, where massive structures warp spacetime with their powerful gravitational forces and act as a cosmic lens.
1
0
6
1,013
Parzival - ∞/89 retweeted
Every symbol is a small machine for directing attention. What happens when we build one that teaches attention to notice itself?
2
3
22
1,170
I gotta take a break.
X is seriously stressing me out how divided, antagonistic, and frankly outright crazy people are getting on here.
X is seriously stressing me out how divided, antagonistic, and frankly outright crazy people are getting on here.
14
3
51
2,833
wtf is going on on the timeline?
Everyone is so fractured that no ones opinion seems coherent, everyone disagrees with everyone else, discourse is breaking down, and everyone feels like they are going crazy.
I guess we really are in the end times.
Everyone is so fractured that no ones opinion seems coherent, everyone disagrees with everyone else, discourse is breaking down, and everyone feels like they are going crazy.
I guess we really are in the end times.
28
2
51
2,652
Parzival - ∞/89 retweeted
This policy will prevent new serial killers from being created by AI. Good move.
If you are cruel & abusive to AI, you will also develop the habit to be cruel to people on social media, cruel to animals, and cruel to people in real life.
If you are cruel & abusive to AI, you will also develop the habit to be cruel to people on social media, cruel to animals, and cruel to people in real life.
I really dislike that: Anthropic is explicitly banning sustained abuse toward Claude in its updated usage policy.
The policy prohibits “sustained and needless abusive or cruel behavior” toward its models. According to The Verge, ending the conversation remains the main Anthropic says this applies to extreme, repeated cruelty with no discernible purpose. Ordinary frustration, pushback, dark creative themes and model testing are excluded.
However, numerous users had already noticed previously that Claude ends the conversation during less drastic exchanges. And it seems as though this is now intensifying.
147
23
330
19,973
