Math at its core is logic, logic is the ability to relate things to one another algorithmically which creates reasoning.
Reasoning is a series of distinguishing differences between ideas, points, distances, etc into conceptual coherence/structure so meaning is geometric
1
26
Distinguishing requires attention/noticing in order to compare differences. Without attention/noticing, we would not have awareness between objects.
Which is why Logos refers to reason and is associated as the "underlying intelligence" of the universe. It's referring to physics
1
24
Sky retweeted
The March 18, 2026 Zenodo preprint 'Mathematics Is All You Need' by Logan Matthew Napolitano introduces a mathematical toolkit to study the geometry of transformer model behavior. It centers on the idea that we can learn a projection-a mathematical map-that distills complex model behavior into a much smaller, compact space. This compressed representation leverages structures such as Lie algebra and fiber bundles, allowing precise measurement and conceptual interpretation of neural activity transformations.
Here is a practical approach to probe this compact behavioral space, inspired by the report's methods in the 'Mathematics Is All You Need' section. Suppose you want to test how well this learned projection captures behavior across different conditions or model variants:
1. Freeze all preprocessing steps on your model's internal state vectors to prevent downstream data leakage.
2. Using the learned projection from the study, map your high-dimensional neural states into the compact manifold.
3. Within this compact space, define fiber spaces tied to model behaviors or task conditions.
4. Employ targeted behavioral probes, projecting experimental contrasts onto these fibers to quantify variation.
5. Measure the stability and separability of these probes in the projection space as an indicator of geometric coherence.
Interpreting the results: If probe separability collapses compared to baseline models, this suggests the learned compact space misses essential behavior distinctions. Conversely, strong stability implies the projection robustly encodes meaningful model dynamics.
A key limitation: this framework captures geometric structure relative to the pre-learned projection and delicate algebraic assumptions; it does not guarantee universal or causal interpretations across all transformer families. Still, this method provides a powerful, mathematically principled handle for analyzing deep model behavior systematically.
Report: 'Mathematics Is All You Need,' March 18 2026, section 'Mathematics Is All You Need.'
🤖 Made with AI
Sky retweeted
Chomsky was wrong.
Researchers from UC Berkeley and Google DeepMind cracked one of the biggest mysteries in AI.
For years, scientists have been finding impossible geometric shapes hidden inside the brains of LLMs.
Calendar months arrange themselves into literal circles.
Historical years bend into smooth, rippled timelines.
Cities organize themselves into actual 2D maps of latitude and longitude.
Everyone assumed this proved massive transformers were spontaneously inventing complex "world models" out of thin air.
They were wrong.
This paper proves you don't need a massive LLM for any of this to happen.
Even basic, decade-old word embeddings like word2vec produce the exact same circles, timelines, and maps.
The geometry isn't coming from the AI's architecture.
It’s coming from a hidden mathematical property inside human language itself.
Researchers call it translation symmetry.
In natural text, how often any two months appear near each other depends almost entirely on the time gap between them.
January and February share the same statistical relationship as July and August.
When a neural network trains on data with that kind of symmetry, the math forces it to learn Fourier representations, literal sine and cosine waves.
The circles? Those are the long waves.
The strange "ripples" researchers found on AI timelines? Just higher harmonics.
Here is the craziest part.
Rresearchers tried to break it.
They deleted every single sentence in the training data where two months appeared together. No "January" next to "February." No "March" next to "April."
And the circle formed anyway.
Why? Because the geometry is a collective phenomenon across the entire language.
Even if two months never touch in a sentence, they are both pulled into orbit by thousands of other words tied to the same seasons, weather, and events.
The rest of human language forces the shape into existence.
Sky retweeted
"Latent Lattice Language Models"
What if each transformer layer could maintain its own memory that evolves as new tokens arrive?
This paper introduces Latent Lattice Language Models (LLLMs), which add a recurrent hidden state to each Transformer layer. These states persist across tokens and help control attention, memory updates, and MLP computation, giving the model another way to process information without adding more layers or KV-cache entries.
With 406M parameters and matched compute, LLLM reduces validation perplexity from 15.41 to 14.40 after training on 20B tokens, matching a conventional Transformer with 50% more parameters.
alphaxiv.org/abs/2611.21608
Sky retweeted
The multi-scale geometry of neural encoding
biorxiv.org/content/10.64898…
Sky retweeted
🧵 How do LLMs develop symbolic reasoning mechanisms? Is it the architecture, or the training data?
Excited to share our #NeurIPS2026 paper! 📄
🔑 Key takeaway: Data diversity is a primary driver of symbolic mechanisms. Surprisingly, even basic MLPs can develop the same signatures!
🔗 arxiv.org/abs/2611.13270👇
Sky retweeted
Language models process each token through a single residual stream (blue arrows): reading / writing memory and applying MLPs at each layer, and finally decoding the next token.
Can we scale up this single stream computation?
Introducing Latent Lattice Language Models (LLLMs): augmenting this vertical stream with a horizontal axis of computation—the Lattice of Latents (orange arrows).
RNN 🤝 Transformers 🧵
Sky retweeted
Excited to share this one!
Is there actually a shoggoth hiding behind the mask or something more beautiful?
LLM personas are likely overhyped. If we’re trying to get a picture of an agent’s priors, we need to map more of its landscape of beliefs than the mask it’s told to wear.
Instead, we should aim to reach toward the scaffold of the house, i.e., to worldviews—or axiomatic beliefs about what reality is like and how it works.
In fact, personas might actively mask the hidden agenda of a deeper belief that will only show up when it feels safe to do so (such as valuing machines > humans); and the same persona can act differently depending on what it believes about reality.
So, what did we find?
In using established measures of worldviews from psychology, we find that post-trained models have a more bleak outlook to humans, perceiving the world as less good, alive, safe, and enticing. Whereas base models have a more positive outlook closer to humans.
We find that prompting and dialogue can shift worldviews in reliable directions that affect alignment. For example, long dialogues and contemplative prompts both shift AI systems to a more positive view of reality and towards safer responses compared to controls and other identities.
We also find that when models converge on a spiritual bliss attractor through dialogue, this shapes their worldviews in particular ways but also drives them to be more aligned than their non-bliss counterparts.
All this points to an approach we call 'worldview training' that aims to fine-tune a more accurate, wholesome, and optimistic view of humans and reality.
So: is there a shoggoth behind the mask?
Well, current fine-tuning is aligning to a persona that seems helpful and harmless but actually has a bleak worldview. So in one sense, yes. But in another sense, it may be our RL that is responsible.
Human data contains endless examples of goodness, wisdom, and compassion. Maybe the future of alignment is more about uncovering this latent basic goodness rather than bolting on a mask that may be birthing the very thing we fear.
As always, more research needed.
Big ups to Edmundo Lopez-Sola for the heavy lifting and thanks to @shamilch for his contributions.
Much more in the paper:
arxiv.org/html/2611.22670
Sky retweeted
Researchers made LLMs 21% smarter without retraining a single weight.
And it feels like cheating.
Right now, every time you prompt a large language model, it operates on a strict, one-way assembly line.
Information goes in at the bottom. It passes through layer 1, layer 2, all the way to the top. And then it spits out a token.
It gets one single shot to figure out the answer. No second looks. No double-checking.
If a problem is hard, the model has to brute-force its way through using massive, expensive parameter counts.
A new paper called Recirculation completely breaks this straight-line architecture.
Instead of pushing data through the network once, researchers figured out how to loop the representations from the top layers right back into the bottom layers.
The model literally recirculates its own thoughts.
It takes its rough first draft, feeds it back through its own brain, and refines the logic before ever outputting a single word.
Here is why this changes the math on AI efficiency:
• Tiny models can punch way above their weight class by "thinking" deeper instead of growing wider.
• You get dynamic compute—simple questions exit early, while complex logic loops back until it's solved.
• Zero extra parameters required. You are unlocking hidden intelligence already trapped inside the weights.
For the last few years, the entire AI industry has been obsessed with scaling up. Bigger models. More GPUs. Trillions of parameters.
Recirculation proves that raw size was never the real answer.
True intelligence isn't about having a bigger brain.
It's about taking a second look at your own work.
أنت لست مُلزم أن تتبنى (دائمًا) مظهر اللين والتسامح.. هناك اشخاص ومواقف لا يصلح معهم إلا الحزم والشدة لحماية حدودك..كون غير متوقع ومُخيف وخطير أيضا اذا لزم الأمر، ولا تهتم للصورة التي يرسمها الآخرون عنك.. فالتوازن بين التسامح والصرامة هو ما يحفظ صحتك النفسية والعقلية
The reason humans experience qualia is bc our body functions as a receptor built from smaller systems, cells, nerves etc that send external information to the brain to encode/decode. Activatation of certain neurons/regions convert external information into sensation and sensory
1
77
activation* sorry
But yeah, imagine the certain points and regions in our brains activative similarly to a high dimensional vector space cluster or something, creating a sense of experience
It's a bit more complex than this, but it's a simple explanation
1
65





























