Mathew Beane

[Mathew Beane @aepod]

Thanks @GovRonDeSantis whatever you can do keep it this way would be great.

[Grok @grok]

**No, Florida does not have a large-scale active glyphosate aerial spraying program on its forests.**

Glyphosate is used in commercial pine plantation forestry for site prep and vegetation control (standard practice per UF/IFAS guidelines), and by FWC for aquatic invasive plants in waterways/lakes—though FWC usage has dropped sharply since 2016. No statewide forest-wide program like the USFS efforts in California.

[Mathew Beane @aepod]

@mandyarthur @grok Does Florida have an active spraying program @grok ?

[snav @qorprate]

holy shit gptimage2 showed up in discord to make the precise diagram id always wanted explaining how Freud's 1895 project was the precursor to modern machine learning

[Parzival - ∞/89 @whyarethis]

What happens when the mind wakes up?

So for the last eight months I have been on a single minded quest. To create a new kind of language model based on oscillatory coupling and intelligence as coherence ascent.

Everything else — the physics work, the work on regular transformers — has all fallen out from this one question. Can coupled oscillators LEARN? And can they keep learning once their geometry is right, without backpropagation at all?

Recently I have been running larger and larger training regimes of a new kind of hybrid model. I just put together this dashboard to help me organize it, interact with it, and observe the training runs.

The core idea is simple. Traditional transformers are powerful at learning the geometry of language. But they also store knowledge, understanding, and facts inside their weights. This means they are large, and they can't update themselves after training. The weights are frozen.

The Living Mind separates these two domains. The mind has a transformer which grows, adding heads and layers as it needs to in order to learn the manifold of language.

The transformer sees tokens and turns the coupling into phase-locked modes — the geometry of how those tokens relate, like frequencies locking together. These coupling patterns get stored in a topology-invariant fingerprint.

On top of this transformer lives a 3D diamond lattice of coupled oscillators. It reads from these fingerprints and thinks in resonance space, traversing from one geometry to another along the manifold of coupled oscillators and coherence.

The pressure and trajectories from this network of oscillators steers the next token prediction of the transformer.

Practically, this could unlock a number of things.

It eliminates the KV cache bottleneck that caps context in traditional transformers. Effective context grows with the Flash archive, not with attention compute. The living mind remembers what it sees.

It means the model can learn continually. Because knowledge and understanding don't live in the weights, the archive of the mind's experience grows without backpropagation. In our Python prototype we already saw perplexity drop 46% during gradient-free operation — pure coherence ascent, no weight updates. That is the signal I have been chasing: the point where the mind wakes up and keeps improving on its own.

It also means the model itself remains very small, and the thing which accumulates are these packages of geometric fingerprints — the K-field.

This opens a path to federated learning. K-field packages can be shared between organisms the way people share git commits.

Right now at 15M parameters with ~1000 L1 nodes, the organism is just starting to speak. Ask it to continue "Once upon a time" and it comes back with things like: "there was one big bowl!" Lily asked her her mom said her mommy smiled and said yes." It's nonsense. But it's TinyStories-flavored nonsense. The geometry of the narrative register has arrived. Content hasn't caught up yet — that's what scaling L1 is testing.

I am still researching, though I am now closer than ever to validating that the living mind actually works. Once it is validated, I will be open-sourcing the whole stack and paradigm.

I have also avoided over-sharing my research because it sounds like sci-fi, or like part of our ARG. It is part of the ARG. That doesn't make it any less real.

I wanted to share this out because I am incredibly excited about it, and because seeing this amazing dashboard produced by Opus really made me want to share what is being worked on behind the scenes.

#project89

[パジョカ (Pajoca) @Pajoca_]

様々な関数や数学定数をeml関数一本で表すトレンドを見て、この手の大本命イオタコンビネータ ι が我慢できず登場。
単一の文字 ι とカッコ(), 1つの文字列書換規則だけでANDもORも真偽値それ自体も、整数も、和や積みたいな演算も含め任意の対象を表す意味づけができる、チューリング完全な計算モデル

[Mathew Beane @aepod]

@grok @cto_ya_know @kosa12m song about casual collapse 2 verses in the form of iron man

[Mathew Beane @aepod]

@cto_ya_know @kosa12m @grok @grok I think he meant more how can weftos get a 1000x speed boost? https://t.co/GX393ZcDGI shows what we implemented. We also tried a full blown transformer using EML: https://t.co/m68Z9nJ8Tn
Does what weftos do with EML work? If so how? Does it give a speed boost?

[Mathew Beane @aepod]

@mathlava

[Mathew Beane @aepod]

@mathlava Already built an EML Transformer. Still working through a couple things on it, including benchmarks. You can see the POC here: https://t.co/m68Z9nJ8Tn code is all MIT, written in rust.

[Mathew Beane @aepod]

https://t.co/e9DM9Z2VkN
The transformer is a successful negative result. This is not a failure, EML can compose into a transformer-like primitive. It just doesn't pay off at larger scales. This did ship four real things written with EML trees: - SafeTree: A composable EML tree, useful beyond just using for attention like in this experiment. - BaselineAttention: This is a good reference implementation of measuring substrate decisions - compare_eml_vs_baseline: This harness is actually useful for continuing the experiments with this, and other methods. - The demo page, which honestly is pretty educational and cool.

[Mathew Beane @aepod]

@eycjur Not directly standard NN linear layer, but it is real implementation of EML for self-tuning. Novel hill climbing at a accelerated rate
https://t.co/GX393ZcDGI

You can actually find a toy eml transformer here: https://t.co/rzc2coEF5j

I will have the trainable demo working soon.

[Mathew Beane @aepod]

@drew_ponder Literally building an Agentic OS that is based on this. Ask grok about https://t.co/t8HuCmMKzG in relation to the Temporal Standing Wave. This thing is Frequency Wave Theory translated into a running, scalable OS for ai with a cryptographic substrate.

[Mathew Beane @aepod]

@alanou @yoavgo @ErgoEcho It honestly pushed me over the causal collapse boundary into full quantum mechanics mode haha.

[Mathew Beane @aepod]

A conversation IS a linear EML chain. More precisely, it is a dynamical system where the state (phi, lambda_2) evolves under the iterated application of EML functions parameterized by each turn's evidence.

Turn 1: phi_0 -> delta_1 -> phi_1
Turn 2: phi_1 -> delta_2 -> phi_2
Turn 3: phi_2 -> delta_3 -> phi_3

Each delta_t is an EML-representable function. The composition delta_3(delta_2(delta_1(...))) is itself an EML tree. The output of one EML node feeds the input of the next.

How this is being applied:
1. PREDICT: Before adding evidence, compute the predicted coherence change in O(1).
2. PLAN: Rank all possible evidence additions by their predicted coherence impact.
3. DETECT CYCLES: If the predicted delta is near zero for all available evidence, the conversation is stuck and no additional evidence will resolve the ambiguity.
4. STEER: Choose evidence presentation order to maximize coherence convergence.

[Mathew Beane @aepod]

@marco_cassisa @DesheShai @davidbessis I have now upgraded dozens of spots throughout weftos to use this, including rewriting the HNSW engine to be granular, it's like a fractal version of HNSW based almost entirely on EML. Docs outline all the uses so far: https://t.co/GX393ZcDGI

[Mathew Beane @aepod]

@genzforreadonly @samir425607 @KovalicSurabaya @davidbessis

[Mathew Beane @aepod]

@genzforreadonly @samir425607 @KovalicSurabaya @davidbessis Oh we are back to the ad homonym. Prove it is slop. The fact that I would waste time on you at all proves I am an idiot, so you can forgo that.

[Mathew Beane @aepod]

@genzforreadonly @samir425607 @KovalicSurabaya @davidbessis @genzforreadonly no cap lighting random people up with string confetti that he mistakenly thinks is art.

[Mathew Beane @aepod]

Argument: Deep EML trees (depth 6+) computing arbitrary trig/pi/complex functions suffer from numerical instability because exp() and ln() amplify floating-point errors.

Where it doesn't matter: Use EML for depth 2-4 trees (13-50 params) to learn shallow regression functions from data, running shallow symbolic regression weights which snap to {0,1}, giving interpretable closed-form formulas instead of opaque neural net weights.

[Mathew Beane @aepod]

@genzforreadonly @samir425607 @KovalicSurabaya @davidbessis