Update README.md

README.md

@@ -6,24 +6,23 @@ language:
 
 # Model Card for ZUNA
 
-ZUNA is a 380M-parameter masked diffusion autoencoder trained to reconstruct, denoise, and upsample scalp-EEG signals. ZUNA represents an early effort to build general foundation models of brain signals. Over time, we see these models forming the foundation towards fully thought-to-outcome agentic systems.
+ZUNA is a 380M-parameter masked diffusion autoencoder trained to reconstruct, denoise, and upsample scalp-EEG signals. ZUNA represents a significant milestone in the development of noninvasive thought-to-text and is an early effort to build general foundation models of brain signals that can be used to understand and decode brain states. Over time, we see these models forming the foundation of thought-to-text agentic systems and a key component in our mission to build human-aligned superintelligence.
 
 Given a subset of EEG channels, ZUNA can:
+- Denoise existing EEG channels
+- Reconstruct missing EEG channels
+- Predict novel channel signals given physical coordinates on the scalp
 
-- Denoise existing channels
-- Reconstruct missing channels
-- Predict novel channel signals given the physical coordinates of electrodes on the scalp
+ZUNA was trained on approximately 2 million channel-hours of EEG data from a wide range of publicly available sources. All data used a standardized preprocessing pipeline to make it suitable for large-scale foundation model training.
 
-ZUNA was trained on a dataset of approximately 2 million channel-hours aggregated from various online sources and harmonized with our standardized preprocessing and electrode location standardization pipeline. 
-
-At only 380M parameters, ZUNA can run comfortably on any consumer GPU or even on CPU. 
+At 380 million parameters, ZUNA is lightweight enough to run quickly on a consumer GPU and can be used on CPU for many workloads.
 
 // TODO channel reconstruction gif
 
 
 ## Performance
 
-ZUNA significantly outperforms existing standard methods for channel denoising, reconstruction, or upsampling. We compared ZUNA to MNE's default spherical spline interpolation method which is ubiquitous in practice. We find that ZUNA outperforms MNE in reconstruction accuracy across a range of unseen datasets, even those with a different preprocessing pipeline. ZUNAs advantage is particularly striking for higher upsampling ratios, demonstrating that ZUNA is effectively usinggeneral priors it has learned through large-scale pretraining. 
+ZUNA significantly outperforms existing standard methods for channel denoising, reconstruction, or upsampling. We compared ZUNA to MNE's default spherical spline interpolation method which is ubiquitous in practice. We find that ZUNA outperforms MNE in reconstruction accuracy across a range of unseen datasets, even those with a different preprocessing pipeline. ZUNA’s advantage is particularly striking for higher upsampling ratios, demonstrating that ZUNA is effectively using general priors it has learned through large-scale pretraining.
 
 ![2x3_panel](https://cdn-uploads.huggingface.co/production/uploads/65c05e75c084467acab2f84a/zURH5s_AQvNhTDYY66_Yf.png)
 
@@ -61,8 +60,6 @@ python 3_tutorial_pt_to_fif.py
 
 For more information please see our [technical whitepaper](arxiv) and [blog](blog). If you find ZUNA useful in your work please cite as
 
-TODO citation. 
-
 Organizations or researchers interested in collaborating with Zyphra to improve future versions for specific needs or use cases should contact bci@zyphra.com.