# May 12, 2025

## Daily

### US Copyright Office found AI companies breach copyright. Its boss was fired
- The **US Copyright Office** has determined that AI companies often exceed fair use by utilizing copyrighted material without consent, leading to the dismissal of its head shortly after the report's release.

### Byte Latent Transformer: Patches Scale Better Than Tokens
- The **Byte Latent Transformer (BLT)** introduces a novel byte-level architecture that achieves **tokenization-level performance** while enhancing **inference efficiency** and **robustness** through the use of dynamically sized patches based on byte entropy.

### Continuous Thought Machines
- The **Continuous Thought Machine (CTM)** introduces a novel neural network architecture that incorporates **neural timing and synchronization**, bridging the gap between biological intelligence and modern AI efficiency, yielding promising results in various tasks.

### Intellect-2 Release: The First 32B Model Trained Through Globally Distributed RL
- **INTELLECT-2 is the first 32B parameter model trained through globally distributed reinforcement learning**, utilizing a unique asynchronous framework that allows for decentralized contributions from various compute sources, enhancing training efficiency and scalability.

### Absolute Zero Reasoner
- The **Absolute Zero Reasoner (AZR)** implements a novel paradigm that enables models to autonomously generate and solve tasks without relying on human-curated data, fostering continuous self-improvement through self-play.

### [R] Continuous Thought Machines: neural dynamics as representation.
- **Continuous Thought Machines (CTMs)** leverage **neural dynamics** as a core representation, challenging traditional deep learning by reintroducing **temporal processing** at the neuron level, which enhances the richness of neuron dynamics.

### Launch HN: ParaQuery (YC X25) – GPU Accelerated Spark/SQL
- **ParaQuery** offers a **GPU-accelerated Spark/SQL solution** that claims to be more cost-efficient and performant than BigQuery, enabling users to save over **60%** on their data processing costs while achieving **2x faster** performance without data migration.

### Why National Labs are investing (heavily) in AI
- **Jason Pruet**, Director of the National Security AI Office at Los Alamos National Laboratory, emphasizes that **AI is not merely a tool** but a transformative force reshaping scientific inquiry and national security, driven by advancements in large AI models.

### Build Your Own Siri. Locally. On-Device. No Cloud
- **Build your own local voice assistant** that respects privacy by executing commands directly on your device without cloud dependency, utilizing models like LLaMA 3.1 and Whisper for seamless interaction.

### SDFs and the Fast sweeping algorithm in Jax
- The **Fast Sweeping Method (FSM)** efficiently solves the Eikonal equation in **O(n)** time, making it suitable for applications like **signed distance functions (SDFs)** in machine learning, where it represents 3D surfaces implicitly.

### In-Memory Ferroelectric Differentiator
- The **in-memory ferroelectric differentiator** utilizes ferroelectric domain dynamics to perform differential calculations directly within memory, significantly enhancing efficiency in data processing and reducing energy consumption to **0.24 fJ** per operation.

### [R] Zero-shot forecasting of chaotic systems (ICLR 2025)
- **Foundation models** demonstrate **zero-shot learning** capabilities in forecasting chaotic systems, achieving competitive results against custom-trained models across **135 chaotic systems** and **108 timepoints**.

### Toward a Sparse and Interpretable Audio Codec
- This work presents a **novel audio encoder** that utilizes a **sparse representation** of audio events, enhancing interpretability compared to traditional codecs like MP3 and Ogg Vorbis.

### [D] Simulating Bias with Bayesian Networks - Feedback wanted!
- The project simulates **bias in machine learning** using a Bayesian network to analyze recruitment, revealing that any observed bias stems from the models rather than the training data itself.

### [P] Llama 3.2 1B-Based Conversational Assistant Fully On-Device (No Cloud, Works Offline)
- The **Llama 3.2 1B Instruct model** powers a **privacy-first mobile assistant** that operates entirely **offline**, ensuring user data remains secure and unclouded.