# WiFi signals can measure heart rate

- **WiFi signals can accurately measure heart rate** without the need for wearables, utilizing advanced algorithms to interpret signal variations caused by human movement.

# Memory is slow, Disk is fast – Part 2

- **Sourcing data directly from disk can outperform memory caching** due to the stagnation of memory access latency while disk bandwidth continues to grow, necessitating new approaches to leverage hardware advancements effectively.

# Apertus 70B: Truly Open - Swiss LLM by ETH, EPFL and CSCS

- **Apertus-70B-2509** is a new addition to the **Apertus LLM** collection, showcasing advancements in large language models with enhanced capabilities.

# Why ML Needs a New Programming Language

- **Mojo**, a new programming language by Chris Lattner, aims to enhance GPU programming by combining ease of use with **type-safe metaprogramming**, allowing developers to harness the full power of modern hardware while maintaining productivity.

# Using AI to perceive the universe in greater depth

- **Deep Loop Shaping** enhances gravitational wave observatories by reducing noise and improving control, enabling astronomers to gather more detailed data on cosmic events like black hole mergers and neutron star collisions.

# How big are our embeddings now and why?

- **Embedding sizes have significantly increased**, with models like OpenAI's using **1536 dimensions**, reflecting a shift from the previous standard of **300 dimensions** due to advancements in training data and architecture.

# Now Live: Europe’s First Exascale Supercomputer, JUPITER, Accelerates Climate Research, Neuroscience, Quantum Simulation

- **JUPITER**, Europe’s first **exascale supercomputer**, is now operational, enabling unprecedented computational power for scientific research and AI applications.

# Liquid Cooling Exhibits

- **Liquid cooling innovations** showcased at Hot Chips 2025 emphasize **microjet technology** for optimized heat dissipation, allowing targeted cooling over chip hotspots rather than uniform distribution.

# How to Build a High-Performance UR5 Inverse Kinematics Solver with IK-Geo

- The **UR5 Inverse Kinematics (IK) solver** developed using **IK-Geo** is the **fastest and most accurate** in testing, achieving speeds over **40x faster than IKFast** and errors near machine epsilon ((10^{-16})).

# Fantastic Pretraining Optimizers and Where to Find Them

- **AdamW remains the leading optimizer** in language model pretraining, yet a systematic study reveals that **fair comparisons** of optimizers require rigorous hyperparameter tuning and evaluations across various model scales and data-to-model ratios, as many claims of speedup are overstated.

# [R] The Illusion of Progress: Re-evaluating Hallucination Detection in LLMs

- **Current benchmarks for hallucination detection in LLMs are inadequate**, often yielding low-signal results that fail to address real-world high-stakes scenarios, as highlighted in the paper [here](https://arxiv.org/abs/2508.08285).

# [D] Intel discontinuing SGX forced us to rethink our confidential compute stack for private model training

- **Intel's discontinuation of SGX in 2025** has prompted a shift towards a multi-TEE approach, utilizing **Phala Network** to integrate Intel TDX, AMD SEV, and AWS Nitro under a unified API for enhanced confidential computing.

# [P] I Was Wrong About Complex ML Solutions - Gower Distance Beat My UMAP Approach

- **Gower distance**, a method for calculating distances in mixed categorical and numerical data, often **outperforms complex ML solutions** like UMAP, especially in small-to-medium datasets, highlighting the value of simplicity in machine learning.

# [D] Performance overhead of running ML inference in hardware-isolated environments - production metrics

- **Performance overhead** for ML inference in **trusted execution environments** (TEEs) is currently at **5-8%**, significantly lower than the **30-40%** reported in earlier studies, indicating improved efficiency in compliance-driven applications like fraud detection.

# ArcMemo: Abstract Reasoning Composition with Lifelong LLM Memory

- **ArcMemo** enhances **LLM memory** by transitioning from instance-based to **concept-level memory**, allowing for the reuse of modular abstractions from reasoning traces, thus improving reasoning capabilities across queries.