# Smartphone Cameras Go Hyperspectral

- **Smartphone cameras can now function as advanced sensors** through an innovative algorithm that extracts detailed spectral information from standard photographs, enhancing their utility beyond traditional imaging.

# Python on the Edge: Fast, sandboxed, and powered by WebAssembly

- **Wasmer Edge now supports full Python functionality**, enabling the execution of popular frameworks like **FastAPI, Django, and Streamlit** directly in a sandboxed environment powered by WebAssembly and WASIX, addressing the growing demand for Python in edge computing.

# How AWS S3 serves 1 petabyte per second on top of slow HDDs

- **AWS S3 achieves an impressive throughput of over 1 petabyte per second** by leveraging a vast array of slow HDDs, utilizing **massive parallelism** and **erasure coding** to optimize data access and storage efficiency.

# Improved Gemini 2.5 Flash and Flash-Lite

- **Gemini 2.5 Flash and Flash-Lite** models have been enhanced for **better instruction following**, reduced verbosity, and improved multimodal capabilities, making them more efficient for high-throughput applications.

# Video models are zero-shot learners and reasoners

- **Veo 3 exhibits emergent zero-shot capabilities**, enabling it to perform a variety of visual tasks such as object segmentation and edge detection without explicit training, positioning it as a potential **generalist vision foundation model** akin to LLMs in language processing.

# Gemini Robotics 1.5 brings AI agents into the physical world

- **Gemini Robotics 1.5** introduces advanced AI agents capable of **perceiving, planning, and executing complex tasks** in the physical world, enhancing robot transparency and learning across different embodiments.

# Ollama Web Search

- **Ollama's new web search API** enhances model accuracy by integrating real-time web data, significantly reducing hallucinations during information retrieval tasks.

# Launch HN: Flywheel (YC S25) – Waymo for Excavators

- **Flywheel AI** is developing a **remote teleoperation and autonomy stack** for excavators, retrofitting existing hydraulic machines to enhance safety and productivity on construction sites.

# Apple Research Debuts Manzano — a Unified Multimodal LLM

- **Manzano** introduces a **scalable unified framework** that effectively merges image understanding and generation, utilizing a **hybrid image tokenizer** to enhance performance without compromising capabilities.

# Can a model trained on satellite data really find brambles on the ground?

- **A model trained on satellite data successfully identified brambles on the ground**, demonstrating the effectiveness of combining **TESSERA embeddings** with **iNaturalist data** for habitat mapping.

# The Harvard-Emory ECG Database

- The **Harvard-Emory ECG Database (HEEDB) version 4.0** comprises over **10 million 12-lead ECG recordings** from nearly **2 million unique patients**, enhancing research on cardiac abnormalities and sleep-related health outcomes.

# Porting a library to a different language with a sentence

- **Slate autonomously ported the Browser Use library to TypeScript in under two hours, achieving a 1:1 conversion for less than $60.00.** This showcases the efficiency of coding agents in handling complex tasks with minimal human intervention.

# Cosmic simulations that once needed supercomputers now run on a laptop

- **Effort.jl** is a groundbreaking emulator that allows complex cosmic simulations to run on standard laptops, achieving results comparable to supercomputers in both speed and accuracy.

# [R] ShinkaEvolve: Towards Open-Ended And Sample-Efficient Program Evolution

- **ShinkaEvolve** is a cutting-edge, **open-source framework** for program optimization, designed for seamless integration into scientific codebases, enabling users to explore its capabilities directly in a remote Google Colab instance without installation. [Open source code](https://github.com/SakanaAI/ShinkaEvolve) \| [Technical report](https://arxiv.org/abs/2509.19349) \| [Blog](https://sakana.ai/shinka-evolve/) \| [Colab instance](https://colab.research.google.com/github/SakanaAI/ShinkaEvolve/blob/main/examples/shinka_tutorial.ipynb)

# [R] A 4-bit reasoning model outperforming full-precision models

- **Alpie Core**, a **32B model** fine-tuned in **4-bit precision**, achieves a **~75% VRAM reduction** compared to FP16 baselines, demonstrating that aggressive quantization can maintain strong reasoning capabilities.