# Jun 20, 2026

## Project Valhalla, Explained: How a Decade of Work Arrives in JDK 28
- **Project Valhalla introduces JEP 401: Value Classes and Objects** into JDK 28, allowing developers to create classes that behave like primitives while maintaining readability and safety, with over **197,000 lines of code** added to the OpenJDK repository.

## GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2
- **AI labs are shifting away from larger models**, as evidenced by the **US ban on Claude Fable 5**, highlighting the risks associated with high parameter counts and the need for a more cautious approach to AI development.

## To study how chips work, MIT researchers built their own operating system
- **Fractal**, a new operating system kernel developed by MIT, provides unprecedented insights into processor behavior, revealing previously unknown vulnerabilities in Apple’s M1 chip, including evidence of **Phantom speculation** attacks.

## Building a robotics research setup that lives next to my desk
- **Robotics research is now feasible for individuals** with setups costing under **€5,000**, thanks to affordable hardware and accessible foundation models like Hugging Face’s [LeRobot](https://huggingface.co/lerobot), enabling meaningful experimentation on real hardware.

## DVD-JEPA: an open-source, fully-reproducible JEPA world model 
- **DVD-JEPA** demonstrates a novel approach to world modeling by predicting the **32-dimensional representation** of future states rather than pixel-by-pixel, allowing for efficient learning from video data.

## How does torch.compile() achieve massive speedups despite highly optimized NumPy functions? 
- **torch.compile** achieves **massive speedups** through **operator fusion**, which optimizes the execution of multiple operations into a single step, enhancing performance beyond that of highly optimized **NumPy functions**.

## An open handbook on LLM inference at scale (GPU internals, KV cache, batching, vLLM/SGLang/TensorRT-LLM) 
- The handbook delves into **GPU execution** and **memory internals**, revealing that **GPUs often remain idle** during inference due to memory hierarchy constraints and throughput bottlenecks, with visual aids enhancing comprehension.

## Built a Global AQ (PM2.5) Forecaster ML Model 
- The **end-to-end Air Quality (PM2.5) forecasting pipeline** utilizes over **1.6M rows** of OpenAQ and NASA weather data, achieving a **MASE below 1.0** globally by employing a **Horizon aligned architecture** to decouple forecasting horizons and mitigate variance issues.

## Time Series Modeling Needs a Dynamical Systems Perspective