# Jun 27, 2026

- **DeepSeek open-sources inference optimizations with 60–85% faster generation**

- The **DSpark paper** from the **DeepSpec** project presents innovative methodologies for enhancing deep learning model performance, particularly in the context of data efficiency and scalability.

- **U.S. allows Anthropic to release Mythos AI to ‘trusted’ US organizations**

- The **US government** has lifted its block on **Anthropic's Claude Mythos 5 AI model**, allowing access to over **100 US institutions**, marking a significant shift in regulatory stance amid previous export controls due to security concerns.

- **Ultrasound imaging of the brain**

- **Ultrasound imaging** of the brain offers a groundbreaking method to achieve **MRI-level detail** without invasive procedures, utilizing the relationship between blood flow and neuronal activity to create detailed vascular maps.

- **AI in mathematics is forcing big questions**

- **AI is transforming mathematics**, enabling machines to tackle complex problems and even produce publishable research, as demonstrated by Google DeepMind's Aletheia and OpenAI's recent breakthroughs in combinatorial geometry.

- **Show HN: Smart model routing directly in Claude, Codex and Cursor**

- **Workweave's Router** is a **drop-in proxy** that intelligently selects the best model for each request using a compact on-box embedder, enhancing efficiency across various AI platforms like Anthropic and OpenAI.

- **AI Is Designing Radio Chips That Humans Couldn't Even Imagine**

- **AI is revolutionizing RFIC design** by employing reinforcement learning and inverse design techniques, enabling the rapid creation of radio chips that outperform traditional designs in both performance and efficiency.

- **Modern GPU Programming for MLSys**

- **Modern GPU programming** is essential for optimizing machine learning systems, focusing on high-performance kernels like attention and GEMM, which are critical for AI workloads.

- **Asian AI startups launch Mythos-like models**

- **Asian AI startups** like 360 and Sakana AI have launched models, Tulongfeng and Fugu, respectively, to compete with **Anthropic’s Mythos**, amid ongoing U.S. export bans that limit access to advanced AI technologies.

- **Researchers have developed pixels that can emit and analyse light together**

- **ETH Zurich researchers have developed a groundbreaking pixel that can both create and analyze images**, potentially leading to devices that function as both cameras and displays simultaneously.

- **Show HN: Autofit2 – End-to-end pipeline for multilingual text classification**

- **autofit2** enables **few-shot text classification** with **95–99% precision** using a streamlined, automated pipeline that supports over **50 languages** through `setfit` and SBERT embeddings.

- **MathFormer: Testing whether symbolic math is pattern matching or reasoning**

- A **4M parameter seq2seq model** achieves ~ **98.6% accuracy** on symbolic math tasks, indicating it relies on **structural token transformations** rather than understanding operators or variables.

- **Showcase: geolocating a dashcam video without GPS, only from the footage**

- **Third Eye** is a project that enables **visual geolocation** of dashcam videos without GPS, utilizing image content to determine filming locations and map routes effectively.

- **Live Continual Learning in Machine Learning**

- **Live continual learning** addresses the challenge of _catastrophic forgetting_, a critical issue in machine learning where models lose previously learned information when new data is introduced.

- **Built an LLM training framework that actually runs on older GPUs without crashing**

- **Picotron** is a newly developed LLM training framework that eliminates **GPU-specific dependencies**, enabling it to run on older GPUs like T4 and V100 without crashing during import.

- **Previewing GPT-5.6 Sol: a next-generation model**