# Apr 3, 2026

## Daily Updates

- Google releases Gemma 4 open models
  - **Gemma 4** represents a breakthrough in **compute and memory efficiency**, enhancing intelligence for mobile and IoT devices, which allows for more sophisticated applications in constrained environments.

- Lemonade by AMD: a fast and open source local LLM server using GPU and NPU
  - **Lemonade** is an **open-source local AI** platform that supports text, images, and speech, designed for rapid deployment on any PC, emphasizing **privacy** and **speed**.

- New Rowhammer attacks give complete control of machines running Nvidia GPUs
  - **New Rowhammer attacks, GDDRHammer and GeForge, enable full control of machines with Nvidia GPUs by exploiting memory vulnerabilities, allowing attackers to manipulate CPU memory.** This marks a significant evolution in Rowhammer techniques, which have historically targeted CPU DRAM but now extend to GPU memory, demonstrating a serious security threat across platforms.

- ESP32-S31: 320MHz 2C RV32IMAFCP+CLIC, 512KB SRAM, GbE, 802.11ax, 61 GPIO
  - **ESP32-S31** is a **dual-core RISC-V SoC** featuring **Wi-Fi 6**, **Bluetooth 5.4**, and advanced **HMI capabilities**, designed for next-gen IoT applications like smart appliances and automation systems.

- We replaced RAG with a virtual filesystem for our AI documentation assistant
  - **ChromaFs** revolutionizes the assistant's interaction with documentation by creating a **virtual filesystem** that allows for instant access and efficient querying, reducing session creation time from **~46 seconds to ~100 milliseconds**.

- What Category Theory Teaches Us About DataFrames
  - **Category theory reveals that over 85% of pandas operations can be expressed through just 15 fundamental operators**, significantly simplifying the understanding of DataFrame operations and their relationships.

- What we learned building 100 API integrations with OpenCode
  - **Nango's background agent** can autonomously generate **~200 API integrations** across five platforms in **15 minutes** at a cost of less than **$20**, significantly reducing the time from a week for manual engineering.

- Show HN: TurboQuant for vector search – 2-4 bit compression
  - **TurboQuant** offers a **Rust-based** implementation for vector search, compressing high-dimensional vectors to **2-4 bits** per coordinate with **zero training** and **fast indexing** times, making it a compelling alternative to existing solutions like FAISS.

- [R] Is autoresearch really better than classic hyperparameter tuning?
  - **Autoresearch** outperforms **Optuna** in terms of speed, cost-efficiency, and generalization, demonstrating superior sample efficiency in experiments conducted on **NanoChat**.

- [P] PhAIL (phail.ai) – an open benchmark for robot AI on real hardware. Best model: 5% of human throughput, needs help every 4 minutes.
  - **PhAIL** (phail.ai) reveals that current robot AI models achieve only **5% of human throughput**, requiring human intervention every **4 minutes**, highlighting significant limitations in autonomous operation.

- Automatic Textbook Formalization
  - **RepoProver** is a **multi-agent framework** designed for the **large-scale formalization** of mathematics textbooks in Lean, utilizing LLM agents for tasks like translating definitions and filling proofs, ensuring collaborative efficiency through a git-based system.

- ContextBudget: Budget-Aware Context Management for Long-Horizon Search Agents
  - **Budget-Aware Context Management (BACM)** addresses the challenge of limited context size in LLM-based agents by framing context management as a sequential decision problem, allowing agents to optimize their interaction history within a defined budget.

- From RTX to Spark: NVIDIA Accelerates Gemma 4 for Local Agentic AI
  - **Gemma 4** models, optimized for **NVIDIA GPUs**, deliver **efficient local execution** across devices, enhancing capabilities in reasoning, coding, and multimodal interactions.