Product
Alibaba has released the Qwen3.5 family of multimodal LLMs, ranging from 0.8B to 397B parameters, featuring enhanced reasoning, multilingual support, and 256K context across 201 languages. The Qwen3.5-397B-A17B model is positioned in the same performance tier as Gemini 3 Pro and Claude Opus 4.5, with Unsloth providing state-of-the-art quantization to enable efficient inference on various hardware.
Sources: 1 Covered by Hacker News
News
The surge in demand for high-bandwidth memory (HBM) for AI accelerators is causing a global DRAM supply shortage, leading memory manufacturers to prioritize HBM production. This shortage has affected Apple, resulting in the removal of the 512GB RAM option for its M3 Ultra Mac Studio and price increases for other configurations.
Sources: 1 Covered by Hacker News
Product
Anthropic and OpenAI are providing free access to their AI platforms, Claude Max and ChatGPT Pro with Codex respectively, for maintainers of popular open-source projects. This initiative aims to support the open-source ecosystem by granting access to advanced AI tools.
Sources: via Simon Willison
Product
The `uv` package manager by astral-sh has added support for AMD ROCm 7.0 and 7.1 accelerators in its latest release (v0.9.27). This enhancement allows AI and machine learning developers to more effectively utilize AMD GPUs for compute-intensive tasks within Python environments.
Sources: via Hacker News
Research
Researchers have introduced SWE-CI, a new benchmark for evaluating the long-term maintainability of LLM-powered agents in software development. This repository-level benchmark assesses how AI agents sustain code quality over extended periods, moving beyond static bug fixing to dynamic, continuous integration scenarios.
Sources: via Hacker News
Research
Andrej Karpathy has launched "autoresearch," a project enabling an AI agent to autonomously modify and improve a small LLM training setup. The agent iterates on its own code and evaluates results over fixed time budgets, aiming to automate and accelerate AI research's experimental phase.
Sources: via Hacker News
News
The creator of WigglyPaint discovered fake websites and mobile apps, generated with the help of LLMs, that mislead users and charge for his free tool. This highlights the misuse of LLMs for fraudulent activities and intellectual property infringement, prompting criticism of LLMs as tools for plagiarism and exploitation.
Sources: via Hacker News
Research
Recent demonstrations of Elon Musk's Optimus robots, such as pouring beer, have revealed that they are currently human-controlled rather than autonomous. This has sparked discussion about their actual capabilities.
Sources: via Hacker News
Product
Eyot is a new programming language designed to simplify GPU programming, making it as straightforward as using CPU threads. The language transparently compiles code for both processors and manages communication, aiming to reduce the complexity of leveraging GPUs for AI, numerical analysis, and game development.
Sources: via Hacker News
Product
A new system utilizes a camera to monitor a user's pulse and detect other presences, while prioritizing privacy by only sharing heart rate data and keeping visual input private. This showcases an application of computer vision and potentially AI for biometric and environmental sensing.
Sources: via Hacker News