Measuring the impact of AI on experienced open-source developer productivity

• Early-2025 AI significantly enhances experienced open-source developer productivity, with metrics indicating a 20% increase in code efficiency and collaboration speed.

• Not So Fast: AI Coding Tools Can Reduce Productivity

  • A recent METR study reveals that AI coding tools can lead to a 19% decrease in productivity among experienced developers, contradicting their self-reported belief of a 20% increase in efficiency.

• Grok 4

  • Grok 4 features 256,000 context length and excels in reasoning tasks, outperforming competitors in key benchmarks, as indicated by an Artificial Analysis Intelligence Index of 73.

• ETH Zurich and EPFL to release a LLM developed on public infrastructure

  • ETH Zurich and EPFL are set to release a fully open large language model (LLM) in late summer 2025, developed on the “Alps” supercomputer, emphasizing transparency and multilingual capabilities across over 1,000 languages.

• LLM Inference Handbook

  • LLM Inference in Production serves as a comprehensive resource, consolidating fragmented knowledge on LLM inference, including core concepts and performance metrics like Time to First Token and Tokens per Second.

• Show HN: Cactus – Ollama for Smartphones

  • Cactus is a cross-platform framework that enables the deployment of LLMs, VLMs, and TTS models directly on smartphones, enhancing privacy and supporting real-time inference without latency.

• Show HN: Cactus – Ollama for Smartphones

  • Cactus is a cross-platform framework enabling local deployment of LLM, VLM, and TTS models in Flutter and React-Native applications, supporting a wide range of models from Huggingface.

• How to prove false statements: Practical attacks on Fiat-Shamir

  • Computer scientists have developed a method to exploit the Fiat-Shamir transformation, allowing them to prove false statements within blockchain and cryptographic systems, fundamentally challenging the security of these technologies.

• AI Agent Benchmarks Are Broken

  • Current AI agent benchmarks are unreliable, with up to 100% misestimation of agents' capabilities due to flawed evaluation methods, as evidenced by the WebArena benchmark incorrectly validating an agent's answer to a simple math problem.

• [D] Views on DIfferentiable Physics

  • Differentiable Physics offers a more reliable approach to simulations by combining traditional numerical methods with autodiff, addressing issues of reproducibility and weak baselines prevalent in the Scientific ML community.

• [P] PrintGuard - SOTA Open-Source 3D print failure detection model

  • PrintGuard is a new open-source 3D print failure detection model that significantly enhances real-time performance, achieving 15 FPS on a Raspberry Pi 4b, which is over 40 times faster than existing solutions like Obico’s “Spaghetti Detective.”

• Researchers create 3D interactive digital room from simple video

  • Cornell researchers have developed an AI-driven method to create interactive 3D simulations from simple videos, enabling users to engage with digital twins of real spaces, enhancing applications in gaming and robotics training.

• SEO Is Dead. Long Live Geo

  • SEO is being overshadowed by GEO, as evidenced by Google's AI Overviews now appearing in over 50% of searches, reflecting a significant shift in how users interact with search engines and AI technologies.

• [D] Modelling continuous non-Gaussian distributions?

  • Modeling non-Gaussian labels requires specialized approaches, such as density mixture networks for bimodal distributions and hurdle models for zero-inflated data, to effectively capture the underlying patterns in sparse ratings.

• How AI will accelerate biomedical research and discovery

  • AI is revolutionizing biomedical research by enhancing drug discovery processes, particularly in identifying targets for complex diseases like ALS, through advanced machine learning techniques that analyze vast datasets for biological patterns.