CVE program faces swift end after DHS fails to renew contract [fixed]

The Common Vulnerabilities and Exposures (CVE) program, which tracks and organizes computer vulnerabilities, was set to shut down on April 16, 2025, after the Department of Homeland Security (DHS) did not renew its funding contract. However, the Cybersecurity and Infrastructure Security Agency (CISA) stepped in and executed an 11-month contract extension, averting a shutdown and ensuring the continuation of critical CVE services.

Read more

JetBrains IDEs Go AI: Coding Agent, Smarter Assistance, Free Tier

JetBrains has made all its AI tools, including the improved AI Assistant and new coding agent Junie, available within its IDEs under a single subscription with a free tier. This integration aims to redefine the developer experience in the AI era, allowing developers to focus on complex tasks while AI handles basic coding tasks, with a strong focus on privacy, security, and transparency.

Read more

AI Is Facing a Crisis

The production and use of Artificial Intelligence systems require significant amounts of water and electricity, with examples including 500ml of water to write a 100-word email and 50 GWh of electricity to train a model like ChatGPT-4. To help mitigate this issue, individuals are encouraged to reduce their own consumption of these resources and raise awareness about the environmental impact of AI through campaigns like #SavetheAI.

Read more

We're Raising Kids to Prefer AI over People–and No One's Noticing

The provided text appears to be a collection of articles and posts from various authors on the Substack platform, covering a wide range of topics including technology, politics, culture, and personal development. The text includes excerpts from different authors, each with their own unique perspective and style, and invites readers to subscribe to their newsletters and join the discussion. Overall, the text showcases the diversity and creativity of the Substack community, with authors sharing their thoughts and ideas on various subjects.

Read more

Can LLMs earn $1M from real freelance coding work?

Researchers from OpenAI evaluated the performance of frontier AI models on real-world software development tasks, collecting over 1,400 freelance jobs worth over $1 million, and found that even the best-performing model earned approximately $403,000, solving only 33.7% of all tasks. The study suggests that AI models are better at picking the best solutions than creating them, and that current limitations may be computational rather than fundamental, providing a benchmark to measure AI progress and understand its current capabilities and impact.

Read more

No Free Lunch with Guardrails: Evaluating LLM Safety Tradeoffs

The adoption of large language models and generative AI has led to the development of guardrails to ensure their safe use, but these measures often involve tradeoffs between security and usability. Researchers have evaluated various industry guardrails and language models, confirming that strengthening security often comes at the cost of usability, and propose a blueprint for designing better guardrails that minimize risk while maintaining usability.

ArXiv

Epistemic Alignment: A Mediating Framework for User-LLM Knowledge Delivery

Current interfaces for large language models (LLMs) lack a structured way for users to specify how they want information presented, leading to workarounds and unverified methods. The proposed Epistemic Alignment Framework addresses this issue by providing a set of challenges and a common vocabulary to bridge the gap between user needs and system capabilities, offering guidance for AI developers and more personalized information delivery for users.

ArXiv

HybridRAG: Integrating Knowledge Graphs and Vector RAG

Large language models face challenges when extracting information from unstructured financial text data, such as earnings call transcripts, due to domain-specific terminology and complex formats. A novel approach called HybridRAG, which combines Knowledge Graphs and VectorRAG techniques, has been shown to outperform traditional methods in generating accurate and contextually relevant answers from financial documents.

ArXiv

What Should We Engineer in Prompts:Training Humans in Requirement-Driven LLM Use

The Requirement-Oriented Prompt Engineering (ROPE) paradigm is introduced to improve the process of prompting large language models (LLMs) by focusing on generating clear and complete requirements. ROPE has been shown to be effective, outperforming conventional prompt engineering training in an experiment and demonstrating a direct correlation between the quality of input requirements and LLM outputs.

ArXiv

From Text to Graph: Leveraging Graph Neural Networks for Explainability in NLP

Transformer-type models, particularly generative models, have become the go-to choice for natural language processing tasks due to their high versatility and outstanding results, but their large size and token-based input interpretation make them computationally expensive and difficult to explain. This study proposes a novel approach to achieve explainability by converting sentences into graphs that maintain semantic meaning, allowing for the identification of critical components in text structure and enabling a deeper understanding of how the model associates elements with the explained task.

ArXiv

Show HN: Plandex v2 – open source AI coding agent for large projects and tasks

Plandex is a terminal-based AI development tool that can plan and execute large coding tasks, handling up to 2M tokens of context and indexing directories with 20M tokens or more. It offers a range of features, including cumulative diff review, command execution control, and the ability to combine models from multiple providers, making it a powerful tool for developers working on complex projects.

Repo

Show HN: We Put Chromium on a Unikernel (OSS Apache 2.0)

Kernel provides deployment-ready, sandboxed Chrome browser environments for automated workflows, allowing users to run browser-based tasks, develop and test AI agents, and build custom tools. The platform offers pre-configured Chrome browsers with GUI access, support for Chrome DevTools-based frameworks, and the option to use either Docker or unikernel implementations for added benefits like automated standby mode and fast restarts.

Repo

Show HN: I built a deep learning engine from scratch in Python

Dolphin is a minimalist, zero-dependency machine learning system that implements deep learning architecture and training logic from scratch in Python, without relying on external libraries like NumPy or PyTorch. It provides a transparent and flexible platform for understanding transformers, automatic differentiation, and the inner workings of modern ML, making it ideal for teaching, learning, and experimenting with symbolic tensor operations.

Repo

Anubis weighs the soul of incoming requests to stop AI crawlers

Anubis is a tool that protects websites from scraper bots by using a sha256 proof-of-work challenge, which may prevent some search engines from indexing the site. It's intended as a last resort for sites that can't use Cloudflare or other protection methods, and can be tried out by connecting to the Anubis test site.

Repo

LlmTornado – The .NET library to consume 100 LLM APIs

LLM Tornado is a .NET library that allows developers to consume over 100 APIs, including OpenAI, Anthropic, and Google, making it easy to switch between different large language models. The library provides features such as API harmonization, vendor extensions, and easy-to-grasp primitives for building Agentic systems, Chatbots, and RAG-based applications, enabling developers to create rich and resilient applications with minimal vendor lock-in.

Repo