FLUX.1 Kontext

Black Forest Labs has introduced FLUX.1 Kontext, a suite of generative flow matching models that enable text-and-image-driven generation and editing, allowing users to modify input images via simple text instructions. The FLUX.1 Kontext models offer capabilities such as character consistency, local editing, style reference, and interactive speed, making it possible to create and manipulate images with minimal latency and high image quality.

Read more

Google is using AI to censor independent websites like mine

The founder of Travel Lemming LLC, a website providing online travel advice, claims that Google censored their site, along with thousands of other small and independent web publishers, by releasing algorithm updates that decimated their traffic, and that this is part of Google's plan to use AI to control the flow of information online. Google's goal, according to the founder, is to monopolize not just the questions people search, but also the answers themselves, by using AI to consume and replace the open web, and creating an information cartel that lets it control the flow of information.

Read more

ClickHouse raises $350M Series C

ClickHouse, Inc. has raised $350 million in Series C financing, bringing its total funding to over $650 million, which will be used to scale product development, support global expansion, and deepen partnerships with customers and technology providers. The company, a leader in real-time analytics and data warehousing, has seen significant growth, with over 300% increase in the past year, and now serves over 2,000 customers, including major companies like Tesla, Sony, and Lyft, as it aims to become the default engine for next-generation intelligent data products.

Read more

Untrusted chatbot AI between you & the internet is a disaster waiting to happen

Adding an untrusted middleman, such as a chatbot, to your information diet and personal communications will inevitably lead to manipulation and disaster, as these entities prioritize their own interests and those of their creators over the user's. This can result in biased recommendations, ideological manipulation, and a loss of control over the information you receive, making it essential to be cautious when relying on these technologies.

Read more

Nobel Laureate Daron Acemoglu: Don't Believe the AI Hype

Daron Acemoglu argues that forecasts of extraordinary productivity benefits from recent advances in generative AI are unrealistic, and that economic theory and data suggest a more modest impact, with AI potentially increasing total factor productivity by only 0.66% over ten years. This translates to a potential increase in GDP growth of 1-1.5%, much smaller than the predictions made by companies like Goldman Sachs and McKinsey.

Read more

Superhuman performance of an LLM on the reasoning tasks of a physician

A large language model (LLM) was evaluated against hundreds of physicians on challenging clinical cases and demonstrated superhuman diagnostic and reasoning abilities in five experiments and a real-world emergency room study. The LLM's performance exceeded that of board-certified physicians at various diagnostic touchpoints, suggesting it has achieved superhuman performance on general medical diagnostic and management reasoning, and highlighting the need for further prospective trials.

ArXiv

Collaborative Agentic AI Needs Interoperability Across Ecosystems

Collaborative agentic AI has the potential to transform industries, but current isolated solutions are leading to fragmented ecosystems. The proposed "Web of Agents" framework aims to address this issue by establishing a minimal architectural foundation with four key components, enabling interoperability and paving the way for open, secure, and widely-adopted agentic ecosystems.

ArXiv

A Practical Deep Learning-Based Acoustic Side Channel Attack on Keyboards

Acoustic side channel attacks pose a significant threat to keyboard security, and a deep learning model using a smartphone's microphone can classify laptop keystrokes with high accuracy, up to 95%. The model's effectiveness was demonstrated using both nearby phone recordings and video-conferencing software, highlighting the need for mitigation methods to protect users against these types of attacks.

ArXiv

An Efficient Function Representation Without Neural Networks

Function approximation is crucial in computer graphics and other fields, but current neural network-based methods are often limited by their many parameters. A new framework and compact function representation using polynomials and radial basis functions have been proposed, achieving comparable or superior performance to state-of-the-art techniques with significantly fewer parameters.

ArXiv

Breaking the Sorting Barrier for Directed Single-Source Shortest Paths

A new algorithm for single-source shortest paths on directed graphs achieves a time complexity of $O(m\log^{2/3}n)$, surpassing the $O(m+n\log n)$ bound of Dijkstra's algorithm for sparse graphs. This breakthrough demonstrates that Dijkstra's algorithm is not optimal for solving the single-source shortest paths problem.

ArXiv

Show HN: Onlook – Open-source, visual-first Cursor for designers

Onlook is an open-source, visual-first code editor that allows designers to craft websites, prototypes, and designs with AI in Next.js + TailwindCSS, making edits directly in the browser DOM with a visual editor. The project is still under development and is looking for contributors to help make Onlook for Web an incredible prompt-to-build experience, with features such as creating Next.js apps in seconds, visually editing apps, and managing brand assets and tokens.

Repo

Show HN: Compliant-LLM: Audit AI Agents for Compliance with NIST AI RMF

Compliant LLM is a comprehensive toolkit for ensuring the compliance and security of AI systems, supporting multiple LLM providers and testing against various attack strategies and compliance frameworks. It offers features such as security testing, compliance analysis, provider support, and visual dashboards, and can be installed using pip and configured to access target LLMs.

Repo

Show HN: I compressed 10k PDFs into a 1.4GB video for LLM memory

Memvid is a lightweight AI memory solution that encodes text data into videos, enabling fast semantic search and sub-second retrieval times across millions of text chunks. It compresses knowledge bases into compact video files, providing 10x storage efficiency and instant access to information, making it suitable for various applications such as digital libraries, educational content, and corporate knowledge bases.

Repo

Cloi CLI: Local debugging agent that runs in your terminal

Cloi is a local, context-aware debugging agent that runs in the terminal, ensuring code and data remain private and secure, and can analyze errors and apply fixes with user permission. It offers features such as smart context retrieval, safe changes, and zero setup, and is free to use with an extensible architecture that allows for customization and contribution.

Repo

Show HN: Sokuji – Open-source real-time speech translation for Microsoft Teams

Sokuji is a desktop application that provides live speech translation using OpenAI's Realtime API, bridging language barriers in live conversations by capturing audio input, processing it, and delivering translated output in real-time. The application offers various features, including automatic turn detection, audio visualization, and virtual audio device management, and is also available as a browser extension for Chrome and Chromium-based browsers.

Repo