In 2025, venture capital can't pretend everything is fine any more

Venture capital is struggling, with the industry relying on a potential bubble in AI as its last hope, and even that is largely dependent on a single company, OpenAI. The latest Pitchbook-NVCA Venture Monitor report reveals a bleak picture, with venture capital investments down and few opportunities for returns outside of AI, which is being propped up by large deals like the $40 billion OpenAI investment led by Softbank.

Read more

Klarna changes its AI tune and again recruits humans for customer service

Klarna, a buy now, pay later firm, is shifting its customer service strategy to prioritize human interaction, after previously relying heavily on AI chatbots to handle customer inquiries. The company is now recruiting human customer service representatives to work alongside its AI technology, acknowledging that while AI can provide speed and efficiency, human empathy and expertise are essential for delivering high-quality customer service.

Read more

Avoiding AI is hard – but our freedom to opt out must be protected

As AI becomes increasingly integrated into daily life, from job applications to healthcare and finance, it is becoming clear that opting out of its influence is no simple matter, and those who choose to avoid it may face significant disadvantages. The growing reliance on AI is creating a divide between those who can navigate and benefit from these systems and those who are left behind, highlighting the need for policies that respect individual freedoms and provide the option to live without AI without facing exclusion or discrimination.

Read more

Continuous Thought Machines

Neurons in biological brains use timing and synchronization to compute, a property that is essential for flexibility and adaptability, but is often discarded in modern AI systems for efficiency and simplicity. Researchers have introduced the Continuous Thought Machine (CTM), a novel neural network architecture that incorporates neural timing as a foundational element, and have found surprising and encouraging results that may help bridge the gap between biological intelligence and modern AI.

Read more

Intellect-2 Release: The First 32B Model Trained Through Globally Distributed RL

INTELLECT-2 is a 32B parameter model trained using globally distributed reinforcement learning, marking a paradigm shift in decentralized training. The model was trained using a custom framework called PRIME-RL, which enables fully asynchronous reinforcement learning across a dynamic, heterogeneous swarm of permissionless compute contributors, and has been open-sourced along with its code and data.

Read more

Scoring the European Citizen in the AI Era

The AI Act bans social scoring, a practice inspired by China's Social Credit System, but this ban is unlikely to affect existing scoring practices in Europe, such as creditworthiness assessments, as they are considered "high-risk AI systems" subject to regulations. However, the ban is argued to be a flexible and useful tool to protect individuals against disproportionate use of AI-based scoring, similar to Article 22 of the General Data Protection Regulation.

ArXiv

AI Powered Energy Management Systems – Prospects and Challenges

Microgrids, a key solution for a sustainable energy future, face challenges such as forecasting renewable energy demand and protecting against cyberattacks, which can be addressed through the use of artificial intelligence (AI) in energy management systems (EMS). The integration of AI in microgrid EMS has shown immense potential for optimizing energy management, and future research directions include the development of self-healing microgrids, blockchain integration, and the use of Internet of Things (IoT) technology.

ArXiv

Absolute Zero: Reinforced Self-Play Reasoning with Zero Data

The Absolute Zero Reasoner (AZR) is a new reinforcement learning paradigm that enables a single model to propose and solve its own tasks, improving its reasoning capabilities without relying on external data or human supervision. AZR achieves state-of-the-art performance on coding and mathematical reasoning tasks, outperforming existing models that rely on large amounts of human-curated examples, and can be applied to different model scales and classes.

ArXiv

Codetations: Intelligent, Persistent Notes and UIs for Programs and Documents

Software developers often rely on memory to recall design decisions and other context about their code, which can be obstructive when working with unfamiliar code. Codetations is a system that helps developers contextualize documents with dynamic, interactive notes that stay outside the code and can improve large language model performance during code repair.

ArXiv

Sound, Precise, and Fast Abstract Interpretation with Tristate Numbers

The Extended Berkeley Packet Filter (BPF) uses a static analyzer to ensure the safety of user code running in the kernel, relying on abstract domains such as tnums to reason about bitwise uncertainty in program values. This paper formally specifies the tnum abstract domain, provides proofs of soundness and optimality for its arithmetic operators, and introduces a novel sound algorithm for tnum multiplication that has been merged into the Linux kernel.

ArXiv

Show HN: One-liner CLI for batched PDF-to-Markdown at $1 per ~6k pages

llm-food is a Python package that provides a FastAPI-based microservice for converting various input document formats into clean Markdown text, optimized for Large Language Model pipelines. The package includes a server, a Python client library, and a command-line interface, supporting formats such as PDF, DOC/DOCX, RTF, PPTX, and HTML, with features like asynchronous batch processing and task status tracking.

Repo

All-in-one AI marketing agent and copilot desktop app to cross-post to any sites

Jaaz is a free, local desktop app that serves as an all-in-one AI-powered marketing assistant, offering features such as AI content editing, 1-click cross-posting, and automated reply generation. The app is currently in early beta and available for macOS and Windows, with users able to join a waitlist for early access and provide feedback to shape the app's development.

Repo

Show HN: LLM Agents Play Among Us-Like Game

The Danganronpa Simulator is a text-based social deduction game where 12 AI agents, each with unique personalities and motivations, compete against each other in a game of lies, strategy, and survival. The game can be played online, and users can also host the game themselves by cloning the repository, installing dependencies, and setting up their own LLM, with options to use Google Gemini or other language models.

Repo

Scraipe: AI Scraping and Analysis Framework

Scraipe is a Python framework for scraping and analyzing data, featuring versatile scraping capabilities, LLM analysis, and workflow management, with a modular design and high-performance asynchronous IO-bound tasks. It can be installed using pip and provides a simple example workflow for scraping and analyzing text data from URLs, with detailed documentation and contribution guidelines available.

Repo

I built a native Windows Todo app in pure C (278 KB, no frameworks)

The Simple Todo application is a native Windows Todo app built with C and the Win32 API, featuring a modern GUI, system tray integration, and persistent storage. It allows users to create, edit, and delete todo items, mark tasks as complete, and set priority levels, all while maintaining a native Windows look and feel.

Repo