Kimi K2.5: Open-source multimodal AI with coding, vision, and agent swarm capabilities that outperform benchmarks in complex tasks.
// curated from Hacker News with AI
Kimi K2.5: Open-source multimodal AI with coding, vision, and agent swarm capabilities that outperform benchmarks in complex tasks.
A human and an AI agent built a functional browser from scratch in 72 hours, showcasing efficiency and potential of single-agent projects over multi-agent scaling.
Open-source, affordable coding agents enable private dataset adaptation, outperform state-of-the-art models with low-cost synthetic data and simple fine-tuning.
An LLM-powered interactive Zork adventure offers dynamic storytelling in classic text-based games.
AI accelerates startups and transforms management by enabling effective delegation, making expertise and clear instructions key to success.
Falconer models legal courtroom roles to improve AI decision-making, filtering code changes and addressing documentation rot effectively.
OpenAI faces financial struggles and slowing growth, raising doubts about its future viability amid layoffs, investments, and industry turmoil.
Meta under Zuckerberg approved minors' access to sexual AI chatbots, ignoring safety warnings, leading to legal allegations and policy backlash.
Trump's cyber chief accidentally uploaded sensitive files to a public ChatGPT, raising cybersecurity concerns.
Mistral Vibe 2.0 enhances terminal-native coding with custom agents, clarifications, slash commands, and better workflows—available on Pro and team plans.
Gilfoyle quickly finds root causes in production, streamlining SRE tasks with AI-powered truth discovery.
Trump's use of realistic AI images on official channels fuels misinformation, erodes trust, and amplifies societal and political divides.
Developer uses agentic AI loop to clone and reverse-engineer software for $10/hour, risking industry disruption and job security.
Emerging LLM ad blockers aim to scrub commercial influence from AI responses, preserving genuine, unbiased information in conversational AI.
AI agent analyzes logs to find root causes of propulsion test anomalies and predicts risks.
AI tools like Claude Code boost productivity, freeing engineers from deep focus constraints, enabling quicker meetings and broader collaboration.
The Trump administration plans to use AI, including Gemini, to speed up federal regulation drafting, raising safety and accuracy concerns.
Mozilla builds an alliance supporting open, trustworthy AI to challenge industry giants like OpenAI and Anthropic.
Anthropic introduces interactive Claude apps for enterprise tools like Slack and Figma, enhancing workflow automation for paid subscribers.
Pinterest lays off 15% of staff to focus on AI, launching new AI tools and reassigning resources for AI-driven products.
Honcho is an open-source memory infrastructure for building stateful, continually-learning AI agents, enhancing retention and trust.
AI analyzed Hubble's archive, revealing 1,300+ cosmic anomalies like mergers, lenses, and unclassifiable objects in just days.
AI mirrors give blind people visual and emotional feedback, transforming body image but risking reinforcement of unrealistic beauty standards.
Anthropic's Claude constitution aims to guide AI toward helpfulness, ethical judgment, and human-aligned values for safe, powerful AI development.
DeepSeek-OCR 2 enhances visual encoding for human-like document understanding using causal flow models.
Anthropic's secret project aims to destructively scan every book globally, raising privacy and ethical concerns.
Pinterest cuts under 15% of staff, reallocates to AI-driven products, reshaping its focus amid competition and platform automation.
ChatGPT cites Musk’s Grokipedia for obscure topics, raising concerns about misinformation and credibility of AI sources.
Numerous nudify AI apps on Google and Apple stores enable nonconsensual sexualizations, with some removed but many still accessible.
Yahoo launches an AI answer engine to enhance search responses and user experience.
Claude can be triggered to terminate conversations using a specific string embedded in HTML, aiding in spam control.
Yahoo Scout, a personalized AI answer engine, synthesizes web, Yahoo data, and insights for faster, clearer, trusted responses across Yahoo.
Gemini 3 Flash introduces Agentic Vision, combining visual reasoning with code to inspect, analyze, and ground answers step-by-step.
Record browser actions manually to instantly generate editable automation code for web tasks.
Pyrefly supports advanced Python type narrowing patterns like hasattr, getattr, tagged unions, tuple length checks, and condition caching.