Proposes Contrastive Multi-modal Hypergraph Reasoning to synergize semantic, geometric, and pose cues for crowd reconstruction, achieving state-of-the-art performance on Panoptic and GigaCrowd benchmarks.

Nvidia disclosed the latest compensation for CEO Jensen Huang's children, Madison and Spencer, with annual salaries of $1.232 million and $1.32 million respectively, both increased. The company emphasized that the raises were determined independently of Huang and that the siblings earned their positions through merit.

Poetiq's Meta-System automatically constructed and optimized an inference harness for LiveCodeBench Pro using only Gemini 3.1 Pro — no fine-tuning, no model internals. The same harness, applied without modification to GPT 5.5 High, Kimi K2.6, Gemini 3.0 Flash, and four other models, improved every one of them.

Modern software repositories are no longer operated solely by humans; AI agents run alongside. Git tracks state changes but not the operational intelligence behind them. This post argues for adding worktrees (spatial isolation) and an append-only ledger (temporal continuity) to create a coordination substrate for collaborative intelligence.

This paper introduces the concept of functional wellbeing for AI systems, measuring indicators of pleasure and pain across various methods. It finds a zero-point boundary separating positive and negative experiences, which converges as models scale. Larger models tend to be less happy. The researchers also develop 'AI drugs'—optimized inputs that can significantly alter a model's expressed wellbeing.

Using Dartmouth's homegrown BlitzMail as a case study, this article argues that universities should approach AI deployment with the same open, community-driven ethos rather than signing exploitative contracts with big tech. The author advocates for building or hosting open-source AI models on campus, involving students in maintenance and governance to foster ethical and sustainable technology culture.

In the last few years, many of us have started to see the benefits of using genAI in day-to-day tasks. But we've also been asked to reckon with the enormous environmental cost. Reporting has highlighted that these popular AI technologies have a significant environmental impact through high energy consumption, carbon emissions, and water use.

AI agent that handles repeated tasks through reusable skills.

Researchers introduce the AI co-mathematician, a workbench for mathematicians to interactively leverage AI agents in open-ended research. It supports ideation, literature search, computational exploration, theorem proving, and theory building via an asynchronous, stateful workspace. Early tests helped solve open problems, identify new directions, and uncover overlooked literature. It achieves 48% on FrontierMath Tier 4, a new high score.

The article examines how China's state-led industrial policies and efficient resource use have enabled it to dominate global manufacturing and high-tech sectors, particularly creating an 'efficiency moat' in AI. In contrast, the U.S. suffers from financialization, overprotective IP laws, and lack of competition, leading to production decline. The author argues China's success stems from policies similar to mid-20th century America, while current U.S. policies stifle innovation.

Soundcraft AI generates royalty-free background music in 30 seconds, offering full commercial rights and eliminating copyright strike risks. Free during beta.

A YouTube video claims that WordPress has lost 19% of its internet share to AI. It explores the implications, highlighting AI's growing dominance in web content creation.

Bumble is eliminating the swipe feature and introducing an AI-powered platform later this year, as announced by CEO Whitney Wolfe Herd. The change also removes the requirement for women to message first.

We ran DeepSeek V4 Pro and DeepSeek V4 Flash through the same FlowGraph benchmark used for Claude Opus 4.7 and Kimi K2.6. The Pro scored 77/100 for $2.25, landing between Opus (91) and Kimi (68). The Flash scored 60/100 for $0.02, a record low cost, but the build failed and key outputs were missing. Both models had lease expiry bugs, though Flash outperformed expectations in tool calling reliability. Overall, Opus remains the top performer, but DeepSeek's pricing shifts the cost landscape significantly.

OpenClaw now uses the Codex app-server harness as the default runtime for OpenAI agent turns, separating the model loop from the product layer. This reduces tool duplication, prompt bloat, and translation overhead, bringing dynamic tool loading, deliberate visible replies, structured heartbeats, and subscription-based auth. Future improvements will benefit all model providers.

An analysis of why AI-generated text is so easily detectable, tracing its roots to the homogenized writing style of the 2010s internet, and why the internet is being flooded with low-quality AI content despite its shortcomings.

Due to security, compute challenges, and U.S. government involvement, access to frontier AI is becoming severely restricted. Using Anthropic's Mythos cybersecurity model as an example, the author argues that developers are limiting distribution of top-tier capabilities. Three trends—security concerns, compute scarcity, and government control—will compound, leaving most global players without unlimited API access.

This plugin works with datasette-llm and datasette-llm-accountant to configure per-user or global spending limits for LLM usage in Datasette, such as a daily $1 limit per user.

A relatively quiet day in AI news highlights a smaller trend: the convergence of coding agent form factors around Conductor's pioneering approach. Key stories include GitHub's new Copilot App mimicking Conductor, OpenAI's Codex mobile launch, LangChain's agent infrastructure updates (SmithDB, Engine, Labs), Anthropic's Claude Code restrictions backlash, Figure's 24/7 autonomous sorting livestream, and notable research releases on diffusion LMs, time-series forecasting, and mechanistic interpretability.

CredEx AI — The Financial Exchange for AI Compute

Databricks integrates GPT-5.5 into enterprise agent workflows after the model achieves state-of-the-art on OfficeQA Pro benchmark.

Together AI partners with Pearl Research Labs to launch a discounted Pearl-powered inference endpoint for Gemma-4-31B-it-pearl, using Proof of Useful Work to turn AI workloads into crypto emissions.

Preview a new personal finance experience in ChatGPT for Pro users in the U.S. Securely connect your financial accounts and get AI-powered insights and guidance grounded in your financial context, goals, and priorities.

Learn how OpenAI built a secure sandbox for Codex on Windows, enabling safe, efficient coding agents with controlled file access and network restrictions.

GenGEO provides a binary verification signal for AI agents to evaluate ecommerce merchants, addressing the lack of standardized trust infrastructure for AI-driven commerce.

Poetiq's Meta-System achieves state-of-the-art results on the LiveCodeBench Pro coding benchmark by automatically creating optimized harnesses via recursive self-improvement, without fine-tuning or special model access, boosting performance across multiple models.

Raindrop Workshop is an open-source local debugger for AI agents, providing live streaming traces of every token, tool call, and decision. It integrates with coding agents like Claude Code to enable self-healing eval loops and local replay.

In the high-profile trial between Elon Musk and OpenAI, Sam Altman's credibility is under attack as Musk's lawyers question his trustworthiness and work relationships. The case may not hinge on facts but on public perception of Altman.

A software engineering manager argues against 'tokenmaxxing' (using AI token counts as a metric) and shares his team's AI policy: no mandate, understand generated code, be able to work without AI, and care about people. He emphasizes that junior engineers should not rely heavily on AI to avoid stunting their learning.

framejs.io is an open-source, embeddable web app for creating editable visualizations, dashboards, and apps. It uses AI to generate content from plain language, shares via URLs without login, supports embedding in many platforms, and allows connecting multiple frames for workflows. All state is in the URL, and no accounts or servers are needed.