Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
SpaceXAI is reportedly losing top talent across coding, world models, and Grok voice. Rivals like Meta and Thinking Machines Lab are scooping up former staff. Elon Musk's culture of extreme work has led some staff to leave. Several of the exits could have been driven by a desire to cash out.
Microsoft signed a deal with OpenAI late April that amended the company's exclusive license to OpenAI models, freed OpenAI to sell on any other cloud, and removed the AGI clause that would have triggered changes to Microsoft's IP rights once OpenAI's board declared the threshold reached. Microsoft's IP license, a 27% stake worth roughly $135 billion, will be kept through 2032. Microsoft is reportedly looking to purchase Inception, a company that builds diffusion-based language models. It is interesting that Microsoft would spend $13 billion on a partner and then immediately start a shadow procurement process for a replacement.
Sponsored
Raw latency doesn't tell you if the answer was right. Learn the metric that actually matters in production.
Nvidia has announced a partnership with Ineffable Intelligence, a startup pursuing superintelligence that was founded in late 2025 by UCL professor and former lead of DeepMind's reinforcement learning team, David Silver.
Babuschkin, an xAI cofounder, is putting in $100 million of his own money into the company.
TLDR's Applied AI team is tasked with making every process at TLDR legible to code, runnable by anyone, and composable into larger workflows. Join a small, fast moving team using the latest AI tools with an unlimited token budget. Learn more [23].
OpenSquilla has introduced an open-source AI agent runtime designed to reduce unnecessary token spend by reusing context efficiently.
Datadog's Toto 2.0, a scalable time series forecasting model family, is now available on Hugging Face.
Codex is now available in the ChatGPT mobile app, enabling seamless remote access to ongoing work on laptops, devboxes, or remote environments.
Sponsored
Unwrap [4] is the AI-powered customer intelligence platform trusted at scale by STRIPE, SOUTHWEST AIRLINES, PERPLEXITY, and other customer-centric companies.
Anthropic launched Claude for Small Business, a package of connectors and workflows that embeds Claude into tools like QuickBooks, PayPal, HubSpot, Google Workspace, and Microsoft 365.
Recursive Superintelligence has raised more than $650 million at a valuation of more than $4 billion to build AI that can improve itself with little or no help from human developers. Its seven co-founders include notable researchers from many of the industry's leading AI companies. Many of these researchers specialize in AI systems that can run for long periods in pursuit of goals. While the researchers are bullish on the idea of AIs recursively improving themselves, the current technology is a long way from the point where humans can be removed from the loop.
Cerebras raised $5.5 billion in its US IPO at a market valuation of about $40 billion. This was the largest IPO this year so far. The IPO drew orders for more than 20 times the number of shares available. The offering was led by Morgan Stanley, Citigroup, Barclays, and UBS.
Anthropic has passed OpenAI in business adoption. More businesses used Anthropic than OpenAI in April. Anthropic has quadrupled business adoption over the last year, while OpenAI grew business adoption by only 0.3%. The pace of development in the AI industry is overriding the typical forces of vendor stickiness.
Superstar researchers at frontier labs can earn over a hundred times more than the average AI postdoc. Researcher quantity doesn't easily make up for quality in the field of AI. Even a 2x researcher can earn far more than the median because their contributions easily scale to billions of users. If they can add something that multiple 1x researchers can't, then it's worth paying a lot to capture it.
OpenAI detailed the engineering behind Codex's Windows sandbox, which constrained local commands, file access, and networking permissions while still allowing coding agents to operate effectively on developer machines.
Vercel analyzed seven months of AI Gateway production traffic spanning hundreds of models and over 200,000 teams. The report showed rapid growth in agentic workloads, increasing adoption of open-source models, and heavy multi-model routing in large-scale deployments.
Sponsored
Agents are a mess to manage. With Agent Foundry [13], you get a single place to govern agents like OpenClaw and Claude Code. Real enforcement at the agent level, deployed in your own environment or SentinelOne's for zero lock-in. Powered by Prompt Security from
@cline/sdk is an open-source framework for building agentic applications. It has a plugin architecture that makes it easy to customize, and the framework has all the features expected from agents, like checkpoints, web fetch, MCPs, cron jobs, subagents, and more. The SDK can be used to run agents from CI/CD pipelines, create automations for end-to-end workflows, or embed agents directly inside products.
PyTorch 2.12 shipped major infrastructure updates including faster CUDA eigendecomposition, a unified graph capture API, MX quantization export support, and fused Adagrad optimizers.
Krishna Rao, Anthropic's CFO, joined the company two years ago when run-rate revenue was about $250 million. It is now $30 billion. Rao helped raise around $75 billion and is responsible for the procurement and allocation of compute. This post links to an interview with Rao where he discusses compute, raising funds, pricing dynamics, how Anthropic's finance team uses Claude, Mythos, biotech and healthcare, and much more.
Microsoft's MDASH AI system uses more than 100 specialized AI agents to work together across multiple AI models to find real-world software vulnerabilities. A set of agents scans code for potential vulnerabilities, and then a separate group of agents debates whether each finding is real and exploitable. A final stage constructs proof-of-concept attacks to confirm the bugs exist. MDASH surpassed Anthropic's Mythos model on the CyberGym benchmark, a test that measures how well AI systems can reproduce real-world vulnerabilities.
Sponsored
Your usage changes fast - your spend doesn't have to. Archera offers insured cloud commitments on AWS / Azure / GCP; you get reservation
The model, which will be announced at Google's annual I/O conference on Tuesday, will roughly be on par with GPT-5.5.
AutoScientist helps models learn specific capabilities quickly by using an automated approach to conventional fine-tuning.
DeepSeek V4 Pro scored 77/100 on the FlowGraph spec for $2.25 and lands between Opus 4.7 and Kimi K2.6 in terms of performance.
Paid Claude plans will be able to claim a dedicated monthly credit for programmatic usage starting on June 15.
This post contains a video of Alex Wang's first interview since he started working with Meta.
Sponsored
AI security is on everyone's mind - but few have a roadmap they can stand behind.
Fast mode for Claude Opus 4.7 is now available in research preview in the API and Claude Code, and on Cursor, Emergent, Factory, v0, Warp, and Windsurf. Fast mode is currently opt-in, but it will eventually become the default. A link to join the waitlist for fast mode is available.
Meta's Muse Spark foundational model is now powering Meta AI across the company's services. The model enables faster voice responses, smarter shopping assistance, and real-time visual recognition through device cameras. The initial rollout targets users in the US and Canada.
Google and SpaceX were reportedly discussing orbital data centers as part of broader efforts to expand AI compute infrastructure beyond Earth-based facilities.
Inference workloads are more variable and less predictable than training workloads. This makes them a natural fit for serverless computing. However, serverless computing only works if new replicas can be spun up as fast as demand changes. This article looks at how Modal took AI inference server scaling from multiple kiloseconds to just tens of seconds.
The AI infrastructure boom has driven increased demand for analog and power semiconductors, notably benefiting Multilayer Ceramic Capacitors, amidst a past supply glut and competition. Companies like Texas Instruments and NXP Semiconductors are avoiding capacity expansion, focusing instead on raising prices and improving profitability. The semiconductor supply chain, previously supporting EV and solar industries, is now being leveraged for AI-related demand growth.
Parameter Golf attracted over 1,000 participants and 2,000 submissions focused on minimizing loss on a dataset within strict constraints. Participants leveraged a range of techniques, including careful tuning, quantization, and novel modeling ideas, with AI coding agents playing a significant role. This challenge revealed new talent and highlighted the evolving role of AI agents in research competitions.
Sponsored
With the ability to publish hundreds of CMS pages in a single click, operate at a global scale with seamless localization, and even host unified content across multiple domains, teams have never been able to ship faster. Trusted by companies like Miro, Bilt, and Perplexity
Cactus Needle is a 26M parameter Simple Attention Network distilled from Gemini 3.1 that can be finetuned locally on a Mac/PC. It runs on Cactus at 6,000 tokens per second prefill and 1,200 decode speed. The weights are fully open. Simple Attention Networks are aimed at redefining AI for consumer devices like phones, watches, and glasses.
OpenAI shared a Codex workflow for agents that iteratively review, repair, and validate outputs using structured feedback loops to improve reliability.
The article discusses using reinforcement learning to fine-tune 4B models as recursive language models (RLMs) for production, achieving efficient task-specific behavior at a lower cost. By training a shared policy for both parent and child RLMs, this approach maintains task performance and reduces the need for multiple models. In tests, this method matches the performance of larger models like Claude Sonnet 4.6 but operates with significantly reduced size and cost.
Researchers derived compression-aware neural scaling laws by training nearly 1,300 models, revealing how bytes per token affect compute allocation. This challenges the heuristic that scales models by 20 tokens per parameter, showing it's due to specific tokenizers. The study suggests scaling should use bytes, not tokens, for better compute efficiency across diverse languages.
Google introduced new Gemini-powered Android features that can complete actions across apps, browse the web, fill out forms, and generate custom widgets through natural language prompts.
Today's LLMs may be commercially valuable, but predicting text alone will not lead to human-level intelligence because language is only a small fraction of how humans understand the world. Future AI systems will instead rely on “world models” that learn abstract representations of physics, causality, and consequences, enabling planning, reasoning, and adaptation in real-world environments like robotics, healthcare, factories, and industrial systems.
Sponsored
Granola is the call recorder works locally on Mac and iPhone, no awkward meeting bot required. Take shorthand notes - AI makes them
The Qwen team released Qwen-Image-2.0, their latest multimodal image generation model, showing improved typography, instruction following, photorealism, and long-text rendering across generation and editing tasks.