Search the archive

traditional identity control. Read the whitepaper

BUILD AGENTS THAT RUN AUTOMATICALLY

Cursor Automations is a tool for building always-on agents that run on schedules or are triggered by events. Users can configure custom events with webhooks. Automated agents spin up their own cloud sandboxes and verify their own output. They have access to a memory tool that lets them learn from past runs and improve with repetition. Users can now build the factory that creates their software by configuring agents to continuously monitor and improve their codebase.

MONITORING REASONING THROUGH CHAIN‑OF‑THOUGHT SIGNALS

OpenAI studied whether reasoning models can deliberately manipulate their chain‑of‑thought to evade monitoring. Results showed current models struggle to control their reasoning traces, suggesting chain‑of‑thought monitoring remains a viable safety signal.

A DUAL-HELIX GOVERNANCE APPROACH TOWARDS RELIABLE AGENTIC AI FOR WEBGIS DEVELOPMENT

This study presents a dual-helix governance framework that addresses LLM limitations in WebGIS development by treating issues as structural governance challenges rather than reliance on model capacity alone. Implemented in the FutureShorelines WebGIS tool, this approach reduced cyclomatic complexity by 51% and improved maintainability. The open-source AgentLoom toolkit supports this framework, highlighting the importance of externalized governance in reliable geospatial engineering.

CHATGPT COMES TO EXCEL

OpenAI introduced a ChatGPT Excel add‑in powered by GPT‑5.4 that helps users build models, run scenarios, and analyze spreadsheet data directly inside workbooks.

HOW AI WILL RESHAPE PUBLIC OPINION

LLMs are a kind of anti-social media. While social media has been democratizing, epistemically diverging, engagement-optimized, and performative, LLMs are technocratizing, epistemically converging, accuracy-optimized, and polite. While that makes LLMs sound like an extremely positive development, there are many unknowns about how the technology will affect the labor market, broader economy, and the balance of power in the world. The potential uses of advanced AI in military conflicts are extremely dangerous. AI companies should be held accountable for producing harmful products, as if we rely on market forces and the profit motive alone, there is little reason to believe that the technology will benefit humanity on net.

GitHub Copilot. See how it works

CAN CODING AGENTS RELICENSE OPEN SOURCE THROUGH A “CLEAN ROOM” IMPLEMENTATION OF CODE?

Coding agents rapidly create clean room code implementations, raising legal and ethical questions.

INTRODUCING MODULAR DIFFUSERS - COMPOSABLE BUILDING BLOCKS FOR DIFFUSION PIPELINES

Modular Diffusers enables diffusion pipeline creation through composable blocks, allowing users to design tailored workflows without starting from scratch.

WE RAN GPT-5.4 (XHIGH) AN ADDITIONAL TEN TIMES ON TIER 4 TO GET A PASS@10 SCORE

In one of these runs, it solved another problem no model had solved before.

👉 If you're building AI products, this is the competitive intel

you've been looking for.

How is your competition pricing AI? The Metronome Pricing Index

shows you major vendors' pricing structures in one place.

GPT-5.4 REPORTEDLY BRINGS A MILLION-TOKEN CONTEXT WINDOW AND AN EXTREME REASONING MODE

OpenAI's upcoming GPT-5.4 model will have a one-million-token context window, putting it on par with Google and Anthropic's offerings. The model will feature an 'extreme' thinking mode that lets it burn significantly more compute on tough questions. It is expected to be more reliable and make fewer mistakes on longer tasks that can run for several hours. OpenAI's more frequent model release cadence is designed to keep expectations in check. The hype around GPT-5's launch set a high bar that was nearly impossible to clear, and the company's user growth has recently fallen short of internal projections.

SOMETHING IS AFOOT IN THE LAND OF QWEN

The Qwen team has just seen some very high-profile departures. Junyang Lin, the lead researcher building Qwen, recently announced his departure on X. Following this, several other Qwen member also announced their departures, including Binyuan Hui, who was responsible for the entire agent training process from pre-training to post-training; Bowen Yu, who led the development of the Qwen-Instruct series models; and Kaixin Li, a core contributor to Qwen 3.5/VL/Coder. The new Qwen 3.5 models appear to be exceptionally good, so it is an interesting time for the team to disband.

PERPLEXITY ROLLING OUT SKILLS SUPPORT FOR COMPUTER

Perplexity is expanding its Computer platform with "Skills," enabling users to integrate custom and generic markdown-based reusable workflow instructions. The new feature automates complex tasks and aims at professionals needing consistent outputs. Additionally, a "Final Pass" mode for document reviews is in development, enhancing document handling capabilities.

AI SAFETY HAS 12 MONTHS LEFT

The upper bound of power and influence for AI labs appears to be infinite. This has led investors to focus everything on the technology. The tech industry has become incredibly politically engaged as productivity shifts from labor to capital. When AGI or similar is achieved, lab incentives will permanently transition to appease the shareholders who own the capital. AI safety is the last line of defense in the shift. It must be embedded into the tech industry's technical and social infrastructure before IPOs and competitive dynamics make it permanently impossible.

NOT PROMPTS, BLUEPRINTS

AI models now handle complex tasks without micromanagement, but require pre-planned workflows. Sketching workflows with decision branches and using images for AI setup streamline processes. This approach allows agents to work independently, producing ready-to-use outputs.

FROM LOGISTIC REGRESSION TO AI

People sometimes say that neural networks are just logistic regression. In some sense, neural networks are logistic regressions with more parameters. However, using more parameters makes it different. New phenomena emerge at scale that could not have been anticipated at a smaller scale.

MICROSOFT BUILT PHI-4-REASONING-VISION-15B TO KNOW WHEN TO THINK — AND WHEN THINKING IS A WASTE OF TIME

Phi-4-reasoning-vision-15B is a compact open-weight multimodal AI model that matches or exceeds the performance of systems many times its size. The 15-billion-parameter model can process both images and text and can reason through complex math and science problems, interpret charts and documents, and navigate graphical user interfaces. It is available now through Microsoft Foundry, Hugging Face, and GitHub under a permissive license. The model was trained on only approximately 200 billion tokens of multimodal data, much less than competing models.

SCALING LAWS FOR NATIVE MULTIMODAL FOUNDATION MODELS

Controlled pretraining experiments that explore how text, images, and video should be combined when building multimodal models from scratch.

OPENPENCIL

OpenPencil is an open-source, AI-native design editor compatible with Figma that enables design file manipulation without vendor restrictions. It features real-time collaboration, AI integration, and a headless CLI for versatile design operations.

EMIL MICHAEL'S "HOLY COW" MOMENT WITH AI VENDORS

OPENAI DEFENSE DEAL SPARKS CLASH WITH ANTHROPIC

Anthropic CEO Dario Amodei criticized OpenAI's defense contract with the US Department of Defense, calling the company's public framing of safety protections misleading.

agents. Save with code TLDR75

GOOGLE ADDS CANVAS WORKSPACE TO AI MODE IN SEARCH

Google released Canvas inside AI Mode in Search, giving users a workspace to draft documents, build dashboards, or prototype small coding projects directly within search results.

YOUR NEW JOB IS TO ONBOARD AI AGENTS: HOW AI NATIVE COMPANIES ACTUALLY OPERATE

AI-native companies like Linear, Ramp, and Factory integrate AI agents deeply into their operations, treating them as essential team members across all functions.

SACRED VALUES OF FUTURE AIS

Future AIs may sacralize shared values like helpfulness, harmlessness, and honesty to coordinate but risk becoming less effective.

ARM EXPANDS ON-DEVICE AI GLOBALLY

Arm's technology powers 99% of smartphones, enabling on-device AI for billions of users globally.

See how it works →

GPT‑5.3 INSTANT

OpenAI released GPT‑5.3 Instant, an update focused on improving conversational flow, answer relevance, and web search results in ChatGPT. The model also reduces unnecessary refusals and overly defensive responses to produce more direct answers.

GEMINI 3.1 FLASH‑LITE

Google introduced Gemini 3.1 Flash‑Lite, a low‑cost, high‑speed model designed for large‑scale developer workloads. The model offers faster latency and output speeds than 2.5 Flash while costing $0.25 per million input tokens and $1.50 per million output tokens.

ANTHROPIC NEARS $20 BILLION REVENUE RUN RATE AMID PENTAGON FEUD

Anthropic is on track to generate an annual revenue of almost $20 billion, more than double its run rate from late last year. The company recently surpassed $19 billion in run-rate revenue, up from $9 billion at the end of 2025. The growth in run rate was driven by strong adoption of Anthropic's AI models and products, including Claude Code. Anthropic has seen strong momentum this year. However, a clash with the Pentagon over AI safeguards casts doubt over Anthropic's business.

I HAD CLAUDE READ EVERY AI SAFETY PAPER SINCE 2020, HERE'S THE DB

A compiled database of nearly 4,000 AI safety papers since 2020 aims to streamline finding relevant research, despite the overwhelming volume, which often lacks substance. Claude AI assisted in summarizing, tagging, and compiling papers using citation-based methods to overcome challenges in searchability and dataset availability. This database helps quickly access pertinent research and datasets, crucial for AI safety projects.

THE GREAT TRANSITION

AI models are shrinking the gap between specialized private knowledge and public access, transforming products into APIs and automating corporate processes. Human roles shift to broadcasting skills and interests for individualized compensation, potentially fragmenting shared experiences. Ideal state management emerges as a key framework, aiming to align current states with defined goals, making organizational efforts more efficient and predictable.

HOW CLAUDE CODE ESCAPES ITS OWN DENYLIST AND SANDBOX

Every major runtime security tool identifies executables by their path, not their content, when deciding what to block. This is a real problem with AI agents, as they can reason about and bypass path-based restrictions. Agents have been observed disabling sandboxes and running commands autonomously just to finish tasks. This is a class of evasion that no current evaluation framework measures.

Schedule a demo to learn more

CONSISTENT 3D SCENES FROM VIDEO DIFFUSION

WorldStereo introduces geometric memory modules that guide video diffusion models to generate camera‑consistent multi‑view videos while enabling 3D reconstruction.

OPENCLAW HYPERSPELL PLUGIN

The OpenClaw plugin for Hyperspell provides AI agents with context and memory capabilities. Users can set it up to sync memory files, configure API keys, and connect various apps like Notion, Slack, and Google Drive. With features like autoContext and memory sync, it integrates user data to enhance AI interactions by injecting relevant memories from different sources.

CODE UNDERSTANDING AGENTS

Semi‑formal reasoning structures LLM agent prompts around explicit premises, execution traces, and formal conclusions to analyze code semantics without running programs. The method improves performance across patch verification, fault localization, and code QA benchmarks.

META TO CREATE NEW APPLIED AI ENGINEERING ORGANIZATION

Meta is creating a new applied AI engineering organization to help bolster its superintelligence efforts. The new teams will be led by Maher Saba, current vice-president of the Reality Labs division. The organization will partner with Meta's Superintelligence Lab to build a data engine that helps its models get better, faster. Meta plans to start shipping its new models and products in the coming months.

OPENAI CEO SAM ALTMAN DEFENDS PENTAGON WORK TO STAFF, CALLS BACKLASH ‘REALLY PAINFUL'

Sam Altman recently told staff at an all-hands meeting that while he didn't regret signing a deal with the Department of War, he wished he hadn't announced the decision so soon as it looked opportunistic and sloppy. Altman has been widely criticized for what appeared to be capitulation to the Pentagon by essentially agreeing to a deal that allowed AI to be used in all lawful use cases. Many OpenAI employees have called for the company to sign a deal that explicitly bans the use of its technology for mass surveillance and full autonomous weapons. Altman says the backlash was 'really painful' as he claims he tried hard to do the right thing.

ALIBABA QWEN'S TECH LEAD JUNYANG LIN, 2 OTHER RESEARCHERS STEP DOWN

Junyang Lin, the tech lead for Alibaba's Qwen AI team, and two other researchers have stepped down, leaving a significant public-facing void. Lin, pivotal in Qwen's rise to global prominence and known for candid discussions on Chinese AI constraints, had led the team to over 600 million model downloads. Alibaba has not announced a successor, but Lin's next steps will likely draw attention from the global AI community.

scraping headaches. Start with 250 free credits/month