Search the archive

MEET YOUR NEW AI CODING TEAMMATE: GEMINI CLI GITHUB ACTIONS

Gemini CLI GitHub Actions is an AI coding teammate that acts as both an autonomous agent for critical routine coding tasks and an on-demand collaborator to quickly delegate work to. It can be triggered by events like new issues or pull requests. Gemini CLI GitHub Actions works asynchronously in the background, using the full context of a project to automatically handle tasks. It is now in beta and available to everyone worldwide.

INTRODUCING AI GROUNDING WITH BRAVE SEARCH API, PROVIDING ENHANCED SEARCH PERFORMANCE IN AI APPLICATIONS

Brave has launched AI Grounding with the Brave Search API, an all-in-one solution for connecting AI system outputs to verifiable data sources. AI Grounding anchors responses from large language models to factual information from verifiable web sources, reducing hallucinations and increasing appropriate responses to nuanced inputs. The AI Grounding plan is priced at $4 per thousand web searches plus $5 per million tokens (input and output). It can be accessed via the Pro AI plan with no change in price.

CODE INDEX MCP

Code Index MCP is a Model Context Protocol (MCP) server that helps large language models index, search, and analyze code repositories with minimal setup. It transforms how AI understands codebases with advanced search, analysis, and navigation capabilities. Code Index MCP is perfect for code review, refactoring, documentation generation, debugging assistance, and architectural analysis. It has support for a range of mainstream languages, mobile development platforms, web frontends, databases, and systems.

XBAI O4 MODEL RELEASE

MetaStone AI released XBai o4, its fourth-generation open-source model, which now outperforms OpenAI-o3-mini in Medium mode. The model focuses on complex reasoning tasks and introduces a scalable, parallel test-time inference framework.

FEATURE CORRELATIONS AND MEMORY CAPACITY IN DAMS

WHAT WE'RE OPTIMIZING CHATGPT FOR

ChatGPT will now have session break reminders. OpenAI is consulting physicians and mental health experts to improve model behavior. It emphasized that its focus is on helping users accomplish their goals rather than maximizing time on the platform or model agreeableness.

FOR THE FIRST TIME, OPENAI MODELS ARE AVAILABLE ON AWS

Before the release of its open source models, OpenAI models were only available on Azure. AWS has been losing AI cloud market share to Microsoft due to this exclusivity deal, but this new partnership signals OpenAI's shifting relationship with Microsoft as it seeks additional capacity beyond what Azure alone can provide.

Subscribe now.

AI FOR CYCLONE FORECASTING

Google DeepMind and Google Research unveiled a new AI-based cyclone forecasting model that predicts storm path, size, and intensity significantly faster than traditional physics-based systems.

MICROSOFT'S AI PROTOTYPE CAN REVERSE ENGINEER MALWARE, NO HUMAN NEEDED

Microsoft's Project Ire can (in some cases) reverse engineer software to determine if it's malicious.

DEMYSTIFYING CLAUDE CODE HOOKS

Claude Code Hooks transform Claude Code from a helpful assistant into a development workflow orchestrator.

CHATGPT REMOVES SEARCH ENGINE SHARING FEATURE

OpenAI discontinued an experimental ChatGPT feature that allowed users to make specific chats discoverable by search engines like Google.

Read the full report here ↗️

CLAUDE OPUS 4.1 LIKELY IN INTERNAL TESTING AS ANTHROPIC PREPARES SAFETY CHECKS

Newly surfaced references in Anthropic's configuration files suggest that it may be ramping up internal testing for what could become Claude Opus 4.1. The files hint at improved reasoning or planning capabilities. Anthropic's internal safety system is undergoing red teaming, a process that usually precedes model deployment by at least a week or two. A new Claude release would serve as Anthropic's answer to the highly anticipated GPT-5 launch, which seems imminent based on partner activity.

XAI LAUNCHES GROK IMAGINE FOR AI VIDEO AND IMAGES

CHATGPT NEARS 700M WEEKLY USERS

ChatGPT is approaching 700 million weekly active users, up from 500 million in March, according to OpenAI VP Nick Turley.

AI WILL NOT SUDDENLY LEAD TO AN ALZHEIMER'S CURE

THE RISE OF VERTICALIZED AI COWORKERS

AI coworkers are emerging across industries, focusing on repetitive tasks in sectors like finance, healthcare, and logistics. They integrate into existing workflows, take over specific tasks with human oversight, and are priced based on usage or outcomes. Key areas include document processing, voice-based interactions, content creation, and information retrieval.

👉 Reserve your spot

QWEN-IMAGE: CRAFTING WITH NATIVE TEXT RENDERING

Qwen-Image is a 20B MMDiT image foundation model that excels in complex text rendering and precise image rendering. It can create multi-line text layouts, paragraph-level semantics, and fine-grained details in both alphabetic languages and logographic languages with high fidelity. The model preserves both semantic meaning and visual realism during editing operations. It consistently outperforms existing models across diverse generation and editing tasks. Examples of images generated by the model are available in the post.

IMPROVING LLM CALIBRATION WITH LABEL SMOOTHING

Instruction tuning degrades confidence calibration in LLMs, making outputs less reliable. This paper proposes label smoothing as a remedy and introduces a memory-efficient loss kernel, showing that calibration issues grow with vocabulary size and hidden dimensions.

CCHISTORY: TRACKING CLAUDE CODE SYSTEM PROMPT AND TOOL CHANGES

A reverse-engineering project tracking Claude Code's system prompt evolution reveals Anthropic's iterative approach to AI tool development — removing emoji usage restrictions, tightening security policies, and adding PDF reading capabilities across 67 versions. The project bypasses anti-debugging measures to extract prompts, showing how AI companies continuously refine their models' behavior through detailed instruction changes rather than just model updates.

KAGGLE GAME ARENA

Google has launched Kaggle Game Arena, a new open-source platform for benchmarking AI systems through direct competition in strategic games.

RESTAURANT BOOKING WITH PERPLEXITY

Perplexity now integrates with OpenTable, allowing users to find and reserve restaurants directly through natural language queries. By analyzing user intent and preferences, it surfaces relevant dining options from OpenTable's network of 60,000 global partners.

pricing before it ends. Grab your pass today.

META TO SHARE AI INFRASTRUCTURE COSTS VIA $2 BILLION ASSET SALE

Meta reclassified $2 billion in data center assets as "held-for-sale" to attract financial partners for co-development, signaling a strategic shift as tech giants seek external funding for AI infrastructure.

PERPLEXITY ACCUSED OF IGNORING AI SCRAPING BLOCKS

Cloudflare claims that Perplexity bypassed robots.txt and other block signals to scrape websites that explicitly opted out.

FINE-TUNED SMALL LLMS CAN BEAT LARGE ONES AT 5-30X LOWER COST WITH PROGRAMMATIC DATA CURATION

This post discusses how to reproduce the workflow using open-source LLMOps tools, even without a GPU, as well as best practices for production deployments.

WHY TECH IS RACING TO ADOPT AI CODING

An interview with Michael Truell, CEO of Anysphere, the maker of Cursor, an automated programming platform that integrates with generated AI models to help developers code.

READ THE AWS CASE STUDY

GOOGLE RELEASES GEMINI 2.5 DEEP THINK FOR AI ULTRA SUBSCRIBERS

Google's most powerful AI model generates multiple solution approaches simultaneously before selecting the best answer. A specialized version that can work for hours on complex problems recently achieved IMO gold medal status and is being shared with select mathematicians.

ANTHROPIC CUTS OFF OPENAI'S ACCESS TO ITS CLAUDE MODELS

Anthropic revoked OpenAI's access after discovering the company was using Claude for internal benchmarking and coding assistance ahead of GPT-5's launch, violating terms that forbid using its AI to create and improve competitor models. OpenAI will retain access for safety evaluations and research collaboration on harmful capabilities testing.

OPENAI RAISES $8.3 BILLION AS PAID CHATGPT BUSINESS USERS REACH 5 MILLION

The round was part of the broader $40 billion fundraise and completed ahead of schedule, driven by OpenAI's annual recurring revenue jumping from $10 billion to $13 billion since June and projected to hit $20 billion by year-end. Dragoneer led with a $2.8 billion investment, 10% of its entire fund.

WILL DATA CENTERS CRASH THE ECONOMY?

Big Tech's $102.5 billion quarterly AI spending now exceeds dot-com era investment as a share of GDP, but the real risk lies in financing through "private credit" funds that operate outside traditional banking oversight. Banks have increased their exposure to these shadow lenders 14x since 2013, creating conditions where a synchronized AI market downturn could trigger defaults on data centers, cascading through private credit funds to their bank and insurance company lenders.

TOKENS ARE GETTING MORE EXPENSIVE

Models will always be cheaper next year, but users will expect more of them. The goalpost is always moving ahead. Even Anthropic, which has the most vertically integrated application layer on the planet, can't make a flat subscription with unlimited usage work. Companies without a plan are headed to the graveyard fast, as there are no investors writing checks for negative margin businesses. There's no figuring it out later when your AWS bill is larger than your revenue.

ANDREW NG ON CHINA'S AI MOMENTUM

In a detailed thread, Andrew Ng argued that China's open-model ecosystem, rapid semiconductor advances, and aggressive business dynamics give it a credible path to surpass the U.S. in AI.

→ Register now

DOES THE BITTER LESSON HAVE LIMITS?

General methods that leverage computation are ultimately the most effective by a large margin. The eventual success of approaches based on scaling computation is often tinged with bitterness because it is success over a favored, human-centric approach. Developers need to be aware that their work might be overtaken by compute, but this shouldn't stop them from applying custom, human-crafted logic to get the job done.

PERSONA VECTORS: MONITORING AND CONTROLLING CHARACTER TRAITS IN LANGUAGE MODELS

Researchers extracted "persona vectors" by comparing neural network activity when models exhibit specific traits versus when they don't, revealing that personality changes follow predictable mathematical patterns. This discovery enables a counterintuitive "vaccination" approach that purposefully steers models toward undesirable traits during training to compensate for their tendency to “overfit” to negative traits.

COHERE'S COMMAND A VISION

Cohere has launched Command A Vision, an open-weight 111B parameter vision-language model designed for business use cases. It reportedly surpasses GPT-4.1 and Llama 4 Maverick on enterprise benchmarks.

SEED-PROVER: DEEP AND BROAD REASONING FOR AUTOMATED THEOREM PROVING

ByteDance's Seed-Prover proved 5 of 6 problems at IMO 2025, the third AI system to do so, and the first Chinese model, in recent weeks. The system combines "lemma-style" whole-proof reasoning that breaks problems into modular components with three-tiered inference strategies from quick refinements to days-long reasoning that accumulates thousands of mathematical facts.

HOW WINDSURF'S FOUNDERS GOT PAID FROM THE GOOGLE DEAL

New details show Google's $2.4B licensing deal with Windsurf split evenly between investors and compensation for 40 employees hired by Google, including co-founders Varun Mohan and Douglas Chen.

AI'S OVERLOOKED $97 BILLION CONTRIBUTION TO THE ECONOMY

The US economy grew at an annual rate of 3% in the second quarter. Americans enjoyed roughly $97 billion in consumer surplus (the difference between the maximum a consumer is willing to pay for a good or service and its actual price) from generative AI tools in 2024. Around 40% of US adults are regular users of generative AI. As more digital goods become available for free, measuring benefits as well as costs will become increasingly important.