Search the archive

MCP WRITING CODE TO CALL MCP: MCPS ALL THE WAY DOWN

This post explores how to build a universal engine for any schema-based integration, without manual plumbing.

SEE HOW AHREFS DOES AI

INTRODUCING GPT-5.2

OpenAI's GPT‑5.2 is optimized for professional use and long-term agent workflows. It improves performance across knowledge work, long-context reasoning, tool use, and multimodal understanding.

DISNEY INVESTING $1 BILLION IN OPENAI, LICENSING CHARACTERS FOR SORA

A three-year licensing deal will let Sora and ChatGPT Image users create content with over 200 characters across Disney, Marvel, Pixar, and Star Wars starting next year.

FAST MODEL LOADING

Tensor R-Fork is a zero-copy method for rapidly transferring model weights between running instances using GPU-to-GPU data transfer. It cuts loading time, reduces disk/DRAM usage, and keeps inference services uninterrupted.

PROMPTS FOR OPEN PROBLEMS

This post proposes different research problems in machine learning research. The ideas include design-based machine learning, a theory of competitive testing, going beyond average case evaluation certainty equivalence in LLM reasoning, and open source, open corpus language models. The research areas are all doable, there just needs to be a will to do it.

BUSINESS AI ADOPTION FLATLINES

AI adoption held flat for 45% of businesses in November, and there were slight declines in the finance and technology sectors. Adoption rose in several industries with relatively low adoption rates, like retail, construction, and manufacturing. The slowdown comes after a rapid run-up in adoption rates in 2025. The effects of the latest advancements seem to have faded. Another run-up in adoption will likely require technological gains or implementation gains.

for free.

CURSOR BROWSER VISUAL EDITOR

Cursor released a new visual editor that integrates design and code within the Cursor Browser. Users can drag UI elements, inspect props, and make changes interactively.

BUILD WITH GEMINI DEEP RESEARCH

Gemini Deep Research is now available to developers via the new Interactions API. It is launching alongside DeepSearchQA, a benchmark for complex web search tasks. Gemini Deep Research is an agent optimized for running context gathering and synthesis tasks. It uses Gemini 3 Pro as its reasoning core. The agent autonomously navigates complex information landscapes with high accuracy by scaling multi-step reinforcement learning for search.

LEARNING COLLATZ - THE MOTHER OF ALL RABBIT HOLES

The Collatz conjecture is one of the most famous unsolved problems in mathematics. It asks whether repeating two simple arithmetic operations will eventually transform every positive integer into 1. This article looks at what happens when a transformer is trained on the unsolved puzzle to study how models fail.

200K TOKENS IS PLENTY

Long context windows are overrated for agentic coding. Amp's team ships features using clusters of small threads averaging 80k tokens each rather than jamming everything into one million-token mega-thread.

CHATGPT'S ‘ADULT MODE' IS COMING IN 2026

OpenAI's CEO of Applications recently told reporters that ChatGPT's 'adult mode' will debut during the first quarter of next year. The company had cut back on the chatbot's personality in response to growing concerns about user mental health. It has since added additional parental controls and new efforts to age-gate younger users. The company is still testing its age verification technology to ensure it can accurately identify teens and not misidentify adults before officially rolling out the split experience.

BROADCOM REVEALS ITS MYSTERY $10 BILLION CUSTOMER IS ANTHROPIC

The customer Broadcom signed a $10 billion order for custom chips in September has been revealed to be Anthropic. Broadcom will deliver entire server racks to Anthropic, not just chips. Anthropic has placed an additional $11 billion order with Broadcom in the latest quarter. The deal has drawn significant investor attention amid the AI infrastructure boom.

→ CREATE ONE NOW

RUNWAY CLAIMS ITS GWM-1 “WORLD MODELS” CAN STAY COHERENT FOR MINUTES AT A TIME

Runway's GWM worlds offers an interface for digital environment exploration with real-time user input that affects the generation of coming frames.

TRUMP SIGNS EXECUTIVE ORDER SEEKING TO BLOCK STATES FROM REGULATING AI COMPANIES

Trump created an AI Litigation Task Force to challenge “onerous” state AI regulations and potentially restrict federal funding to states with their own AI laws.

CHINESE AI IN 2025, WRAPPED

A look at the biggest model drops of the year, China's evolving AGI discussion, the biggest swings in the US-China chip war, China's answer to the US' AI Action plan, and robots.

GOOGLE TESTS GEMINI 3 FLASH MODEL ON LM ARENA

Google is actively testing two new models on LM Arena, suggesting a potential Gemini 3 Flash or Gemini 3 Pro general availability launch.

Read the whitepaper

INSIDE META'S PIVOT FROM OPEN SOURCE TO MONEY-MAKING AI MODEL

Meta is developing a closed model codenamed Avocado, expected next spring, after years of loudly supporting open-source. The new model is trained on the outputs (or “distilled”) from rival models, including Google's Gemma, OpenAI's gpt-oss, and Alibaba's Qwen. New Chief AI Officer Alexandr Wang is reportedly chafing under Zuckerberg's “micromanagement”. Long-time AI head Yann LeCun left last month after being sidelined under the new org structure.

GOOGLE DEEPMIND WILL OPEN A ROBOTIC AI LAB IN THE UK TO DISCOVER NEW MATERIALS

Google DeepMind will open a research lab for discovering new materials in the UK as part of Alphabet's newly announced partnership with the British government. The agreement involves Alphabet tailoring several of its AI models for scientists, teachers, and public employees in the UK. The new lab will research materials that could improve medical imaging, solar panels, and chips. Using AI to unearth novel materials could drastically reduce costs and time.

SHOPIFY MERCHANTS CAN NOW SELL PRODUCTS THROUGH AI CHATBOTS

Shopify merchants' products can now be discovered on AI platforms like ChatGPT, Perplexity, and Microsoft Copilot. Agentic storefronts allow merchants to sell their products across popular API chatbots. Agents can answer customers' questions while giving merchants insight into search trends and topics customers are asking about. Shopify plans to make every Shopify store agent-ready by default.

I REVERSE ENGINEERED CHATGPT'S MEMORY SYSTEM

No vector databases, no RAG over conversation history. The system uses four layers: ephemeral session metadata (device, location, and usage patterns), long-term facts stored via a dedicated tool, lightweight summaries of recent messages, and a sliding window of the current conversation. OpenAI prioritizes speed and token efficiency over detailed historical context.

IS IT A BUBBLE?

No one can definitively say whether we are currently in a bubble or not. No one should go all-in without acknowledging that they face the risk of ruin if things go badly. At the same time, no one should stay all-out and risk missing out on one of the great technological steps forward. A moderate position seems like the best approach. Intelligent investment requires sober, insightful judgment and skillful implementation.

Shell. Try Tangle (open source)

CURSOR DEBUG MODE

Cursor has introduced Debug Mode, a new agent loop that incorporates runtime logs and human interaction to fix stubborn bugs. Inspired by expert debugging workflows, the mode generates multiple failure hypotheses, instruments code with logs, and loops in the user to reproduce errors and verify fixes.

VECTORIZED MAXSCORE OVER WAND: BUILDING FASTER TEXT SEARCH FOR HUMANS AND AGENTS

Designing fast search engines is a subtle optimization problem. It takes effort to select the right algorithm and then tune it to get the best out of the CPU. Agents write longer queries than humans, so it is becoming increasingly important for text search to scale well with the number of terms. For text search, this means the cursor has shifted more and more from WAND to MAXSCORE.

UPGRADES TO CLAUDE CODE CLI

Claude Code CLI now has async subagents, instant compact, customer session names, and usage stats. There are also new keyboard shortcuts on the /resume screen. Run 'claude update' to get all of the new features.

TOWARDS A SCIENCE OF SCALING AGENT SYSTEMS

Agents are becoming the dominant paradigm for real-world AI applications. However, practitioners still rely on heuristics rather than principled design choices. This study attempts to derive quantitative scaling principles for agent systems.

GETTING A GEMINI API KEY IS AN EXERCISE IN FRUSTRATION

Gemini is aimed at large organizations, not individual developers. It is designed for maximum compliance with government regulations, not for efficiency or productivity. Google doesn't want your business unless you're an organization that employs tens of thousands of people. OpenAI and Anthropic are much smaller and more nimble. They are able to make the process of setting up a developer account quick and easy.

AI WORLD MODELS RESHAPE INTERACTIVE STORYTELLING

NVIDIA-BACKED STARCLOUD TRAINS FIRST AI MODEL IN SPACE AS ORBITAL DATA CENTER RACE HEATS UP

Nvidia-backed Starcloud successfully trained the first AI model in space, using an Nvidia H100 GPU to run Google's Gemma LLM on their Starcloud-1 satellite. These orbital data centers promise significant energy savings compared to Earth-based facilities, addressing the growing power demands and environmental impact of terrestrial data centers. Starcloud plans a 5-gigawatt orbital data center to capture constant solar energy, with implications for both commercial AI workloads and real-time data analysis for military and emergency response operations.

READ MORE ON THE NEW AMP WEBSITE >

GOOGLE'S AI-POWERED ARTICLE OVERVIEWS

Google is expanding features like Preferred Sources in Search, allowing users to prioritize their favorite outlets in Top Stories results.

ADOBE BRINGS TOOLS TO CHATGPT

Adobe has launched Photoshop, Adobe Express, and Acrobat within ChatGPT, allowing 800 million users to create and edit by simply describing their goals.

GOOGLE UPGRADES GEMINI 2.5 TEXT-TO-SPEECH MODELS

ANNOUNCING MODEL CONTEXT PROTOCOL (MCP) SUPPORT FOR GOOGLE SERVICES

Google has released full-managed, remote MCP servers, providing a unified layer across all Google and Google Cloud services.

catch critical issues early. Read the report

Want to check it out firsthand? Experiment with AI observability tools in the Dynatrace Playground [5] - where you can explore sample data without installing any software.

OPENAI TESTING NEW IMAGE-2 MODELS ON LM ARENA

OpenAI is preparing to roll out its next generation of image generation models along with GPT-5.2. Two new models have appeared on online evaluation platforms. Early testers note a substantial increase in image detail and fidelity. The upgrade brings OpenAI's image generation closer to the standard set by Google Nano Banana 2. Examples of images generated with the models are available in the article.

GLM-4.6V: OPEN SOURCE MULTIMODAL MODELS WITH NATIVE TOOL USE

GLM-4.6V (106B), a foundation model designed for cloud and high-performance cluster scenarios, and GLM-4.6V-Flash (9B), a lightweight model optimized for local deployment and low-latency applications, have been open-sourced. GLM-4.6V scales its context window to 128k tokens in training. It achieves SoTA performance in visual understanding and reasoning among models of similar parameter scales. The model has native Function Calling capabilities.

INTRODUCING DEVSTRAL2 AND MISTRAL VIBE CLI

Devstral 2 hits 72.2% on SWE-bench Verified with 123B parameters, making it one of the best open-weight coding models despite being a fraction of the size of its peers. The 24B Devstral Small 2 scores 68.0% and runs on consumer hardware. Mistral also launched Vibe CLI, an open-source terminal agent that orchestrates multi-file changes across codebases.

HOW PEOPLE USE AI AT WORK

Professionals use AI for grunt work and first drafts. It is never trusted without supervision. Artists use AI to handle admin work so they have more time for creative work. Researchers find that the time required to fact-check AI output often negates the efficiency gains. The primary barrier to AI adoption is hallucination.