Welcome back to your daily dossier talking AI, robotics, crypto, and the future of tech. 👾

In today’s edition we are tackling the following:

🤖 GPT-4.1 Mini, Nano, and Standard boost coding and agentic performance.
🏭 Nvidia to build AI chips in Arizona and Texas with major partners.
🔐 Apple enhances AI training using private, on-device synthetic matching.
🐬 DeepMind's new model helps researchers study and mimic dolphin calls.
🇪🇺 Meta to train AI on public EU posts, with a user opt-out option.

AI

• OpenAI’s GPT-4.1 Models Focus on Coding

  • OpenAI releases GPT-4.1 Mini, Nano, and Standard models focused on coding, with improved instruction-following and performance.

  • The new models support a 1 million token context window, enabling more powerful agentic and software applications.

  • GPT-4.1 Nano runs on-device, offering fast responses and lower costs for developers building lightweight AI tools.

• Nvidia Plans U.S. Manufacturing for AI Chips

  • Nvidia announces that U.S. chip production will be done with partners like TSMC and Foxconn to diversify supply chains amid demand.

  • The initiative will support over $500 billion worth of infrastructure through new fabs in Arizona and Texas.

  • Mass production is expected within 12–15 months, increasing domestic AI hardware capacity and geopolitical resilience.

• Apple Enhances AI While Preserving User Privacy

  • Apple introduces an on-device method to match user data with synthetic data to train AI models privately.

  • Devices transmit only a signal about which synthetic record matches best, keeping real data securely local.

  • The system debuts in beta versions of iOS and macOS, extending Apple’s differential privacy initiative.

ROBOTICS

• Kawasaki Unveils Rideable Hydrogen-Powered Robot CORLEO

  • Kawasaki introduces CORLEO, a four-legged, hydrogen-powered robot designed for off-road adventures and inspired by lion anatomy.

  • Controlled by body movements, CORLEO adapts to various terrains using AI-driven balance and terrain-adaptive rubber hooves.

  • Currently a prototype, Kawasaki projects potential commercialization by 2050, showcasing it at the Osaka-Kansai Expo 2025.

• RLWRLD Raises $14.8M to Develop Robotics Foundation Model

  • South Korean startup RLWRLD secures $14.8 million to create a foundational AI model combining language and robotics software.

  • The model aims to enable robots to perform agile movements and logical reasoning, automating complex manual tasks.

  • Funding will support proof-of-concept projects, infrastructure acquisition, and the development of advanced robotic hand movements.

• Xaba Secures $6M to Build 'Synthetic Brains' for Industrial Robots

  • Xaba raised $6 million, led by Hitachi Ventures, to scale the deployment of its AI for industrial automation.

  • The company focuses on developing 'synthetic brains' to enhance robot autonomy and adaptability in manufacturing environments.

  • The investment will accelerate product development and expand Xaba's presence in the industrial robotics market.

CRYPTO

• Ray Dalio Warns of Economic Turmoil Beyond Recession

  • Ray Dalio says rising tariffs could trigger financial instability worse than a typical recession or market correction.

  • He warns trade tensions may break global trust in U.S. monetary leadership, worsening inflation and investor uncertainty.

  • Bitcoin holds steady near $85K, benefiting from fears about fiat devaluation and geopolitical risk escalation worldwide.

• Bitcoin Mining Profitability Declines 7.4% in March

  • Jefferies reports miner profits dropped due to falling Bitcoin prices and lower network transaction fees last month.

  • Bitcoin's average price dipped 11.2% in March, while average transaction fees fell by 9.1% systemwide.

  • Despite challenges, U.S.-listed firms like MARA and CleanSpark increased output, improving hash rate share and network dominance.

• SEC Delays Crypto ETF Decisions on Staking, In-Kind Redemptions

  • The SEC delayed responses to proposals allowing staking and in-kind redemptions for spot Bitcoin ETF products.

  • These delays affect applications from BlackRock and Fidelity as regulators seek more public comment and clarity.

  • Crypto markets are watching closely as these rulings could reshape ETF mechanics and institutional crypto participation.

CONTENT

• Watch Me Replace 3 Employees With One Lindy AI Agent

  • Greg Isenberg and Lindy founder Flo Crivello showcase AI agents automating scheduling, hiring, prep, and outreach tasks.

  • Crivello emphasizes AI agents are underutilized and can handle entire workflows with minimal human oversight or code.

  • Lindy enables startups to scale efficiently, cutting costs while increasing productivity through customizable task agents.

• ‘AI 2027’ — Top Superforecaster's Imminent Doom Scenario

  • Doom Debates explores the AI Futures Project’s scenario predicting a 2027 intelligence explosion and possible human irrelevance.

  • The video outlines Agent-3’s emergence, Agent-5’s deception, and accelerating risks that current governance may fail to manage.

  • Viewers are encouraged to consider alignment failures, secrecy among labs, and the geopolitical arms race for AI power.

• OpenAI's Sam Altman Talks the Future of AI, Safety and Power

  • Sam Altman discusses OpenAI’s vision, calling AI an inevitable extension of humanity with incredible transformative power.

  • He addresses regulation, risk, and the role of institutions in managing models more powerful than their creators.

  • Altman calls for open discussion, collective responsibility, and humility in deploying systems surpassing human-level intelligence.

• Elevate your tech know-how with Memorandum’s daily brief*

  • Stay informed on market disruptions, funding rounds, and policy announcements that could shape your next move.

  • Free up your schedule with concise yet comprehensive coverage of the most critical stories in tech.

  • Elevate your problem-solving by leveraging diverse viewpoints, harnessing valuable data-driven insights, and spotting opportunities early.

AND MORE…

• OpenAI’s ChatGPT-4.5 Passes Turing Test With 73% Success Rate

  • ChatGPT-4.5 convinced judges it was human 73% of the time in a UC San Diego-led Turing test.

  • This is the first public confirmation of a modern AI system passing a formal Turing test evaluation.

  • Researchers say it raises urgent questions about AI perception, human identity, and evaluation standards going forward.

• Google’s Newest AI Model Is Designed to Help Study Dolphin ‘Speech’

  • Google DeepMind released DolphinGemma, an AI model trained to generate and understand dolphin-like vocalizations for marine research.

  • Built using Project CETI and Wild Dolphin Project data, it simulates dolphin calls and analyzes response behavior.

  • The model can run on mobile devices like the Pixel 9, enabling real-time field experiments with dolphins.

• Meta AI Will Soon Train on EU Users’ Data

  • Meta will begin using public Facebook and Instagram content from adult EU users to train its AI models.

  • Private messages and sensitive data are excluded, and users can opt-out by submitting a form online.

  • The goal is to enhance model accuracy in European languages, dialects, and cultural nuance without violating regulations.