Samwise Tech Newsletter
AI & Technology Intelligence, Curated Daily
Tuesday, April 7, 2026
TOP 5 STORIES
- OpenAI’s 13-page “social contract” asks government to tax robots, create national wealth fund, implement 4-day workweek
- New Yorker surfaces secret memos behind Sam Altman’s firing â 100+ interviews, Sutskever dossier, Amodei notes
- UBTech offers $18M a year for a single AI scientist â one job listing as a signal in the global humanoid arms race
- Japan’s vanishing workforce turns to robots not to cut costs â but because there are no workers left to hire
- Meta’s first Alexandr Wang models are nearly ready to ship â open-source versions planned despite earlier delays
OpenAI’s New “Social Contract” for Society and ASI
OpenAI published a 13-page policy document outlining what CEO Sam Altman calls a “new social contract” for The age of superintelligence. The proposal asks Washington to tax AI-driven profits and create a sovereign-style wealth fund â modeled on Alaska’s oil revenue program â that would pay dividends to every American. Other ideas include taxes on robot labor, a four-day workweek, a “Right to AI” access for all citizens, and containment playbooks for rogue autonomous AI. Axios described it as “the most detailed blueprint any tech titan has ever published for how to tax, regulate, and redistribute wealth from the technology he’s building.” The document explicitly states that the transition toward superintelligence has already begun.
New Yorker Surfaces Secret Memos Behind Altman’s Firing
The New Yorker published a sweeping investigation into Sam Altman, drawing on more than 100 interviews, previously unseen internal memos from former chief scientist Ilya Sutskever, and private notes compiled by Dario Amodei. Sutskever’s dossier â built from 70 pages of Slack messages and HR documents â alleges Altman repeatedly misrepresented AI safety protocols to OpenAI’s board. Amodei’s notes, written independently over years, reach the same conclusion: the problem at OpenAI is Altman himself. A Microsoft executive told reporters there is “a small but real chance” Altman could be remembered as a “Bernie Madoff, Sam Bankman-Fried-level scammer.” No single smoking gun, but a detailed and troubling pattern spanning his entire career.
Wang’s First Meta Models Getting Ready to Ship
Meta is preparing to release the first AI models developed under Alexandr Wang’s Superintelligence team, with Axios reporting that some will be open-sourced while the largest models stay closed. The release follows a March delay of the codenamed “Avocado” model after benchmark performance fell short of rival models across the board. Despite the stumble, Meta is reportedly confident it can carve out consumer-focused areas of strength even where it does not lead overall, planning a hybrid approach with open models for broad distribution and closed models for its core apps. After massive investment and high-profile new hires, another disappointing release would be a painful outcome for a tech giant pushing hard to enter the frontier AI race.
UBTech Offers $18M a Year for One AI Scientist
Chinese humanoid maker UBTech is offering up to $18 million annually for a single chief AI scientist, turning one job posting into a global signal of how extreme the humanoid arms race has become. The role will lead “embodied intelligence” research, translating vision-language-action and robotics models into dependable software for full-size industrial humanoids. In January, Airbus deployed UBTech’s Walker S2 robots on aircraft assembly lines â proving these machines can operate in real factory environments beyond controlled demos. UBTech reports humanoid revenue surging with sales climbing more than 50%. By dangling CEO-level pay, the company appears to be using a single role as both a talent grab and a signal of dominance in a fiercely competitive global race.
Japan’s New Workforce: Robots Wanted
With working-age citizens comprising just 59.6% of Japan’s population â and nearly a third of the country already over 65 â Japan is deploying robots not to cut costs, but because there are no workers left to hire. Robots are filling frontline roles in convenience stores, logistics, and hospitality: stocking shelves, cleaning floors, and delivering room service. Elderly care facilities are adopting robotic assistants for lifting patients, monitoring vital signs, and providing companionship. Policymakers are explicitly reframing automation as critical economic infrastructure rather than an employment threat. Japan’s experiment is now being studied worldwide as a preview of what awaits other rapidly aging economies facing the same mathematical impossibility of full human employment.
This Tiny Bot Grows Its Own Nervous System
Researchers at Tufts University and Harvard have created “neurobots” â microscopic living machines assembled from frog cells that spontaneously develop their own rudimentary nervous systems. Unlike conventional robotic engineering, these organisms grow neurons that wire directly into the outer cell layer and begin shaping movement from within. As the nervous system emerges, gene activity shifts, activating pathways associated with brain formation and even eye development. Neurobots swim with greater intensity and display varied, unpredictable movement patterns compared to earlier frog-cell bots. While early-stage, the research hints at a future class of engineered life that self-repairs and behaves like biological tissue while being designed and deployed like hardware â a fundamentally new category of machine.
Paperclip: The Open-Source Platform Turning AI Agents into an Actual Company
Businesses deploying autonomous AI agents face a problem that has nothing to do with capability: accountability. Legal, operational, and financial accountability structures were never designed for systems that act on their own. Paperclip, an open-source platform profiled by Kristopher Dunham on Medium, attempts to bridge this gap by giving AI agents a formal corporate-style structure â documented decision trails, accountability layers, and operational boundaries that enterprise deployments require. Dunham argues that without this scaffolding, deploying agents in high-stakes business contexts remains fundamentally risky regardless of how capable the underlying model is. The platform is gaining attention as organizations begin moving from AI assistants to autonomous AI agents at scale.
Your AI Coding Mandate Will Backfire. Amazon Proved It.
Alvis Ng’s widely read Medium essay argues that mandatory AI coding adoption programs are structurally counterproductive â and uses Amazon’s internal experience as the primary evidence. The critical distinction: engineers who choose to integrate AI into their workflow internalize its limitations and develop sound judgment about when to use it and when not to. Engineers who are forced to adopt AI by management directive instead learn to perform compliance without developing genuine skill. Ng argues that the productivity gains cited in AI adoption studies are driven overwhelmingly by voluntary adopters, and that top-down mandates suppress the autonomous judgment that makes AI a real force multiplier rather than a performance-review metric that teams learn to game.
This World Model Learns Physics by Watching Videos
Yann LeCun’s team at Meta has published a world model that develops physical intuition by watching videos rather than processing text â building internal representations of mass, trajectory, and collision dynamics from observation alone, without explicit physics instruction. Sumit Pandey’s analysis in Towards Deep Learning on Medium covers how the model generalizes from observed sequences to novel physical scenarios it has not encountered before. The approach represents a meaningful departure from the language-model paradigm that has dominated AI development since GPT-3. For LeCun, who has long argued that generalized intelligence requires grounding in physical reality, the work provides concrete evidence for his thesis that video prediction â not text prediction â is the more promising path to machine common sense.
Artemis II Crew Sets New Human Distance Record
The four-person crew of NASA’s Artemis II set the record for the farthest humans have ever traveled from Earth on Monday, April 7. At 1:57 p.m. ET, the astronauts reached 252,752 miles from Earth, surpassing the record set by the Apollo 13 crew in 1970 by approximately 4,102 miles. The crew then completed a flyby of the moon â including 40 planned minutes without communication as they passed the far side â reaching a maximum distance of 252,756 miles just after 7 p.m. ET. Canadian astronaut Jeremy Hansen said the crew was “challenge this generation and the next to make sure this record is not long-lived.” The crew is scheduled to splash down back on Earth on Friday, April 11.
Anthropic Plans Extra Charges for Third-Party Tools in Claude Code
Anthropic is planning to charge Claude Code subscribers extra to use OpenClaw and other third-party tools, in what Morning Brew describes as a major change to the platform’s pricing structure. Claude Code, which allows developers to delegate coding tasks directly from the terminal, has emerged as one of the most widely adopted AI development tools. The additional charge for third-party tool integrations could meaningfully affect developers who have built workflows dependent on external tools inside the Claude Code ecosystem. The move places Anthropic in a delicate position â simultaneously trying to grow its developer ecosystem while tightening the economics of building on Claude subscriptions, at a moment when OpenAI is actively competing for the same developer audience.
Sources: The Rundown AI • The Rundown Robotics • Medium Daily Digest • Morning Brew