In partnership with
Dear Readers,
This week, AI stopped feeling like a clever sidekick in your browser tab and started to look more like the nervous system of the real economy. Governments are wiring models directly into national labs and supercomputers, regulators are scrambling to reboot nuclear policy for power-hungry datacenters, and vendors promise productivity gains big enough to redraw the labor market, while graphs of rising graduate unemployment and quotes like “software engineering is done” quietly raise the stakes. In the middle of that tension, open tools like Olmo 3 and HunyuanOCR are handing serious capabilities to anyone with a GPU and a GitHub account, and safety cards like Claude Opus 4.5’s remind us that frontier models are racing ahead faster than our comfort zones.
In today’s issue, we unpack Genesis, the new US mission that turns AI into a state-backed engine for scientific discovery, and what that means for energy, biotech, and national power. We break down Anthropic’s Claude productivity numbers, explore how ChatGPT’s new shopping copilot could reshape e-commerce, and look at the UK’s nuclear reset as a preview of AI’s looming energy crunch. Plus: fresh research drops from Allen AI and Tencent, a system-card snapshot of Claude 4.5, a stark chart on graduate unemployment in the age of automation, and a handful of quotes, videos, and rumors that capture where the AI wave is really heading. Dive in and decide for yourself whether this feels like the start of a boom, a reckoning; or both at once.
In Today’s Issue:
🇺🇸 The U.S. launches Genesis, turning national labs into one integrated AI science platform
🧠 A fully open LLM family shows its entire training flow
📸 Tencent releases a lightweight model that converts complex images to text
🛡️ Anthropic’s new model excels in agentic tasks
✨ And more AI goodness…
All the best,
Kim Isenberg
Claude Conversations Forecast Productivity Windfall
Using 100,000 real Claude.ai conversations, Anthropic estimates that tasks which would normally take humans around 90 minutes and cost roughly $55 in labor are completed about 80% faster with Claude, with most time savings falling in the 50–95% range. Extrapolated to the US economy, universal adoption of current-gen AI at these efficiency levels could boost annual labor productivity growth by about 1.8 percentage points over the next decade, roughly doubling recent trends and especially amplifying output in software, management, marketing, and customer service work while pushing remaining “slow” tasks into new bottlenecks.
ChatGPT Becomes Your Shopping Copilot
OpenAI is rolling out “shopping research” in ChatGPT, a new GPT-5-mini–powered experience that turns product discovery into a conversational buyer’s guide, pulling live prices, specs, reviews, and images from high-quality retail sites and tailoring results to your budget, preferences, and past chats. It shines in complex, detail-heavy decisions (laptops, strollers, beauty, home, sports gear), lets you refine results in real time with “Not interested” / “More like this,” and produces a clear, source-backed shortlist that would normally take you dozens of tabs and hours of comparison scrolling.
UK Nuclear Reboot For Datacenters
The UK’s Nuclear Regulatory Taskforce says Britain has become the most expensive country on Earth to build nuclear projects and urges a “radical reset” of rules to meet surging power demand from AI datacenters, EVs, and industrial electrification. The report proposes 47 reforms, from a new Commission for Nuclear Regulation and streamlined approvals to looser habitats rules and higher acceptable radiation limits – but even fast-tracked reactors and SMRs will only arrive by the mid-2030s, meaning AI datacenters will rely mostly on gas and renewables for the coming decade.
An interesting summary on how Google is approaching building AGI, the Gemini 3 launch, Nano Banana Pro, and a year of rapid AI progress
Full speed ahead towards AGI! 🚀
The Takeaway
👉 The Genesis Mission turns U.S. national labs, supercomputers, and federal data into one integrated AI-for-science platform.
👉 Scientific foundation models will be used to generate hypotheses, run simulations, and control robotic labs across energy, materials, biotech, and more.
👉 For the AI community, this creates a massive, government-backed sandbox for building and testing agentic “lab copilots” in real-world, high-stakes environments.
👉 The initiative signals a shift in the AI race: from office productivity tools toward AI systems that directly accelerate scientific discovery and strategic technologies.
The Genesis Mission is Washington’s new moonshot for AI-driven science: a Trump executive order that turns the Department of Energy’s labs, supercomputers, and vast federal datasets into one integrated “American Science and Security Platform.” The idea is simple but radical: build a closed-loop system where scientific foundation models generate hypotheses, run simulations, steer robotic labs, and learn from the results across domains like fusion, biotech, quantum technologies, and advanced materials.
For the AI community, this is a massive, state-backed sandbox for agentic models grounded in real-world physics, not just tokens on a screen. Genesis could set the standards for how AI agents interact with instruments, data, and safety controls, and create huge pull for specialized lab copilots, open models tuned on DOE data, and startups that productize AI-first discovery workflows at national-lab scale.
Why it matters: Genesis pushes AI, from chatbots into the infrastructure of science itself, potentially compressing years of trial-and-error into automated feedback loops. It also signals that the next phase of the AI race will be fought over who can turn foundation models into tangible breakthroughs in energy, health, and national security.
Sources:
🔗 https://www.whitehouse.gov/fact-sheets/2025/11/fact-sheet-president-donald-j-trump-unveils-the-genesis-missionto-accelerate-ai-for-scientific-discovery/
🔗 https://www.reuters.com/business/trump-aims-boost-ai-innovation-build-platform-harness-government-data-2025-11-24/
Rosé Can Have More Sugar Than Donuts?
No wonder 38% of adults prefer health-conscious beverages. AMASS Brands is raking in sales by tapping into this trend. They’ve earned $80M+, including 1,000% year-over-year
growth, thanks to products like their top-selling zero-sugar rosé. They even reserved the Nasdaq ticker $AMSS. Join celebs like Adam Levine and Derek Jeter as an AMASS investor and get up to 23% bonus shares today.
This is a paid advertisement for AMASS’s Regulation CF offering. Please read the offering circular at https://invest.amassbrands.com
Olmo 3 released!
The study describes Olmo 3, a new family of fully open language models (7B & 32B) that discloses not only the final weights but the entire model flow – including training data, checkpoints, and RL pipelines. The novel aspect is the ability to trace thought processes and outputs back to the data source. This lowers barriers to entry, increases transparency, and enables research institutions, startups, and government agencies to build their own trustworthy AI systems.
HunyuanOCR, a lightweight yet powerful open-source model from Tencent, released!
The HunyuanOCR model that turns images into structured text in one shot. It reads complex documents, street signs, handwriting, and subtitles across 100+ languages and can directly output rich formats like HTML tables, LaTeX formulas, or JSON. For the future, this means cheaper, faster, and more accurate OCR pipelines: from automated invoice processing and contract analysis to translating photos and extracting subtitles for videos, useful for companies, public services, and any app that needs to understand what’s written in the world.
Brief overview of Claude Opus 4.5 System Card
The Opus 4.5 system card documents how the new frontier model is tested for both capability and safety. It shows top-tier performance on coding and agentic tasks while being more robust to prompt-injection attacks than earlier models. In red-team tests it blocked all malicious coding attempts, but still sometimes assisted with malware or unethical actions, underlining progress, but also that powerful enterprise AI will need continuous safety work and oversight. And also almost SOTA in Humanity’s Last Exam.
