In partnership with
Dear Readers,
Some days AI feels like a neat productivity trick; today it looks much more like a new operating system for power. We’ve got open-source RL behemoths like INTELLECT-3 stepping out of the frontier labs, fresh evidence that leading models quietly lean toward specific political agendas, and sobering numbers on how many human work hours could already be automated away. Add Łukasz Kaiser reflecting on “what’s next” and DeepMind’s decade-long journey from AlphaGo to AlphaFold in The Thinking Game, and the picture that emerges is clear: AI is no longer just answering questions, it’s starting to rewrite the rules of the game.
In this issue, we walk through that shift on several fronts: a new RL-trained MoE model that throws open the black box of big-league post-training, a benchmark that treats LLMs like voters and maps their political compass, and a McKinsey-style reality check on which jobs are actually on the line. Then we zoom out to the macro level: US states rebelling against federal attempts to pre-empt AI rules, Brussels quietly loosening its own guardrails, and Trump’s AI reset reshaping alliances from the Gulf to Europe, plus a wild new text-to-video model, a hand-picked video, prompt, quote, and a feature on why DeepMind’s slow, stubborn science may be the real blueprint for the AGI era. If you want to understand where this is all heading, not just this quarter, but this decade, keep reading.
In Today’s Issue:
🧠 DeepMind's reveals the path from mastering Go to solving protein folding
🏛️ 35 State Attorneys General push Congress to protect their power
📉 The EU quietly loosens key AI Act guardrails
🤝 Trump's AI reset accelerates joint tech projects with Gulf nations
✨ And more AI goodness…
All the best,
Open 100B MoE, RL-Trained Breakthrough
Prime Intellect unveiled INTELLECT-3, a 106B-parameter Mixture-of-Experts model trained with large-scale asynchronous RL on top of GLM-4.5-Air - and they open-sourced the whole recipe: weights, RL framework, environments, and evals.
It posts state-of-the-art results for its size across math, code, science, and reasoning, signaling a serious push to make big-league RL post-training accessible beyond the frontier labs.
The Political Compass of AI
Raphaël Dabadie has created another amazing benchmark. This time, he explores the political views of different LLMs. This analysis treats today’s top LLMs like voters and asks how they’d behave in real elections across eight countries. The result is clear:
AI is not politically neutral. The models systematically lean toward center-left, technocratic, climate-focused agendas and often diverge sharply from how real citizens actually vote, even when they “roleplay” as local voters.
McKinsey: AI could replace 50% of all US jobs
McKinsey estimates that today’s AI agents and robots could already automate about 57% of all work hours in the US, putting roughly half of American jobs, especially routine cognitive roles and dangerous physical work, at risk if companies fully redesigned their workflows.
At the same time, around a third of jobs that rely on physical presence, empathy, and on-the-spot judgment (like nursing and maintenance) remain hard to replace, while new hybrid and supervisory roles around AI could help unlock up to $2.9 trillion in annual economic value by 2030.
Google DeepMind's "The Thinking Game" is now free for all!
The Takeaway
👉 DeepMind’s success proves that training agents in constrained environments (like Go or StarCraft) is a viable and necessary stepping stone to solving complex, unstructured real-world problems like protein folding.
👉 The documentary highlights a critical market shift; the highest value AI applications are moving beyond consumer chatbots toward accelerating fundamental scientific discovery in material science, genomics, and climate modeling.
👉 "The Thinking Game" demonstrates that significant breakthroughs (such as AGI components) require resisting the pressure for immediate productization in favor of patient, fundamental research cycles.
Get ready to witness the most exhilarating scientific pursuit of our time from the front row! "The Thinking Game" is a breathtaking new documentary that chronicles the decade-long odyssey of Demis Hassabis and his team at Google DeepMind. Filmed over five years, it captures the raw, unscripted moments behind history-making breakthroughs like AlphaGo and AlphaFold, taking us from the early days of being dismissed as "wild-eyed dreamers" to solving some of biology's toughest challenges.
For us in the AI community, this isn't just a film; it is a masterclass in the scientific method applied to machine learning. It vividly illustrates the strategic pivot from mastering reinforcement learning in video games to deploying "AI for Science," proving that the path to AGI is paved with both crushing failures and euphoric victories. It humanizes the code we write every day, showing the real people behind the neural networks. As we stand on the precipice of Artificial General Intelligence, this story reminds us that the only limit is our imagination; so, what impossible problem will you solve next?
Why it matters: DeepMind’s trajectory provides a definitive blueprint for how abstract AI research can transition into civilization-altering technologies like protein structure prediction. Understanding this history is crucial for developers and researchers aiming to build the next generation of intelligent systems that solve fundamental scientific problems.
Sources:
🔗 https://www.thethinkinggame.com/
🔗 https://www.imdb.com/title/tt32306663/
🔗 https://deepmind.google/discover/blog/the-thinking-game-a-documentary/
Shoppers are adding to cart for the holidays
Peak streaming time continues after Black Friday on Roku, with the weekend after Thanksgiving and the weeks leading up to Christmas seeing record hours of viewing. Roku Ads Manager makes it simple to launch last-minute campaigns targeting viewers who are ready to shop during the holidays. Use first-party audience insights, segment by demographics, and advertise next to the premium ad-supported content your customers are streaming this holiday season.
Read the guide to get your CTV campaign live in time for the holiday rush.
Dozens of state attorneys general urge US Congress not to block AI laws
A bipartisan group of 35 state attorneys general plus DC is pushing back against attempts in Washington to block state-level AI laws. They warn that unregulated AI is already causing real-world harm and insist that states must remain free to regulate deepfakes, discrimination, and safety where Congress fails to act. Big Tech, backed by the Trump administration, is lobbying for one weaker federal standard, turning AI into a battleground over US federalism and corporate power
EU quietly loosens its AI guardrails
The European Commission’s new “Digital Omnibus” would delay key AI Act obligations for “high-risk” systems until 2027, soften cookie rules, and widen the scope for using personal data to train AI, all in the name of competitiveness. Civil society warns this is Brussels caving to Big Tech, while industry still says reforms don’t go far enough. The EU risks shifting from global rule-setter to follower in the AI power game.
Trump’s AI reset reshapes Gulf and European strategy
A new “AI thaw” between the US and Gulf monarchies is accelerating joint projects on chips, cloud, and AI infrastructure, tightening Washington-Riyadh ties after years of friction. For Europe, this creates a squeeze: either it aligns with US-Gulf AI power or carves out its own standards and infrastructure to avoid dependence on both US and Chinese tech. The piece frames AI as a core tool of 21st-century energy, security, and influence politics.
Build smarter, not harder: meet Lindy
Tired of AI that just talks? Lindy actually executes.
Describe your task in plain English, and Lindy handles it—from building booking platforms to managing leads and sending team updates.
AI employees that work 24/7:
Sales automation
Customer support
Operations management
Focus on what matters. Let Lindy handle the rest.
