I am increasingly frustrated about the state of conversation about AI in the UK. It feels like we are in never ending cycles of deluded thinking. Under current trends, Britain will be defenceless against the transformative effects of AI and have no leverage over development at the frontier. We are the most exposed economy to AI in the world. We have very little data centre capacity and few credible plans to expand – and our high energy costs mean we would struggle to even if we tried. I am publishing a paper today on what the UK should do to build that leverage, let the economy capture value from AI and prepare the state to adapt, however AI develops. The three strategies are as follows: 1. Build the infrastructure that gives us leverage. Fix the grid, fix planning, make Britain a place frontier AI can run. 2. Let the economy capture the value of AI. Reform the regulatory, labour and talent barriers that stop British firms adopting AI. 3. Build a state that can adapt to a changing world. Build a tax base that survives the labour market changing, public services that use AI to deliver better, and capacity to weather the upcoming geopolitical storms. Two years ago, we published a similar report on AI - many recommendations of which ended up in the action plan. I remember feeling hopeful, but sadly the fundamental problems facing the country on AI have not shifted. I truly hope that this week of momentum culminates in action. There’s still a narrow path for the UK, but everyday that path is narrowing.

Put the right tools in the hands of ambitious builders and they'll show you things you didn't expect. Great seeing that in action at our @awscloud Global Meeting this week. Thanks to @mza for hosting, @odysseyml's @olivercameron for showing us Agora-2, and @Francessca_V and @steipete, creator of @openclaw and now at @OpenAI, for a great conversation on how AI is changing the way developers build. Grateful to everyone pushing the work forward.

One of the killer applications of world models is to train agents themselves. If a world model can simulate infinite, intelligent worlds, we can push agents to new heights of intelligence and utility. Can't wait to see what's built with Odyssey-3 ❤️

I spent 2015-2023 building self-driving cars. It was so painful. It's now 2026, and we tasked just 1 researcher to take Odyssey-3, our latest world model, and teach it to drive the roads of India with just 20 hours of driving data. And…it works! This is insane.

For those that think safety/alignment researchers are fear-mongering, remember they have access to models with zero safeguards, and those models are far more capable than models you have access to. We all get the lobotomized, mostly controlled intelligence.

World models represent a new AI category, learning world dynamics from visual and auditory experience. This enables advances in robotics, gaming, advertising, defense, education, and retail. Here's @jeffrey_hawke of @odysseyml at @raais

I'll be presenting our work on calibration for video world models at WOOP @ ECCV on Tuesday in the MalmoMassan Exhibit Hall at 15:10 – 16:10. tl;dr if you simulate random physics experiments with SOTA world models the odds you measure are wrong! #ECCV2026#

Breakthroughs emerge not from scaling or data alone, but from incorporating human feedback on preferences, a key lesson from LLMs like GPT-2 applied to world models. This preference learning is an underexplored frontier. Here's Fabian Gura of @odysseyml at @raais 2026

World models don’t compress enough. I think LLMs generalise because they are extremely compressed. “Ship” means both rocket ship and naval ship, which forces the abstraction “large vessel.” RL on boats can then transfer to rockets when the task is about large vessels. And that’s just one token. WMs currently use huge representations that strictly do not require this kind of abstraction. I think it should be possible to train WMs to create factorised representations: s, r. s is the super compressed state: ‘ship’. R is the residual, everything required to reconstruct that specific instance: ‘wooden on water’. When it’s not required one could drop the r and model dynamics purely in the abstract space. One could then decode the instance with the residual of an observation.

🦄 Delighted to announce @EmeraldAI_'s $150 million Series A round, valuing us at more than $1 billion with over $220 million raised to date! Read more in The New York Times: 📢Read our full announcement: 📺Watch our launch on CNBC Squawk Box: 🔗And dive deeper at our website: ⚡Emerald AI is transforming AI data centers into flexible grid allies to unlock 100+GW of power for AI while protecting energy affordability and reliability for local communities. After 18 months of demonstrations—across five data centers all over the world—we've entered our commercial scaling phase and are now deployed at multi-MW, full data center scale with customers spanning the world's biggest AI pioneers, global data center platforms, and leading power utilities. We're proud to collaborate with @NVIDIA, @Digitalrealty, @EPRINews, and others to launch the world's first power-flexible AI Factory—roughly 100MW in Manassas, Virginia—later this year, creating the blueprint for a new generation of flexible DSX AI infrastructure that will be good citizens to power grids and local communities. We're also honored to partner with @SantaClaraPower on the nation's first Flexible Load Interconnection Program (FLIP) to provide additional power capacity to flexible AI data centers. 🤝We are so grateful to our incredible consortium of investors, co-led by @EnergizeCapital and @DCVC, as well as to the extraordinary syndicate of leading global financial and strategic investors. Emerald now counts 12 Fortune 500 Global companies as our investors. Many thanks to Series A investors including @NVIDIA, @SamsungVentures, @aramcoventures, @Siemens, @SalesforceVC, @GEVernova, @RWE_AG, @jeraventures, @ADI_News Ventures, @SabanciHolding, In-Q-Tel, @radicalvcfund, @EnergyImpact_, @lowercarbon, Marunouchi Innovation Partners, @EmCollective, @johndoerr, The Olayan Group, The Temerty Group, @EarthshotVC, @collectiveGBL, General Catalyst's scout fund, and others. 🚀This funding will be instrumental as we rapidly scale commercially across the United States and around the world. Our vision is simple: secure larger and faster power connections for AI while making sure AI gives back to the communities that host it. At Emerald AI, we believe that the biggest new user of electricity—AI data centers—can also be the grid's greatest ally, and a boon for local communities by protecting energy affordability and reliability. Thank you to all of our customers, partners, investors, and our incredible team (grateful to the leadership team pictured below Ayse Coskun Mansi Shah Aroon Vijaykar Shayan Sengupta as well as to the whole fantastic team)—this has already been the journey of a lifetime, and we're just getting started

The winning World Model company won’t be the most focused — it’ll be the broadest. @olivercameron, Co-Founder & CEO of @odysseyml, joins @kaliouby to explain why exploring everything from robotics to healthcare could be the key to winning. Watch:

Take all the water being used by UK data centres and double it. Then double that. Then double that. Then double that. Then double that. Then double that. Then double that. Then double that. Congratulations - you now have half as much water as the UK water system loses every year due to leaks.

Starting a new movement called Hyperoptimism • Technological might is economic might • Good fortune is a civic, moral force • Sovereignty through solvency • Prosperity is invented, not legislated • Power for all: Infinite leverage means boundless agency

We have our first @odysseyml paper CaliBench accepted at TMLR 💃🌍 In our Benchmark, we evaluate whether I2V world models sample from the true statistical distribution of physical outcomes or just generate realistic looking frames. 🧵👇 Congrats @JonathanSadeghi and the team!