The best roguelike games offer a serious but rewarding commitment, inviting you to lose time and time again until you reach new heights. And then you get to repeat the process while retaining knowledge that shapes your future sessions, improving your skills with each subsequent run.The best roguelike games follow the philosophy of randomizing existing game elements and providing a different experience every time you start a new run, with the condition that a game over screen means restarting from scratch. The roguelite distinction is similar, but it involves permanent progression, be it in the form of story, unlockable paths, items, and so on.Our selection of the best roguelike games has something for every player, regardless of platform and subgenre of choice. Our picks intersect with rhythm, deckbuilding, platforming, puzzle, and shooter games, to name a few. The roguelike and roguelite genres keep gaining more ground as pillars for design choices, meaning that some of the games on this list will ring familiar with the selections in our best PS5 games and best PC game lists. You'll also find some overlap with our recommendations for the best indie games to play right now.As the middle of 2025 draws near, choosing the best roguelike games becomes tougher with each passing day. The likes of Nuclear Throne, The Binding of Isaac, and Enter the Gungeon progressively marked their place in the genre years ago. Now, roguelikes and roguelites are in abundance, with game developers coming up with novel spins and pushing the genre forward.Whether you're looking for something more traditional or a unique take on a genre that is brimming with creativity, our best roguelike games list is bound to add a few new experiences to have on your radar. BalatroDeveloper: LocalThunkRelease Date: February 20, 2024Platforms: PlayStation 4, PlayStation 5, Xbox One, Xbox Series X|S, Nintendo Switch, Android, iOS, PCIf you've seen your loved ones spending way too much time on their phones during the past year, it's likely due to LocalThunk's engrossing take on poker. Balatro is deceptively simple: Look at your hand, think of the combination that will net you the most points, and see the numbers go up.As you slowly make your way to tougher levels, gaining additional cards that multiply said score while granting an array of often absurd bonuses, you'll immediately understand the appeal. Balatro is not about poker, nor is it just a roguelike. It's a numbers game where you can rig the rules in your favor. Its challenges are hypnotizing, but once you achieve victory for the first time, it'll take you a while to play anything else. See at Amazon Pacific DriveDeveloper: Ironwood StudiosRelease Date: February 22, 2024Platforms: PlayStation 5, PCPacific Drive is not technically a roguelite, even if it shares quite a few similarities in how you navigate its world, picking up key items and tackling objectives as you escape from weather anomalies from the inside of your car. The Endless Expeditions update, however, does push the game toward that direction.Released on April 3 of this year, Expeditions takes you outside of the campaign and into a randomized map with modifiers and rewards, the latter including cosmetics and unique tools. The trick is that you won't be able to stock up on resources, as you're forced to scavenge and build your inventory from scratch each time.Expeditions only end once you've collected enough anchors on a map. The longer this takes, however, the harder the task will become, with anomalies gearing up in difficulty. If you're looking for an extra challenge or an interesting twist on the Pacific Drive formula, Expeditions is the answer.Fanatical and GameSpot are both owned by Fandom. See at Fanatical Dead CellsDeveloper: Motion TwinRelease Date: August 6, 2018Platforms: PlayStation 4, PlayStation 5, Xbox One, Xbox Series X|S, Nintendo Switch, Android, iOS, PCDead Cells is one of the pioneers of the new wave of roguelites of the past few years. Ever since its conception, developer Motion Twin set out to build an ambitious foundation--prioritizing a breakneck rhythm and flow in movement and attacks, rewarding fast reflexes and improvisation on the spot.The team continued to work on the game over the years, delivering a total of 35 major updates, expanding and ironing out possibly every element inside out. If at some point you think the base game doesn't have enough to offer, there are five DLCs (one of them free), including the Return to Castlevania expansion. As it stands, there might never be another game like Dead Cells, and that's okay. We'll still be playing it for years to come, while also witnessing the team trying out new ideas, such as the co-op roguelite Windblown, which is an early access game to watch. See at Fanatical Deep Rock Galactic: SurvivorDeveloper: Funday GamesRelease Date: February 14, 2024Platforms: PCIf you're a Left 4 Dead 2 fan, chances are that you've heard of Deep Rock Galactic, which takes the structure of the zombie-driven shooter and takes it in a different direction, featuring dwarves in space, alien monsters, and destructible environments. Deep Rock Galactic: Survivor, however, takes things even further.Released in Steam early access back on February 14, 2024, you're taken onto increasingly harder levels where your dwarf of choice attacks automatically. The Survivor-like is presented in a top-down perspective, with the mining mechanic taking center stage to unlock upgrades during runs and improve your chances.There have been four sizable updates, but even in its infancy, developer Funday Games struck gold when mining for resources. The combination of its existing setting with the roguelike genre is an enticing one. See at Steam Hades 2Developer: Supergiant GamesRelease Date: May 6, 2024Platforms: PCDeveloper Supergiant Games has famously never done sequels before--until Hades 2. The sequel follows the events of the first game, in which Zagreus, son of the Greek god of the underworld, fought his way to the surface. Now, his sister Melinöe is tasked with defeating Chronos, the god of time itself, who's posing a threat unknown to everyone to this point.The roguelite is similar in nature to its predecessor, advancing the story with each victorious or failed run, gradually uncovering layer upon layer of new characters, powers to combine and grow stronger with, and even more features to further customize your experience. Hades 2 is still in early access, but there's a ridiculous amount of story and challenges to go through. Plus, the sequel is one of the first games confirmed for the Nintendo Switch 2. The 1.0 release date is yet to be confirmed, but Supergiant is targeting 2025. See at Steam Darkest Dungeon 2Developer: Red Hook StudiosRelease Date: May 8, 2024Platforms: PlayStation 4, PlayStation 5, Xbox One, Xbox Series X|S, Nintendo Switch, PCDarkest Dungeon 2 entered early access in October 2021, with the full launch taking place on May 8. It had a tall order--surpassing the inimitable Darkest Dungeon, a tough roguelike centered around stress as a mechanic, with characters being driven mad or, on occasion, becoming inspired by the embrace of darkness and presence of heinous monsters.The sequel moves away from the structure of its predecessor, where you explored different biomes while managing resources and upgrading a home base. Now, it follows a structure similar to modern roguelite conventions, where you choose from different paths that are labeled with the rewards and challenges awaiting on the roads.While the combat shares some similarities, many new features shake up previous foundations, from playable origin stories for each hero to the affinity system, which takes the stress mechanic of the first game and adds a communal element to it by leading to different relationship archetypes between party members. Despite the change in presentation with more bells and whistles, Darkest Dungeon 2 retains the spirit of the first game--it's a visceral, tough-as-nails adventure that demands patience to overcome. See at Fanatical Rogue Legacy 2Developer: Cellar Door GamesRelease Date: April 28, 2022Platforms: PlayStation 4, PlayStation 5, Xbox One, Xbox Series X|S, Nintendo Switch, PCWhen the original Rogue Legacy launched in 2013, its combination of metroidvania with roguelite elements was already promising. Yet, it pushed things further with a genius feature where, each time you die, one of your children succeeds you. The trick? Everybody has unique traits, from gigantism and baldness to color blindness and dyslexia, which would have gameplay impacts from the beneficial to the hilarious.Rogue Legacy 2 pushes the novelty with even more traits and classes, including a bard and a dragon lancer. The metroidvania aspect has also been improved with the addition of unique items that permanently unlock abilities to further explore the world and unveil its secrets. It's a sequel that successfully ticks every box a sequel must, and does so with a familiar grace and humor that still has a lot to offer over a decade later. See at Steam Spelunky 2Developer: Mossmouth and BlitWorksRelease Date: September 29, 2020Platforms: PlayStation 4, PlayStation 5, Xbox One, Xbox Series X|S, Nintendo Switch, PCSpelunky is the classic go-to recommendation for a roguelike with ironclad design. The platformer with an Indiana Jones spirit made for one of the most challenging roguelites out there, requiring patience and the willingness to study every possible enemy pattern, trap, and miscalculation that can end your run, and then memorize it all for the next attempt.Spelunky 2 is a showcase of refinement upon refinement. Sure, it's touted with a bigger world with new areas to study, multiple routes to find and take advantage of, and an assortment of features meant to subvert long-standing fans' expectations. The sequel's greatest trick, however, is once again proving that meticulously thought design can elevate good ideas to admirable heights. The only way to improve is to keep trying, over and over, until surpassing the obstacles in your way becomes second nature. And then you get to do it all over again in the following area. See at Steam Slay the SpireDeveloper: Mega CritRelease Date: January 23, 2019Platforms: PlayStation 4, PlayStation 5, Xbox One, Xbox Series X|S, Nintendo Switch, Android, iOS, PCDeck-building has become a popular marriage candidate for roguelites, and Slay the Spire might be one of the games responsible for its popularity. The premise is simple: You pick a character who has a predefined card deck and jump into a procedurally generated run. The rewards, obstacles, and challenges roaming your chosen paths are always different, demanding different strategies.As you make your way through a game session, you'll collect more cards for your deck, slowly customizing the options available to fit different playstyles and adapt yourself to the danger at hand. Do you invest in dealing as much damage as possible, neglecting your defense? Do you try to come up with different synergies to create a jack-of-all-trades card deck? There's always something to learn and discover, and the number of combinations available to experiment with is seducingly daunting. The best part? There's a sequel in the works, slated for 2025. See at Steam Risk of Rain 2Developer: Hopoo GamesRelease Date: August 11, 2020Platforms: PlayStation 4, PlayStation 5, Xbox One, Xbox Series X|S, Nintendo Switch, PCRisk of Rain 2 is one of the few roguelites that explored the idea of taking the often pixel art, 2D-driven presentations of the genre and experimenting with 3D instead. Taking the elements that made its predecessor great, from enemy design to the different survivors you control, the change of perspective led to a breath of fresh air in the genre, and one of the most entertaining online games out there.The 1.0 release on August 11, 2020 has been followed up with hefty updates and paid DLC since. You can spend hours just fooling around with friends and seeing how far you can go. Once you get invested in the game's most intricate secrets and mechanics, however, you'll find out that the thrill comes down to becoming an unstoppable force challenged by a dynamic difficulty meter that's constantly keeping you on your toes. The use of 3D, then, isn't a mere novelty--it literally shines a new light on what roguelites can achieve when looked at from a different perspective, adding a different sense of movement and scale to the usual roguelite chaos. See at Steam Crypt of the NecrodancerDeveloper: Brace Yourself GamesRelease Date: April 23, 2015Platforms: PlayStation 4, Xbox One, Nintendo Switch, Android, iOS, PCCrypt of the Necrodancer might be one of the oldest picks of our best roguelike games list, but the reason is simple: No other roguelite has managed to combine the genre conventions with the mechanics of a rhythm game to such success. Every movement and action in the game is tied to the beat of the soundtrack blasting in the background.Enemies have different patterns that you must learn, all while carefully moving on tiles as if you were tapping the floor with your foot following a song. Even if you're not rhythm game savvy, the mix of both genres is accommodating enough to make you a believer--right until you meet a dragon for the first time, that is. After the original release, Cadence of Hyrule followed up on the concept, with the studio collaborating with Nintendo for a different take on The Legend of Zelda. But Crypt of the Necrodancer remains a worthy rogue classic. See at Steam Into the BreachDeveloper: Subset GamesRelease Date: February 27, 2018Platforms: Nintendo Switch, Android, iOS, PCThe alien invasion grows in numbers. You must assemble a squad of a handful of units and try to beat the odds. When you inevitably meet your demise, the last person standing creates a rift and travels to a different timeline. Time to start again.Into the Breach has a steep learning curve. Its design conventions, however, make it worth the effort to learn how to best use the space given to you and how your units can counterattack the alien push. The game is clear about the consequences of your movements on each map grid--whether or not you'll be able to land an attack, if one of your mechs will be caught in a tidal wave or an enemy projectile, and so on.Often, playing the strategy roguelike feels like a series of elaborate board games, where you carefully move pieces and spend your time thinking of the best possible plan. When it all comes together, the satisfaction is unparalleled. But even when you fail and you're sent to another timeline to commence anew, there's a thrill in knowing that next time might be the one if you take the time to analyze your movements and execute with care.Into the Breach is also the rare game to receive a 10/10 from GameSpot. See at Steam Loop HeroDeveloper: Four QuartersRelease Date: March 4, 2021Platforms: Xbox One, Xbox Series X|S, Nintendo Switch, Android, iOS, PCThe concept of loops and the repetitive nature of roguelites are a natural pairing. Loop Hero not only understands it, but extends it further by turning you into a dungeon master of sorts. You're given the choice of a hero and a map layout, as well as a deck of cards. You can place different types of terrain and structures, each having its own pros and cons. But you also need to place enemies, slowly making each loop--where the character walks a full cycle of the map--more intricate to navigate.The randomized nature of roguelites is heightened by giving you agency over the dangers that your hero will have to overcome. With a large number of unlockables to pursue after each session, as well as a captivating presentation, Loop Hero is one of the most inventive roguelites out there. See at Fanatical ReturnalDeveloper: HousemarqueRelease Date: April 30, 2021Platforms: PlayStation 5, PCHousemarque's expertise lies in games with an arcade nature, from Resogun to Nex Machina. At first glance, Returnal seems different--its presentation is powered by a level of production that screams AAA game. Initially introduced as a PlayStation 5 exclusive, the third-person shooter makes an interesting use of the roguelite concept, intertwining story events with each death. Similarly to Hades, meeting your demise rarely means taking a step back.Don't let its prestigious look deceive you, though--Returnal is the living proof of a modern arcade game, taking cues from the bullet hell genre. This means that you must be in constant movement, carefully timing dodges and narrow jumps to avoid a barrage of projectiles coming your way from all directions. A sequel is slated for 2026, but the first game deserves all of your attention. It's one of the best games in the PlayStation catalog, and a thrilling showcase of how a roguelite structure can inform a story, blending into one coexisting vehicle for a narrative that wouldn't work elsewhere. See at Fanatical Shogun ShowdownDeveloper: RoboatinoRelease Date: September 5, 2024Platforms: PlayStation 4, PlayStation 5, Xbox One, Xbox Series X|S, Nintendo Switch, PCDeveloper Roboatino saw the synergy of roguelike and deck-building elements and decided to add turn-based combat to the mix. You command a lone hero who must take positioning and time into consideration to defeat multiple enemies and make it out unscathed.Using an inventive mechanic regarding the tiles you set foot on, Shogun Showdown hides a surprising level of depth, which you gradually uncover the more time you spend with it. You can upgrade and sacrifice different movements and skills in between battles, and as you die, you'll unlock new characters and attacks to experiment with. If you want to test the game's combat by yourself, there's a free prologue available on PC. See at Fanatical Blue PrinceDeveloper: DogubombRelease Date: April 10, 2025Platforms: PlayStation 5, Xbox Series X|S, PCIn Blue Prince, you're a fourteen-year-old boy next in line for an inheritance. The catch is that you first need to explore a manor that changes its inner structure each day, hiding access to the elusive room 46 somewhere within its bowels. Dogubomb's title is part puzzle game, part roguelite, part investigative game.When you start a new day, you're given a number of steps that you can take inside the manor. Upon interacting with a door, the game grants you a random selection of rooms, each containing a puzzle, resource items, or a clue to a larger mystery. Sometimes all three of them. The deeper you plunge into Blue Prince, the higher the chances of becoming engrossed by the sheer amount of layers upon layers of puzzles to solve. See at Fanatical Caves of QudDeveloper: Freehold GamesRelease Date: December 5, 2024Platforms: PCDevelopment for Caves of Qud began back in 2007, with the first public beta being released to the world in 2010. Then, after almost a decade in Steam early access, the game was fully launched on December 5, 2024. The science-fantasy roguelike is brimming with emergent stories, offering a deeply simulated world where you can shape the environment as you see fit, join one of over 70 factions, or simply get lost in the overwhelming number of possible actions and outcomes available.In Caves of Qud, every NPC and monster is as fully simulated as you, meaning that they all have their own skills, equipment, body parts, and levels. The body parts are important, as there are multiple mutations at play, from two heads to the power of cloning oneself. The sandbox nature and painstaking level of detail have added an unmatched identity to the roguelike over its long lifespan. Now, there's never been a better time to take a plunge and become a part of its labyrinthine systems and intricate synergies that are happening in the game without your input. See at Steam FTL: Faster Than LightDeveloper: Subset GamesRelease Date: September 14, 2021Platforms: iOS, PCCommanding a party is a classic go-to for roguelites. Being the captain of a spaceship where you must attend to your crew and rooms individually, however, is an idea that is still novel to this day. Before Into the Breach, developer Subset Games came up with a different adventure in outer space.Presented with randomly generated galaxies, you must pick different paths to warp to, taking care and managing your spaceship in your ventures. Everything from the state of the hull to the level of oxygen must be accounted for. While you might be lucky with the galaxy destinations you pick, your crew will inevitably have to confront other ships.It's during these moments that FTL: Faster Than Light showcases its exhilarating mix of mechanics, forcing you to act fast by putting up fires, deciding which rooms to open and which ones to close, and sending crew members to repair the ship, all while using similar strategies on the enemy at hand. There's no other game like FTL. See at Steam InscryptionDeveloper: Daniel Mullins GamesRelease Date: October 19, 2021Platforms: PlayStation 4, PlayStation 5, Xbox One, Xbox Series X|S, Nintendo Switch, PCThe creator of the eerie Pony Island took a stab at the roguelike genre in 2021 with Inscryption. At first, you don't know where you are--all you can see is the inside of a dark cabin, and a strange figure that invites you to play a card game. The creepiness sets in more and more over time, as you use cards depicting animals that seem to be alive, trying to make progress in the game while also looking for an opportunity to try and figure out how to escape the cabin.Inscription has multiple twists that are best left as a secret. If you want a general indication of what to expect, however, this roguelike pulls you into an obscure setting that becomes darker the more time you spend with it, subverting existing genre conventions and familiar mechanics with a horror twist. See at Fanatical Dome KeeperDeveloper: BippinbitsRelease Date: September 27, 2022Platforms: PCDome Keeper is a great game for people who enjoy multitasking. As the name implies, you must protect a dome from enemy attacks. In order to do so, you must dig underneath the surface to search for resources and artifacts, which are used to choose upgrades and different ways to defend your base.Enemies won't just sit and wait, however. You only have a limited time to dig in between attack waves. Picking your upgrades carefully will make or break your chances of survival to gain another chance at plunging through the surface and build better defenses. Dome Keeper is an ambitious survival game that takes cue from roguelikes and tower defense games to create a different kind of challenge to overcome. See at Fanatical Monster TrainDeveloper: Shiny ShoeRelease Date: May 20, 2020Platforms: PlayStation 5, Xbox One, Xbox Series X|S, Nintendo Switch, iOS, PCMonster Train is the distant cousin of Slay the Spire. You're given a deck of cards that you slowly grow during runs, as well as different paths to choose from. The twist is how combat encounters are structured, turning elements like positioning and card strategies on their head.Whenever you face enemies, you're presented with a large vertical structure that has three play fields. You must carefully plan where to place your cards to defend the train's core. It's an idea that's pushed to its limit time and time again with the cards available and the ways in which enemies can also strategize around your defenses. Monster Train is a prime example of how much innovation is still in the genre. See at Fanatical Vampire SurvivorsDeveloper: PoncleRelease Date: October 20, 2022Platforms: PlayStation 4, PlayStation 5, Xbox One, Xbox Series X|S, Nintendo Switch, Android, iOS, PCTaking inspiration from a mobile game called Magic Survival, in which the main character attacked automatically without the player's input, developer Poncle leaned on a fascination for Castlevania's aesthetic to iterate on the concept. The simple idea led to an absurd amount of characters, levels, and items to unlock--including, of course, an official collaboration with Konami to include more than just indirect references to the Castlevania series.Playing a Vampire Survivors stage can last anything from 15 to 20 or 30 minutes--if you can survive long enough with the items you've acquired. Even if you fail, however, you're constantly unlocking items, characters, and stages to explore in subsequent runs. It's a game that's best described as a Pandora's Box equivalent. Especially considering the chaos that it will likely inflict on your schedule. See at Steam Hitman World of Assassination: Freelancer ModeDeveloper: IO InteractiveRelease Date: January 26, 2023Platforms: PlayStation 4, PlayStation 5, Xbox One, Xbox Series X|S, PCThe Hitman series has always excelled in presenting puzzle boxes that double down as playgrounds to blend in, follow your target, and take them down in increasingly ridiculous ways. Freelancer Mode, introduced to Hitman World of Assassination on January 26, 2023, adds a different spin to the premise with a persistent and highly replayable experience.In this mode, targets are always random, and you're given access to different bonus objectives. Making use of a hub exclusive to Freelancer, Agent 47 must choose a crime syndicate to pursue, which sets the mood of the campaign, and then get started without any equipment or weapons. If you've already mastered the classic Hitman levels or you just want a different, ridiculously polished roguelike experience, Freelancer Mode is a distinct and ambitious take on the genre. Worth mentioning that Hitman World of Assassination is also slated to launch on Nintendo Switch 2. See at Steam InkboundDeveloper: Shiny ShoeRelease Date: April 9, 2024Platforms: PCFrom the makers of Monster Train comes Inkbound, a turn-based tactical roguelike that offers co-op, and synergies that are heightened by coordinating with others online. Players can move freely and act simultaneously in multiplayer, picking from eight different classes to try out different combinations.There's also a deck-building element of sorts involved in the vein of draftable abilities, upgrades, and an array of items to use in-game. It's an intriguing blend of genres that might take some time to get used to. Once things click, however, you and the rest of the party will be working in tandem to see how far you can push your strategies. See at Steam God of War Ragnarok: ValhallaDeveloper: Santa Monica StudiosRelease Date: December 12, 2023Platforms: PlayStation 4, PlayStation 5, PCGod of War Ragnarok is a behemoth of an action-RPG, featuring a lengthy campaign across open areas with side quests, collectibles, and dozens of corners to explore. If you're looking for a break from the main story or just want to try a new roguelite that elevates some of the game's strengths in a different structure, the free Valhalla DLC is a must play.Taking place after the events of the main story (although you can jump into it at any point from the main menu without worrying about spoilers of the main campaign), Valhalla has Kratos going through a series of punishing trials, which slowly serve as a therapy session of sorts as he reminiscences of the events that happened during the original trilogy of God of War games.The bite-sized rendition of combat encounters shines a different light on combat mechanics that you may already be familiar with. And, taking cues from the likes of Hades and Returnal, each death pushes the story forward in meaningful ways. Valhalla is a rare combination of genres that shouldn't work as well as they do, and it deserves your time before the next adventure of Kratos inevitably rolls in. See at Fanatical NoitaDeveloper: Nolla GamesRelease Date: October 15, 2020Platforms: PCAfter a short early access period in 2019, Noita was fully released in October 2020, and provided an escape from the horrors of that year with, well, more horrors. Its punishing difficulty and procedurally-generated worlds provide deep and engaging gameplay. Noita places you in the shoes of a wand-wielding alchemist navigating a deadly, destructible world. The seemingly simple pixel art style leads to complexity where water flows, acid burns, fire spreads, and cave-ins can be triggered by a single misplaced explosive.Unlike many roguelikes that focus on stat-based progression, Noita appeals to player creativity and experimentation. The game's wand crafting system is a particular highlight, allowing you to mix and match spells, modifiers, and triggers to create devastating effects or strange contraptions that alter the way the game plays with every decision. The fun of Noita is learning how to break it in delightfully chaotic ways. This customization means each run feels unique, not just because of the randomized levels, but because the tools at your disposal and the way you choose to implement them define each playthrough. Noita is deeper than it first appears, with plenty of secrets and mysteries to uncover. Its expansive hidden world and lore begs to be uncovered, and the community engagement and conversation around it continues even half a decade on from its release. Experiment with fire, mess around with slime, and Noita will reward your curiosity. You'll find that, most of the time, death in Noita stems from your own creation, giving the game a cruel sense of irony and punishment. Despite that, or perhaps even because of that, Noita is a beautiful experience. See at Steam

You might think that a nuclear explosion is not something you need a detector for, but clearly not everyone agrees. [Bigcrimping] has not only built one, the BhangmeterV2, but he …read more

In April 2025, at the ITER Private Sector Fusion Workshop in Cadarache, something remarkable unfolded. In a room filled with scientists, engineers and software visionaries, the line between big science and commercial innovation began to blur.   Three organisations – Microsoft Research, Arena and Brigantium Engineering – shared how artificial intelligence (AI), already transforming everything from language models to logistics, is now stepping into a new role: helping humanity to unlock the power of nuclear fusion.  Each presenter addressed a different part of the puzzle, but the message was the same: AI isn’t just a buzzword anymore. It’s becoming a real tool – practical, powerful and indispensable – for big science and engineering projects, including fusion.  “If we think of the agricultural revolution and the industrial revolution, the AI revolution is next – and it’s coming at a pace which is unprecedented,” said Kenji Takeda, director of research incubations at Microsoft Research.  Microsoft’s collaboration with ITER is already in motion. Just a month before the workshop, the two teams signed a Memorandum of Understanding (MoU) to explore how AI can accelerate research and development. This follows ITER’s initial use of Microsoft technology to empower their teams. A chatbot in Azure OpenAI service was developed to help staff navigate technical knowledge, on more than a million ITER documents, using natural conversation. GitHub Copilot assists with coding, while AI helps to resolve IT support tickets – those everyday but essential tasks that keep the lights on.  But Microsoft’s vision goes deeper. Fusion demands materials that can survive extreme conditions – heat, radiation, pressure – and that’s where AI shows a different kind of potential. MatterGen, a Microsoft Research generative AI model for materials, designs entirely new materials based on specific properties. “It’s like ChatGPT,” said Takeda, “but instead of ‘Write me a poem’, we ask it to design a material that can survive as the first wall of a fusion reactor.”  The next step? MatterSim – a simulation tool that predicts how these imagined materials will behave in the real world. By combining generation and simulation, Microsoft hopes to uncover materials that don’t yet exist in any catalogue.  While Microsoft tackles the atomic scale, Arena is focused on a different challenge: speeding up hardware development. As general manager Michael Frei put it: “Software innovation happens in seconds. In hardware, that loop can take months – or years.”  Arena’s answer is Atlas, a multimodal AI platform that acts as an extra set of hands – and eyes – for engineers. It can read data sheets, interpret lab results, analyse circuit diagrams and even interact with lab equipment through software interfaces. “Instead of adjusting an oscilloscope manually,” said Frei, “you can just say, ‘Verify the I2C [inter integrated circuit] protocol’, and Atlas gets it done.”  It doesn’t stop there. Atlas can write and adapt firmware on the fly, responding to real-time conditions. That means tighter feedback loops, faster prototyping and fewer late nights in the lab. Arena aims to make building hardware feel a little more like writing software – fluid, fast and assisted by smart tools.  Fusion, of course, isn’t just about atoms and code – it’s also about construction. Gigantic, one-of-a-kind machines don’t build themselves. That’s where Brigantium Engineering comes in. Founder Lynton Sutton explained how his team uses “4D planning” – a marriage of 3D CAD models and detailed construction schedules – to visualise how everything comes together over time. “Gantt charts are hard to interpret. 3D models are static. Our job is to bring those together,” he said.  The result is a time-lapse-style animation that shows the construction process step by step. It’s proven invaluable for safety reviews and stakeholder meetings. Rather than poring over spreadsheets, teams can simply watch the plan come to life.  And there’s more. Brigantium is bringing these models into virtual reality using Unreal Engine – the same one behind many video games. One recent model recreated ITER’s tokamak pit using drone footage and photogrammetry. The experience is fully interactive and can even run in a web browser. “We’ve really improved the quality of the visualisation,” said Sutton. “It’s a lot smoother; the textures look a lot better. Eventually, we’ll have this running through a web browser, so anybody on the team can just click on a web link to navigate this 4D model.”  Looking forward, Sutton believes AI could help automate the painstaking work of syncing schedules with 3D models. One day, these simulations could reach all the way down to individual bolts and fasteners – not just with impressive visuals, but with critical tools for preventing delays.  Despite the different approaches, one theme ran through all three presentations: AI isn’t just a tool for office productivity. It’s becoming a partner in creativity, problem-solving and even scientific discovery.  Takeda mentioned that Microsoft is experimenting with “world models” inspired by how video games simulate physics. These models learn about the physical world by watching pixels in the form of videos of real phenomena such as plasma behaviour. “Our thesis is that if you showed this AI videos of plasma, it might learn the physics of plasmas,” he said.  It sounds futuristic, but the logic holds. The more AI can learn from the world, the more it can help us understand it – and perhaps even master it. At its heart, the message from the workshop was simple: AI isn’t here to replace the scientist, the engineer or the planner; it’s here to help, and to make their work faster, more flexible and maybe a little more fun. As Takeda put it: “Those are just a few examples of how AI is starting to be used at ITER. And it’s just the start of that journey.”  If these early steps are any indication, that journey won’t just be faster – it might also be more inspired. 

June 13, 20256 min readAir-Conditioning Can Surprisingly Help the Power Grid during Extreme HeatSwitching on air-conditioning during extreme heat doesn’t have to make us feel guilty—it can actually boost power grid reliability and help bring more renewable energy onlineBy Johanna Mathieu & The Conversation US Imagedepotpro/Getty ImagesThe following essay is reprinted with permission from The Conversation, an online publication covering the latest research.As summer arrives, people are turning on air conditioners in most of the U.S. But if you’re like me, you always feel a little guilty about that. Past generations managed without air conditioning – do I really need it? And how bad is it to use all this electricity for cooling in a warming world?If I leave my air conditioner off, I get too hot. But if everyone turns on their air conditioner at the same time, electricity demand spikes, which can force power grid operators to activate some of the most expensive, and dirtiest, power plants. Sometimes those spikes can ask too much of the grid and lead to brownouts or blackouts.On supporting science journalismIf you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.Research I recently published with a team of scholars makes me feel a little better, though. We have found that it is possible to coordinate the operation of large numbers of home air-conditioning units, balancing supply and demand on the power grid – and without making people endure high temperatures inside their homes.Studies along these lines, using remote control of air conditioners to support the grid, have for many years explored theoretical possibilities like this. However, few approaches have been demonstrated in practice and never for such a high-value application and at this scale. The system we developed not only demonstrated the ability to balance the grid on timescales of seconds, but also proved it was possible to do so without affecting residents’ comfort.The benefits include increasing the reliability of the power grid, which makes it easier for the grid to accept more renewable energy. Our goal is to turn air conditioners from a challenge for the power grid into an asset, supporting a shift away from fossil fuels toward cleaner energy.Adjustable equipmentMy research focuses on batteries, solar panels and electric equipment – such as electric vehicles, water heaters, air conditioners and heat pumps – that can adjust itself to consume different amounts of energy at different times.Originally, the U.S. electric grid was built to transport electricity from large power plants to customers’ homes and businesses. And originally, power plants were large, centralized operations that burned coal or natural gas, or harvested energy from nuclear reactions. These plants were typically always available and could adjust how much power they generated in response to customer demand, so the grid would be balanced between power coming in from producers and being used by consumers.But the grid has changed. There are more renewable energy sources, from which power isn’t always available – like solar panels at night or wind turbines on calm days. And there are the devices and equipment I study. These newer options, called “distributed energy resources,” generate or store energy near where consumers need it – or adjust how much energy they’re using in real time.One aspect of the grid hasn’t changed, though: There’s not much storage built into the system. So every time you turn on a light, for a moment there’s not enough electricity to supply everything that wants it right then: The grid needs a power producer to generate a little more power. And when you turn off a light, there’s a little too much: A power producer needs to ramp down.The way power plants know what real-time power adjustments are needed is by closely monitoring the grid frequency. The goal is to provide electricity at a constant frequency – 60 hertz – at all times. If more power is needed than is being produced, the frequency drops and a power plant boosts output. If there’s too much power being produced, the frequency rises and a power plant slows production a little. These actions, a process called “frequency regulation,” happen in a matter of seconds to keep the grid balanced.This output flexibility, primarily from power plants, is key to keeping the lights on for everyone.Finding new optionsI’m interested in how distributed energy resources can improve flexibility in the grid. They can release more energy, or consume less, to respond to the changing supply or demand, and help balance the grid, ensuring the frequency remains near 60 hertz.Some people fear that doing so might be invasive, giving someone outside your home the ability to control your battery or air conditioner. Therefore, we wanted to see if we could help balance the grid with frequency regulation using home air-conditioning units rather than power plants – without affecting how residents use their appliances or how comfortable they are in their homes.From 2019 to 2023, my group at the University of Michigan tried this approach, in collaboration with researchers at Pecan Street Inc., Los Alamos National Laboratory and the University of California, Berkeley, with funding from the U.S. Department of Energy Advanced Research Projects Agency-Energy.We recruited 100 homeowners in Austin, Texas, to do a real-world test of our system. All the homes had whole-house forced-air cooling systems, which we connected to custom control boards and sensors the owners allowed us to install in their homes. This equipment let us send instructions to the air-conditioning units based on the frequency of the grid.Before I explain how the system worked, I first need to explain how thermostats work. When people set thermostats, they pick a temperature, and the thermostat switches the air-conditioning compressor on and off to maintain the air temperature within a small range around that set point. If the temperature is set at 68 degrees, the thermostat turns the AC on when the temperature is, say, 70, and turns it off when it’s cooled down to, say, 66.Every few seconds, our system slightly changed the timing of air-conditioning compressor switching for some of the 100 air conditioners, causing the units’ aggregate power consumption to change. In this way, our small group of home air conditioners reacted to grid changes the way a power plant would – using more or less energy to balance the grid and keep the frequency near 60 hertz.Moreover, our system was designed to keep home temperatures within the same small temperature range around the set point.Testing the approachWe ran our system in four tests, each lasting one hour. We found two encouraging results.First, the air conditioners were able to provide frequency regulation at least as accurately as a traditional power plant. Therefore, we showed that air conditioners could play a significant role in increasing grid flexibility. But perhaps more importantly – at least in terms of encouraging people to participate in these types of systems – we found that we were able to do so without affecting people’s comfort in their homes.We found that home temperatures did not deviate more than 1.6 Fahrenheit from their set point. Homeowners were allowed to override the controls if they got uncomfortable, but most didn’t. For most tests, we received zero override requests. In the worst case, we received override requests from two of the 100 homes in our test.In practice, this sort of technology could be added to commercially available internet-connected thermostats. In exchange for credits on their energy bills, users could choose to join a service run by the thermostat company, their utility provider or some other third party.Then people could turn on the air conditioning in the summer heat without that pang of guilt, knowing they were helping to make the grid more reliable and more capable of accommodating renewable energy sources – without sacrificing their own comfort in the process.This article was originally published on The Conversation. Read the original article.

Transcript [MUSIC]      [BOOK PASSAGE]   PETER LEE: “In ‘The Little Black Bag,’ a classic science fiction story, a high-tech doctor’s kit of the future is accidentally transported back to the 1950s, into the shaky hands of a washed-up, alcoholic doctor. The ultimate medical tool, it redeems the doctor wielding it, allowing him to practice gratifyingly heroic medicine. … The tale ends badly for the doctor and his treacherous assistant, but it offered a picture of how advanced technology could transform medicine—powerful when it was written nearly 75 years ago and still so today. What would be the Al equivalent of that little black bag? At this moment when new capabilities are emerging, how do we imagine them into medicine?”   [END OF BOOK PASSAGE]     [THEME MUSIC]     This is The AI Revolution in Medicine, Revisited. I’m your host, Peter Lee.    Shortly after OpenAI’s GPT-4 was publicly released, Carey Goldberg, Dr. Zak Kohane, and I published The AI Revolution in Medicine to help educate the world of healthcare and medical research about the transformative impact this new generative AI technology could have. But because we wrote the book when GPT-4 was still a secret, we had to speculate. Now, two years later, what did we get right, and what did we get wrong?     In this series, we’ll talk to clinicians, patients, hospital administrators, and others to understand the reality of AI in the field and where we go from here.   [THEME MUSIC FADES] The book passage I read at the top is from “Chapter 10: The Big Black Bag.”  In imagining AI in medicine, Carey, Zak, and I included in our book two fictional accounts. In the first, a medical resident consults GPT-4 on her personal phone as the patient in front of her crashes. Within seconds, it offers an alternate response based on recent literature. In the second account, a 90-year-old woman with several chronic conditions is living independently and receiving near-constant medical support from an AI aide.    In our conversations with the guests we’ve spoken to so far, we’ve caught a glimpse of these predicted futures, seeing how clinicians and patients are actually using AI today and how developers are leveraging the technology in the healthcare products and services they’re creating. In fact, that first fictional account isn’t so fictional after all, as most of the doctors in the real world actually appear to be using AI at least occasionally—and sometimes much more than occasionally—to help in their daily clinical work. And as for the second fictional account, which is more of a science fiction account, it seems we are indeed on the verge of a new way of delivering and receiving healthcare, though the future is still very much open.  As we continue to examine the current state of AI in healthcare and its potential to transform the field, I’m pleased to welcome Bill Gates and Sébastien Bubeck.   Bill may be best known as the co-founder of Microsoft, having created the company with his childhood friend Paul Allen in 1975. He’s now the founder of Breakthrough Energy, which aims to advance clean energy innovation, and TerraPower, a company developing groundbreaking nuclear energy and science technologies. He also chairs the world’s largest philanthropic organization, the Gates Foundation, and focuses on solving a variety of health challenges around the globe and here at home.  Sébastien is a research lead at OpenAI. He was previously a distinguished scientist, vice president of AI, and a colleague of mine here at Microsoft, where his work included spearheading the development of the family of small language models known as Phi. While at Microsoft, he also coauthored the discussion-provoking 2023 paper “Sparks of Artificial General Intelligence,” which presented the results of early experiments with GPT-4 conducted by a small team from Microsoft Research.    [TRANSITION MUSIC]   Here’s my conversation with Bill Gates and Sébastien Bubeck.  LEE: Bill, welcome.  BILL GATES: Thank you.  LEE: Seb …  SÉBASTIEN BUBECK: Yeah. Hi, hi, Peter. Nice to be here.  LEE: You know, one of the things that I’ve been doing just to get the conversation warmed up is to talk about origin stories, and what I mean about origin stories is, you know, what was the first contact that you had with large language models or the concept of generative AI that convinced you or made you think that something really important was happening?  And so, Bill, I think I’ve heard the story about, you know, the time when the OpenAI folks—Sam Altman, Greg Brockman, and others—showed you something, but could we hear from you what those early encounters were like and what was going through your mind?   GATES: Well, I’d been visiting OpenAI soon after it was created to see things like GPT-2 and to see the little arm they had that was trying to match human manipulation and, you know, looking at their games like Dota that they were trying to get as good as human play. And honestly, I didn’t think the language model stuff they were doing, even when they got to GPT-3, would show the ability to learn, you know, in the same sense that a human reads a biology book and is able to take that knowledge and access it not only to pass a test but also to create new medicines.  And so my challenge to them was that if their LLM could get a five on the advanced placement biology test, then I would say, OK, it took biologic knowledge and encoded it in an accessible way and that I didn’t expect them to do that very quickly but it would be profound.   And it was only about six months after I challenged them to do that, that an early version of GPT-4 they brought up to a dinner at my house, and in fact, it answered most of the questions that night very well. The one it got totally wrong, we were … because it was so good, we kept thinking, Oh, we must be wrong. It turned out it was a math weakness [LAUGHTER] that, you know, we later understood that that was an area of, weirdly, of incredible weakness of those early models. But, you know, that was when I realized, OK, the age of cheap intelligence was at its beginning.  LEE: Yeah. So I guess it seems like you had something similar to me in that my first encounters, I actually harbored some skepticism. Is it fair to say you were skeptical before that?  GATES: Well, the idea that we’ve figured out how to encode and access knowledge in this very deep sense without even understanding the nature of the encoding, …  LEE: Right.   GATES: … that is a bit weird.   LEE: Yeah.  GATES: We have an algorithm that creates the computation, but even say, OK, where is the president’s birthday stored in there? Where is this fact stored in there? The fact that even now when we’re playing around, getting a little bit more sense of it, it’s opaque to us what the semantic encoding is, it’s, kind of, amazing to me. I thought the invention of knowledge storage would be an explicit way of encoding knowledge, not an implicit statistical training.  LEE: Yeah, yeah. All right. So, Seb, you know, on this same topic, you know, I got—as we say at Microsoft—I got pulled into the tent. [LAUGHS]  BUBECK: Yes.   LEE: Because this was a very secret project. And then, um, I had the opportunity to select a small number of researchers in MSR [Microsoft Research] to join and start investigating this thing seriously. And the first person I pulled in was you.  BUBECK: Yeah.  LEE: And so what were your first encounters? Because I actually don’t remember what happened then.  BUBECK: Oh, I remember it very well. [LAUGHS] My first encounter with GPT-4 was in a meeting with the two of you, actually. But my kind of first contact, the first moment where I realized that something was happening with generative AI, was before that. And I agree with Bill that I also wasn’t too impressed by GPT-3.  I though that it was kind of, you know, very naturally mimicking the web, sort of parroting what was written there in a nice way. Still in a way which seemed very impressive. But it wasn’t really intelligent in any way. But shortly after GPT-3, there was a model before GPT-4 that really shocked me, and this was the first image generation model, DALL-E 1.  So that was in 2021. And I will forever remember the press release of OpenAI where they had this prompt of an avocado chair and then you had this image of the avocado chair. [LAUGHTER] And what really shocked me is that clearly the model kind of “understood” what is a chair, what is an avocado, and was able to merge those concepts.  So this was really, to me, the first moment where I saw some understanding in those models.   LEE: So this was, just to get the timing right, that was before I pulled you into the tent.  BUBECK: That was before. That was like a year before.  LEE: Right.   BUBECK: And now I will tell you how, you know, we went from that moment to the meeting with the two of you and GPT-4.  So once I saw this kind of understanding, I thought, OK, fine. It understands concept, but it’s still not able to reason. It cannot—as, you know, Bill was saying—it cannot learn from your document. It cannot reason.   So I set out to try to prove that. You know, this is what I was in the business of at the time, trying to prove things in mathematics. So I was trying to prove that basically autoregressive transformers could never reason. So I was trying to prove this. And after a year of work, I had something reasonable to show. And so I had the meeting with the two of you, and I had this example where I wanted to say, there is no way that an LLM is going to be able to do x.  And then as soon as I … I don’t know if you remember, Bill. But as soon as I said that, you said, oh, but wait a second. I had, you know, the OpenAI crew at my house recently, and they showed me a new model. Why don’t we ask this new model this question?   LEE: Yeah. BUBECK: And we did, and it solved it on the spot. And that really, honestly, just changed my life. Like, you know, I had been working for a year trying to say that this was impossible. And just right there, it was shown to be possible.   LEE: [LAUGHS] One of the very first things I got interested in—because I was really thinking a lot about healthcare—was healthcare and medicine.  And I don’t know if the two of you remember, but I ended up doing a lot of tests. I ran through, you know, step one and step two of the US Medical Licensing Exam. Did a whole bunch of other things. I wrote this big report. It was, you know, I can’t remember … a couple hundred pages.   And I needed to share this with someone. I didn’t … there weren’t too many people I could share it with. So I sent, I think, a copy to you, Bill. Sent a copy to you, Seb.   I hardly slept for about a week putting that report together. And, yeah, and I kept working on it. But I was far from alone. I think everyone who was in the tent, so to speak, in those early days was going through something pretty similar. All right. So I think … of course, a lot of what I put in the report also ended up being examples that made it into the book.  But the main purpose of this conversation isn’t to reminisce about [LAUGHS] or indulge in those reminiscences but to talk about what’s happening in healthcare and medicine. And, you know, as I said, we wrote this book. We did it very, very quickly. Seb, you helped. Bill, you know, you provided a review and some endorsements.  But, you know, honestly, we didn’t know what we were talking about because no one had access to this thing. And so we just made a bunch of guesses. So really, the whole thing I wanted to probe with the two of you is, now with two years of experience out in the world, what, you know, what do we think is happening today?  You know, is AI actually having an impact, positive or negative, on healthcare and medicine? And what do we now think is going to happen in the next two years, five years, or 10 years? And so I realize it’s a little bit too abstract to just ask it that way. So let me just try to narrow the discussion and guide us a little bit.   Um, the kind of administrative and clerical work, paperwork, around healthcare—and we made a lot of guesses about that—that appears to be going well, but, you know, Bill, I know we’ve discussed that sometimes that you think there ought to be a lot more going on. Do you have a viewpoint on how AI is actually finding its way into reducing paperwork?  GATES: Well, I’m stunned … I don’t think there should be a patient-doctor meeting where the AI is not sitting in and both transcribing, offering to help with the paperwork, and even making suggestions, although the doctor will be the one, you know, who makes the final decision about the diagnosis and whatever prescription gets done.   It’s so helpful. You know, when that patient goes home and their, you know, son who wants to understand what happened has some questions, that AI should be available to continue that conversation. And the way you can improve that experience and streamline things and, you know, involve the people who advise you. I don’t understand why that’s not more adopted, because there you still have the human in the loop making that final decision.  But even for, like, follow-up calls to make sure the patient did things, to understand if they have concerns and knowing when to escalate back to the doctor, the benefit is incredible. And, you know, that thing is ready for prime time. That paradigm is ready for prime time, in my view.  LEE: Yeah, there are some good products, but it seems like the number one use right now—and we kind of got this from some of the previous guests in previous episodes—is the use of AI just to respond to emails from patients. [LAUGHTER] Does that make sense to you?  BUBECK: Yeah. So maybe I want to second what Bill was saying but maybe take a step back first. You know, two years ago, like, the concept of clinical scribes, which is one of the things that we’re talking about right now, it would have sounded, in fact, it sounded two years ago, borderline dangerous. Because everybody was worried about hallucinations. What happened if you have this AI listening in and then it transcribes, you know, something wrong?  Now, two years later, I think it’s mostly working. And in fact, it is not yet, you know, fully adopted. You’re right. But it is in production. It is used, you know, in many, many places. So this rate of progress is astounding because it wasn’t obvious that we would be able to overcome those obstacles of hallucination. It’s not to say that hallucinations are fully solved. In the case of the closed system, they are.   Now, I think more generally what’s going on in the background is that there is something that we, that certainly I, underestimated, which is this management overhead. So I think the reason why this is not adopted everywhere is really a training and teaching aspect. People need to be taught, like, those systems, how to interact with them.  And one example that I really like, a study that recently appeared where they tried to use ChatGPT for diagnosis and they were comparing doctors without and with ChatGPT (opens in new tab). And the amazing thing … so this was a set of cases where the accuracy of the doctors alone was around 75%. ChatGPT alone was 90%. So that’s already kind of mind blowing. But then the kicker is that doctors with ChatGPT was 80%.   Intelligence alone is not enough. It’s also how it’s presented, how you interact with it. And ChatGPT, it’s an amazing tool. Obviously, I absolutely love it. But it’s not … you don’t want a doctor to have to type in, you know, prompts and use it that way.  It should be, as Bill was saying, kind of running continuously in the background, sending you notifications. And you have to be really careful of the rate at which those notifications are being sent. Because if they are too frequent, then the doctor will learn to ignore them. So you have to … all of those things matter, in fact, at least as much as the level of intelligence of the machine.  LEE: One of the things I think about, Bill, in that scenario that you described, doctors do some thinking about the patient when they write the note. So, you know, I’m always a little uncertain whether it’s actually … you know, you wouldn’t necessarily want to fully automate this, I don’t think. Or at least there needs to be some prompt to the doctor to make sure that the doctor puts some thought into what happened in the encounter with the patient. Does that make sense to you at all?  GATES: At this stage, you know, I’d still put the onus on the doctor to write the conclusions and the summary and not delegate that.  The tradeoffs you make a little bit are somewhat dependent on the situation you’re in. If you’re in Africa, So, yes, the doctor’s still going to have to do a lot of work, but just the quality of letting the patient and the people around them interact and ask questions and have things explained, that alone is such a quality improvement. It’s mind blowing.   LEE: So since you mentioned, you know, Africa—and, of course, this touches on the mission and some of the priorities of the Gates Foundation and this idea of democratization of access to expert medical care—what’s the most interesting stuff going on right now? Are there people and organizations or technologies that are impressing you or that you’re tracking?  GATES: Yeah. So the Gates Foundation has given out a lot of grants to people in Africa doing education, agriculture but more healthcare examples than anything. And the way these things start off, they often start out either being patient-centric in a narrow situation, like, OK, I’m a pregnant woman; talk to me. Or, I have infectious disease symptoms; talk to me. Or they’re connected to a health worker where they’re helping that worker get their job done. And we have lots of pilots out, you know, in both of those cases.   The dream would be eventually to have the thing the patient consults be so broad that it’s like having a doctor available who understands the local things.   LEE: Right.   GATES: We’re not there yet. But over the next two or three years, you know, particularly given the worsening financial constraints against African health systems, where the withdrawal of money has been dramatic, you know, figuring out how to take this—what I sometimes call “free intelligence”—and build a quality health system around that, we will have to be more radical in low-income countries than any rich country is ever going to be.   LEE: Also, there’s maybe a different regulatory environment, so some of those things maybe are easier? Because right now, I think the world hasn’t figured out how to and whether to regulate, let’s say, an AI that might give a medical diagnosis or write a prescription for a medication.  BUBECK: Yeah. I think one issue with this, and it’s also slowing down the deployment of AI in healthcare more generally, is a lack of proper benchmark. Because, you know, you were mentioning the USMLE [United States Medical Licensing Examination], for example. That’s a great test to test human beings and their knowledge of healthcare and medicine. But it’s not a great test to give to an AI.  It’s not asking the right questions. So finding what are the right questions to test whether an AI system is ready to give diagnosis in a constrained setting, that’s a very, very important direction, which to my surprise, is not yet accelerating at the rate that I was hoping for.  LEE: OK, so that gives me an excuse to get more now into the core AI tech because something I’ve discussed with both of you is this issue of what are the right tests. And you both know the very first test I give to any new spin of an LLM is I present a patient, the results—a mythical patient—the results of my physical exam, my mythical physical exam. Maybe some results of some initial labs. And then I present or propose a differential diagnosis. And if you’re not in medicine, a differential diagnosis you can just think of as a prioritized list of the possible diagnoses that fit with all that data. And in that proposed differential, I always intentionally make two mistakes.  I make a textbook technical error in one of the possible elements of the differential diagnosis, and I have an error of omission. And, you know, I just want to know, does the LLM understand what I’m talking about? And all the good ones out there do now. But then I want to know, can it spot the errors? And then most importantly, is it willing to tell me I’m wrong, that I’ve made a mistake?   That last piece seems really hard for AI today. And so let me ask you first, Seb, because at the time of this taping, of course, there was a new spin of GPT-4o last week that became overly sycophantic. In other words, it was actually prone in that test of mine not only to not tell me I’m wrong, but it actually praised me for the creativity of my differential. [LAUGHTER] What’s up with that?  BUBECK: Yeah, I guess it’s a testament to the fact that training those models is still more of an art than a science. So it’s a difficult job. Just to be clear with the audience, we have rolled back that [LAUGHS] version of GPT-4o, so now we don’t have the sycophant version out there.  Yeah, no, it’s a really difficult question. It has to do … as you said, it’s very technical. It has to do with the post-training and how, like, where do you nudge the model? So, you know, there is this very classical by now technique called RLHF [reinforcement learning from human feedback], where you push the model in the direction of a certain reward model. So the reward model is just telling the model, you know, what behavior is good, what behavior is bad.  But this reward model is itself an LLM, and, you know, Bill was saying at the very beginning of the conversation that we don’t really understand how those LLMs deal with concepts like, you know, where is the capital of France located? Things like that. It is the same thing for this reward model. We don’t know why it says that it prefers one output to another, and whether this is correlated with some sycophancy is, you know, something that we discovered basically just now. That if you push too hard in optimization on this reward model, you will get a sycophant model.  So it’s kind of … what I’m trying to say is we became too good at what we were doing, and we ended up, in fact, in a trap of the reward model.  LEE: I mean, you do want … it’s a difficult balance because you do want models to follow your desires and …  BUBECK: It’s a very difficult, very difficult balance.  LEE: So this brings up then the following question for me, which is the extent to which we think we’ll need to have specially trained models for things. So let me start with you, Bill. Do you have a point of view on whether we will need to, you know, quote-unquote take AI models to med school? Have them specially trained? Like, if you were going to deploy something to give medical care in underserved parts of the world, do we need to do something special to create those models?  GATES: We certainly need to teach them the African languages and the unique dialects so that the multimedia interactions are very high quality. We certainly need to teach them the disease prevalence and unique disease patterns like, you know, neglected tropical diseases and malaria. So we need to gather a set of facts that somebody trying to go for a US customer base, you know, wouldn’t necessarily have that in there.  Those two things are actually very straightforward because the additional training time is small. I’d say for the next few years, we’ll also need to do reinforcement learning about the context of being a doctor and how important certain behaviors are. Humans learn over the course of their life to some degree that, I’m in a different context and the way I behave in terms of being willing to criticize or be nice, you know, how important is it? Who’s here? What’s my relationship to them?   Right now, these machines don’t have that broad social experience. And so if you know it’s going to be used for health things, a lot of reinforcement learning of the very best humans in that context would still be valuable. Eventually, the models will, having read all the literature of the world about good doctors, bad doctors, it’ll understand as soon as you say, “I want you to be a doctor diagnosing somebody.” All of the implicit reinforcement that fits that situation, you know, will be there. LEE: Yeah. GATES: And so I hope three years from now, we don’t have to do that reinforcement learning. But today, for any medical context, you would want a lot of data to reinforce tone, willingness to say things when, you know, there might be something significant at stake.  LEE: Yeah. So, you know, something Bill said, kind of, reminds me of another thing that I think we missed, which is, the context also … and the specialization also pertains to different, I guess, what we still call “modes,” although I don’t know if the idea of multimodal is the same as it was two years ago. But, you know, what do you make of all of the hubbub around—in fact, within Microsoft Research, this is a big deal, but I think we’re far from alone—you know, medical images and vision, video, proteins and molecules, cell, you know, cellular data and so on.  BUBECK: Yeah. OK. So there is a lot to say to everything … to the last, you know, couple of minutes. Maybe on the specialization aspect, you know, I think there is, hiding behind this, a really fundamental scientific question of whether eventually we have a singular AGI [artificial general intelligence] that kind of knows everything and you can just put, you know, explain your own context and it will just get it and understand everything.  That’s one vision. I have to say, I don’t particularly believe in this vision. In fact, we humans are not like that at all. I think, hopefully, we are general intelligences, yet we have to specialize a lot. And, you know, I did myself a lot of RL, reinforcement learning, on mathematics. Like, that’s what I did, you know, spent a lot of time doing that. And I didn’t improve on other aspects. You know, in fact, I probably degraded in other aspects. [LAUGHTER] So it’s … I think it’s an important example to have in mind.  LEE: I think I might disagree with you on that, though, because, like, doesn’t a model have to see both good science and bad science in order to be able to gain the ability to discern between the two?  BUBECK: Yeah, no, that absolutely. I think there is value in seeing the generality, in having a very broad base. But then you, kind of, specialize on verticals. And this is where also, you know, open-weights model, which we haven’t talked about yet, are really important because they allow you to provide this broad base to everyone. And then you can specialize on top of it.  LEE: So we have about three hours of stuff to talk about, but our time is actually running low. BUBECK: Yes, yes, yes.   LEE: So I think I want … there’s a more provocative question. It’s almost a silly question, but I need to ask it of the two of you, which is, is there a future, you know, where AI replaces doctors or replaces, you know, medical specialties that we have today? So what does the world look like, say, five years from now?  GATES: Well, it’s important to distinguish healthcare discovery activity from healthcare delivery activity. We focused mostly on delivery. I think it’s very much within the realm of possibility that the AI is not only accelerating healthcare discovery but substituting for a lot of the roles of, you know, I’m an organic chemist, or I run various types of assays. I can see those, which are, you know, testable-output-type jobs but with still very high value, I can see, you know, some replacement in those areas before the doctor.   The doctor, still understanding the human condition and long-term dialogues, you know, they’ve had a lifetime of reinforcement of that, particularly when you get into areas like mental health. So I wouldn’t say in five years, either people will choose to adopt it, but it will be profound that there’ll be this nearly free intelligence that can do follow-up, that can help you, you know, make sure you went through different possibilities.  And so I’d say, yes, we’ll have doctors, but I’d say healthcare will be massively transformed in its quality and in efficiency by AI in that time period.  LEE: Is there a comparison, useful comparison, say, between doctors and, say, programmers, computer programmers, or doctors and, I don’t know, lawyers?  GATES: Programming is another one that has, kind of, a mathematical correctness to it, you know, and so the objective function that you’re trying to reinforce to, as soon as you can understand the state machines, you can have something that’s “checkable”; that’s correct. So I think programming, you know, which is weird to say, that the machine will beat us at most programming tasks before we let it take over roles that have deep empathy, you know, physical presence and social understanding in them.  LEE: Yeah. By the way, you know, I fully expect in five years that AI will produce mathematical proofs that are checkable for validity, easily checkable, because they’ll be written in a proof-checking language like Lean or something but will be so complex that no human mathematician can understand them. I expect that to happen.   I can imagine in some fields, like cellular biology, we could have the same situation in the future because the molecular pathways, the chemistry, biochemistry of human cells or living cells is as complex as any mathematics, and so it seems possible that we may be in a state where in wet lab, we see, Oh yeah, this actually works, but no one can understand why.  BUBECK: Yeah, absolutely. I mean, I think I really agree with Bill’s distinction of the discovery and the delivery, and indeed, the discovery’s when you can check things, and at the end, there is an artifact that you can verify. You know, you can run the protocol in the wet lab and see [if you have] produced what you wanted. So I absolutely agree with that.   And in fact, you know, we don’t have to talk five years from now. I don’t know if you know, but just recently, there was a paper that was published on a scientific discovery using o3- mini (opens in new tab). So this is really amazing. And, you know, just very quickly, just so people know, it was about this statistical physics model, the frustrated Potts model, which has to do with coloring, and basically, the case of three colors, like, more than two colors was open for a long time, and o3 was able to reduce the case of three colors to two colors.   LEE: Yeah.  BUBECK: Which is just, like, astounding. And this is not … this is now. This is happening right now. So this is something that I personally didn’t expect it would happen so quickly, and it’s due to those reasoning models.   Now, on the delivery side, I would add something more to it for the reason why doctors and, in fact, lawyers and coders will remain for a long time, and it’s because we still don’t understand how those models generalize. Like, at the end of the day, we are not able to tell you when they are confronted with a really new, novel situation, whether they will work or not.  Nobody is able to give you that guarantee. And I think until we understand this generalization better, we’re not going to be willing to just let the system in the wild without human supervision.  LEE: But don’t human doctors, human specialists … so, for example, a cardiologist sees a patient in a certain way that a nephrologist …  BUBECK: Yeah. LEE: … or an endocrinologist might not. BUBECK: That’s right. But another cardiologist will understand and, kind of, expect a certain level of generalization from their peer. And this, we just don’t have it with AI models. Now, of course, you’re exactly right. That generalization is also hard for humans. Like, if you have a human trained for one task and you put them into another task, then you don’t … you often don’t know. LEE: OK. You know, the podcast is focused on what’s happened over the last two years. But now, I’d like one provocative prediction about what you think the world of AI and medicine is going to be at some point in the future. You pick your timeframe. I don’t care if it’s two years or 20 years from now, but, you know, what do you think will be different about AI in medicine in that future than today?  BUBECK: Yeah, I think the deployment is going to accelerate soon. Like, we’re really not missing very much. There is this enormous capability overhang. Like, even if progress completely stopped, with current systems, we can do a lot more than what we’re doing right now. So I think this will … this has to be realized, you know, sooner rather than later.  And I think it’s probably dependent on these benchmarks and proper evaluation and tying this with regulation. So these are things that take time in human society and for good reason. But now we already are at two years; you know, give it another two years and it should be really …   LEE: Will AI prescribe your medicines? Write your prescriptions?  BUBECK: I think yes. I think yes.  LEE: OK. Bill?  GATES: Well, I think the next two years, we’ll have massive pilots, and so the amount of use of the AI, still in a copilot-type mode, you know, we should get millions of patient visits, you know, both in general medicine and in the mental health side, as well. And I think that’s going to build up both the data and the confidence to give the AI some additional autonomy. You know, are you going to let it talk to you at night when you’re panicked about your mental health with some ability to escalate? And, you know, I’ve gone so far as to tell politicians with national health systems that if they deploy AI appropriately, that the quality of care, the overload of the doctors, the improvement in the economics will be enough that their voters will be stunned because they just don’t expect this, and, you know, they could be reelected [LAUGHTER] just on this one thing of fixing what is a very overloaded and economically challenged health system in these rich countries.  You know, my personal role is going to be to make sure that in the poorer countries, there isn’t some lag; in fact, in many cases, that we’ll be more aggressive because, you know, we’re comparing to having no access to doctors at all. And, you know, so I think whether it’s India or Africa, there’ll be lessons that are globally valuable because we need medical intelligence. And, you know, thank god AI is going to provide a lot of that.  LEE: Well, on that optimistic note, I think that’s a good way to end. Bill, Seb, really appreciate all of this.   I think the most fundamental prediction we made in the book is that AI would actually find its way into the practice of medicine, and I think that that at least has come true, maybe in different ways than we expected, but it’s come true, and I think it’ll only accelerate from here. So thanks again, both of you.  [TRANSITION MUSIC]  GATES: Yeah. Thanks, you guys.  BUBECK: Thank you, Peter. Thanks, Bill.  LEE: I just always feel such a sense of privilege to have a chance to interact and actually work with people like Bill and Sébastien.    With Bill, I’m always amazed at how practically minded he is. He’s really thinking about the nuts and bolts of what AI might be able to do for people, and his thoughts about underserved parts of the world, the idea that we might actually be able to empower people with access to expert medical knowledge, I think is both inspiring and amazing.   And then, Seb, Sébastien Bubeck, he’s just absolutely a brilliant mind. He has a really firm grip on the deep mathematics of artificial intelligence and brings that to bear in his research and development work. And where that mathematics takes him isn’t just into the nuts and bolts of algorithms but into philosophical questions about the nature of intelligence.   One of the things that Sébastien brought up was the state of evaluation of AI systems. And indeed, he was fairly critical in our conversation. But of course, the world of AI research and development is just moving so fast, and indeed, since we recorded our conversation, OpenAI, in fact, released a new evaluation metric that is directly relevant to medical applications, and that is something called HealthBench. And Microsoft Research also released a new evaluation approach or process called ADeLe.   HealthBench and ADeLe are examples of new approaches to evaluating AI models that are less about testing their knowledge and ability to pass multiple-choice exams and instead are evaluation approaches designed to assess how well AI models are able to complete tasks that actually arise every day in typical healthcare or biomedical research settings. These are examples of really important good work that speak to how well AI models work in the real world of healthcare and biomedical research and how well they can collaborate with human beings in those settings.  You know, I asked Bill and Seb to make some predictions about the future. You know, my own answer, I expect that we’re going to be able to use AI to change how we diagnose patients, change how we decide treatment options.   If you’re a doctor or a nurse and you encounter a patient, you’ll ask questions, do a physical exam, you know, call out for labs just like you do today, but then you’ll be able to engage with AI based on all of that data and just ask, you know, based on all the other people who have gone through the same experience, who have similar data, how were they diagnosed? How were they treated? What were their outcomes? And what does that mean for the patient I have right now? Some people call it the “patients like me” paradigm. And I think that’s going to become real because of AI within our lifetimes. That idea of really grounding the delivery in healthcare and medical practice through data and intelligence, I actually now don’t see any barriers to that future becoming real.  [THEME MUSIC]  I’d like to extend another big thank you to Bill and Sébastien for their time. And to our listeners, as always, it’s a pleasure to have you along for the ride. I hope you’ll join us for our remaining conversations, as well as a second coauthor roundtable with Carey and Zak.   Until next time.   [MUSIC FADES]

Ethan Hunt and the IMF team race against time to find a rogue artificial intelligence (why is AI always the bad guy now if films? ) that can destroy mankind. AI, IMF & VFX: A Mission Worth Rendering In the latest episode of The VFXShow podcast, hosts Matt Wallin, Jason Diamond, and Mike Seymour reunite to dissect the spectacle, story, and seamless visual effects of Mission: Impossible – The Final Reckoning. As the eighth entry in the franchise, this chapter serves as a high-stakes, high-altitude crescendo to Tom Cruise’s nearly 30-year run as Ethan Hunt,  the relentless agent of the Impossible Mission Force. Cruise Control: When Practical Meets Pixel While the narrative revolves around the existential threat of a rogue AI known as The Entity, the real heart of the film lies in its bold commitment to visceral, real-world action. The VFX team discusses how Cruise’s ongoing devotion to doing his own death-defying stunts, from leaping between bi-planes to diving into the wreckage of a sunken submarine,  paradoxically increases the importance of invisible VFX. From seamless digital stitching to background replacements and subtle physics enhancements, the effects work had to serve the story without ever betraying the sense of raw, in-camera danger. Matt, Jason, and Mike explore how VFX in this film plays a critical supporting role, cleaning up stunts, compositing dangerous sequences, and selling the illusion of globe-spanning chaos. Whether it’s simulating the collapse of a Cold War-era submarine, managing intricate water dynamics in Ethan’s deep-sea dive, or integrating AI-driven visualisations of nuclear catastrophe, the film leans heavily on sophisticated post work to make Cruise’s practical stunts feel even more grounded and believable. The team also reflects on the thematic evolution of the franchise. While the plot may twist through layers of espionage, betrayal, and digital apocalypse, including face-offs with Gabriel, doomsday cults, and geopolitical brinkmanship,  it is not the team’s favourite MI film. And yet, they note, even as the story veers into sci-fi territory with sentient algorithms and bunker-bound AI traps, the VFX never overshadows the tactile performance at the film’s centre. Falling, Flying, Faking It Beautifully For fans of the franchise, visual effects, or just adrenaline-fueled cinema, this episode offers a thoughtful cinematic critique on how modern VFX artistry and old-school stuntwork can coexist to save a film that has lost its driving narrative direction. This week in our lineup is (or are they really??) Matt Wallin *            @mattwallin    www.mattwallin.com Follow Matt on Mastodon: @[email protected] Jason Diamond  @jasondiamond           www.thediamondbros.com Mike Seymour   @mikeseymour             www.fxguide.com. + @mikeseymour Special thanks to Matt Wallin for the editing & production of the show with help from Jim Shen.

In a nutshell: As nations set ever more ambitious targets for renewable energy and electrification, the humble high-voltage cable has emerged as a linchpin – and a potential chokepoint – in the race to decarbonize the global economy. A Bloomberg interview with Claes Westerlind, CEO of NKT, a leading cable manufacturer based in Denmark, explains why. A global surge in demand for high-voltage electricity cables is threatening to stall the clean energy revolution, as the world's ability to build new wind farms, solar plants, and cross-border power links increasingly hinges on a supply chain bottleneck few outside the industry have considered. At the center of this challenge is the complex, capital-intensive process of manufacturing the giant cables that transport electricity across hundreds of miles, both over land and under the sea. Despite soaring demand, cable manufacturers remain cautious about expanding capacity, raising questions about whether the pace of electrification can keep up with climate ambitions, geopolitical tensions, and the practical realities of industrial investment. High-voltage cables are the arteries of modern power grids, carrying electrons from remote wind farms or hydroelectric dams to the cities and industries that need them. Unlike the thin wires that run through a home's walls, these cables are engineering marvels – sometimes as thick as a person's torso, armored to withstand the crushing pressure of the ocean floor, and designed to last for decades under extreme electrical and environmental stress. "If you look at the very high voltage direct current cable, able to carry roughly two gigawatts through two pairs of cables – that means that the equivalent of one nuclear power reactor is flowing through one cable," Westerlind told Bloomberg. The process of making these cables is as specialized as it is demanding. At the core is a conductor, typically made of copper or aluminum, twisted together like a rope for flexibility and strength. Around this, manufacturers apply multiple layers of insulation in towering vertical factories to ensure the cable remains perfectly round and can safely contain the immense voltages involved. Any impurity in the insulation, even something as small as an eyelash, can cause catastrophic failure, potentially knocking out power to entire cities. // Related Stories As the world rushes to harness new sources of renewable energy, the demand for high-voltage direct current (HVDC) cables has skyrocketed. HVDC technology, initially pioneered by NKT in the 1950s, has become the backbone of long-distance power transmission, particularly for offshore wind farms and intercontinental links. In recent years, approximately 80 to 90 percent of new large-scale cable projects have utilized HVDC, reflecting its efficiency in transmitting electricity over vast distances with minimal losses. But this surge in demand has led to a critical bottleneck. Factories that produce these cables are booked out for years, Westerlind reports, and every project requires custom engineering to match the power needs, geography, and environmental conditions of its route. According to the International Energy Agency, meeting global clean energy goals will require building the equivalent of 80 million kilometers (around 49.7 million miles) of new grid infrastructure by 2040 – essentially doubling what has been constructed over the past century, but in just 15 years. Despite the clear need, cable makers have been slow to add capacity due to reasons that are as much economic and political as technical. Building a new cable factory can cost upwards of a billion euros, and manufacturers are wary of making such investments without long-term commitments from utilities or governments. "For a company like us to do investments in the realm of €1 or 2 billion, it's a massive commitment... but it's also a massive amount of demand that is needed for this investment to actually make financial sense over the next not five years, not 10 years, but over the next 20 to 30 years," Westerlind said. The industry still bears scars from a decade ago, when anticipated demand failed to materialize and expensive new facilities sat underused. Some governments and transmission system operators are trying to break the logjam by making "anticipatory investments" – committing to buy cable capacity even before specific projects are finalized. This approach, backed by regulators, gives manufacturers the confidence to expand, but it remains the exception rather than the rule. Meanwhile, the industry's structure itself creates barriers to rapid expansion, according to Westerlind. The expertise, technology, and infrastructure required to make high-voltage cables are concentrated in a handful of companies, creating what analysts describe as a "deep moat" that is difficult for new entrants to cross. Geopolitical tensions add another layer of complexity. China has built more HVDC lines than any other country, although Western manufacturers, such as NKT, maintain a technical edge in the most advanced cable systems. Still, there is growing concern in Europe and the US about becoming dependent on foreign suppliers for such critical infrastructure, especially in light of recent global conflicts and trade disputes. "Strategic autonomy is very important when it comes to the core parts and the fundamental parts of your society, where the grid backbone is one," Westerlind noted. The stakes are high. Without a rapid and coordinated push to expand cable manufacturing, the world's clean energy transition could be slowed not by a lack of wind or sun but by a shortage of the cables needed to connect them to the grid. As Westerlind put it, "We all know it has to be done... These are large investments. They are very expensive investments. So also the governments have to have a part in enabling these anticipatory investments, and making it possible for the TSOs to actually carry forward with them."

June 13, 20253 min readCould Iran Have Been Close to Making a Nuclear Weapon? Uranium Enrichment ExplainedWhen Israeli aircraft recently struck a uranium-enrichment complex in the nation, Iran could have been days away from achieving “breakout,” the ability to quickly turn “yellowcake” uranium into bomb-grade fuel, with its new high-speed centrifugesBy Deni Ellis Béchard edited by Dean VisserMen work inside of a uranium conversion facility just outside the city of Isfahan, Iran, on March 30, 2005. The facility in Isfahan made hexaflouride gas, which was then enriched by feeding it into centrifuges at a facility in Natanz, Iran. Getty ImagesIn the predawn darkness on Friday local time, Israeli military aircraft struck one of Iran’s uranium-enrichment complexes near the city of Natanz. The warheads aimed to do more than shatter concrete; they were meant to buy time, according to news reports. For months, Iran had seemed to be edging ever closer to “breakout,” the point at which its growing stockpile of partially enriched uranium could be converted into fuel for a nuclear bomb. (Iran has denied that it has been pursuing nuclear weapons development.)But why did the strike occur now? One consideration could involve the way enrichment complexes work. Natural uranium is composed almost entirely of uranium 238, or U-238, an isotope that is relatively “heavy” (meaning it has more neutrons in its nucleus). Only about 0.7 percent is uranium 235 (U-235), a lighter isotope that is capable of sustaining a nuclear chain reaction. That means that in natural uranium, only seven atoms in 1,000 are the lighter, fission-ready U-235; “enrichment” simply means raising the percentage of U-235.U-235 can be used in warheads because its nucleus can easily be split. The International Atomic Energy Agency uses 25 kilograms of contained U-235 as the benchmark amount deemed sufficient for a first-generation implosion bomb. In such a weapon, the U-235 is surrounded by conventional explosives that, when detonated, compress the isotope. A separate device releases a neutron stream. (Neutrons are the neutral subatomic particle in an atom’s nucleus that adds to their mass.) Each time a neutron strikes a U-235 atom, the atom fissions; it divides and spits out, on average, two or three fresh neutrons—plus a burst of energy in the form of heat and gamma radiation. And the emitted neutrons in turn strike other U-235 nuclei, creating a self-sustaining chain reaction among the U-235 atoms that have been packed together into a critical mass. The result is a nuclear explosion. By contrast, the more common isotope, U-238, usually absorbs slow neutrons without splitting and cannot drive such a devastating chain reaction.On supporting science journalismIf you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.To enrich uranium so that it contains enough U-235, the “yellowcake” uranium powder that comes out of a mine must go through a lengthy process of conversions to transform it from a solid into the gas uranium hexafluoride. First, a series of chemical processes refine the uranium and then, at high temperatures, each uranium atom is bound to six fluorine atoms. The result, uranium hexafluoride, is unusual: below 56 degrees Celsius (132.8 degrees Fahrenheit) it is a white, waxy solid, but just above that temperature, it sublimates into a dense, invisible gas.During enrichment, this uranium hexafluoride is loaded into a centrifuge: a metal cylinder that spins at tens of thousands of revolutions per minute—faster than the blades of a jet engine. As the heavier U-238 molecules drift toward the cylinder wall, the lighter U-235 molecules remain closer to the center and are siphoned off. This new, slightly U-235-richer gas is then put into the next centrifuge. The process is repeated 10 to 20 times as ever more enriched gas is sent through a series of centrifuges.Enrichment is a slow process, but the Iranian government has been working on this for years and already holds roughly 400 kilograms of uranium enriched to 60 percent U-235. This falls short of the 90 percent required for nuclear weapons. But whereas Iran’s first-generation IR-1 centrifuges whirl at about 63,000 revolutions per minute and do relatively modest work, its newer IR-6 models, built from high-strength carbon fiber, spin faster and produce enriched uranium far more quickly.Iran has been installing thousands of these units, especially at Fordow, an underground enrichment facility built beneath 80 to 90 meters of rock. According to a report released on Monday by the Institute for Science and International Security, the new centrifuges could produce enough 90 percent U-235 uranium for a warhead “in as little as two to three days” and enough for nine nuclear weapons in three weeks—or 19 by the end of the third month.