Brockman said that focusing on the game was his idea, in a project worked on by several people, later explaining that DeepMind was working on something similar with a different game but “had nothing” yet, contributing to their decision not to open-source the technology.

The most notable part of the project, however, is how its development led to understanding that increasing the scale of their compute could rapidly advance the AI capabilities. “…the first Dota bot Jakub Pachocki trained was on 16 CPU cores … every week they had 2x the CPU cores, and the AI got 2x better. There was no limit. We kept increasing the scale, thinking this would peter out, but it never did,” said Brockman.

","freeContentEncoded","textContent","published","strategy","curl-ready","locked","deletedAt","isSyncedFromOwnBlog","original","duplicatesCount","duplicateVersion","originalArticleId","errorsCount","appearance","color","backgroundColor","shortSummary","The project's success in Dota 2 revealed that scaling compute power dramatically improved AI capabilities, with performance doubling alongside compute resources.","summary","Focusing on the game Dota 2 was a deliberate project choice, with DeepMind working on a similar project without immediate results. A key finding was that increasing compute scale rapidly advanced AI capabilities, as AI performance doubled weekly with doubled CPU cores, showing no apparent limit.","bullets",[306,307,308],"The decision to focus on the game Dota 2 was a project idea, with DeepMind pursuing a similar but less advanced project.","Increasing compute scale was found to rapidly advance AI capabilities.","AI performance doubled weekly with doubled CPU cores, indicating no limit to scaling benefits.","persons",[311,312],"Brockman","Jakub Pachocki","places",[],"words",139,"sentiment",2.5,"subject","tech","meta",{"_323":324,"_325":326,"_355":333},"type","bee:get-index-of-articles","schedule",{"_27":327,"_50":-5,"_328":175,"_83":-5,"_329":330,"_331":199,"_332":333,"_334":89,"_86":87,"_319":51,"_335":-5,"_336":-5,"_337":-5,"_338":87,"_287":288,"_339":340,"_54":340,"_341":342,"_343":89,"_344":345,"_56":340,"_346":-5,"_347":87,"_85":124,"_348":349,"_350":-5,"_351":352,"_353":-5,"_354":-5},"420afd2e-c7af-4502-876b-d55f4a27200f","npub","jobId","0aa5206c-740b-4121-8c83-cc05ff79c4d3","order","rssUrl","https://www.theverge.com/ai-artificial-intelligence","system","userRef","xHandle","xUserId","autosign","syncedAt","2025-12-08T13:10:53.526Z","frequency","daily","isOwnBlog","nextRunAt","2026-05-05T21:57:22.588Z","xUserName","autopublish","sourceVariant","rss","systemVariant","scheduleVariant","scraping-bee-classic-proxy","retryArticlesIds","ownershipConfirmedAt","dataSourceUrl","eventId","f01cfbbdf6800a3956824df417f41cf9b478fa930444d8f9c5d0255e916e65e9","nostrRootEvent","event",{"_27":357,"_361":362,"_141":199,"_83":363,"_375":374,"_280":376,"_377":378},"sig","b4a947b830fe41188c1c21d65ef11223ed5bda718b3833c92790b6a1bfa1a5fd8ee52527609b41bd4253838534a9dbc552339aabc66239bde8edb5fab37c3ae9",[364,365,366,370,373],[319,266],[44,273],[367,368,51,369],"a","30023:c3875abd667e3d145a490ef13c953323db26b96b9da091a409902bf5ee629c9c:8cf4fbfb-4341-4f81-adba-24108b9e10ff","mention",[371,372],"client","Layer3.press",[144,374,51],"c3875abd667e3d145a490ef13c953323db26b96b9da091a409902bf5ee629c9c","pubkey","Brockman talks Dota 2.\nLive updates from Elon Musk and Sam Altman’s court battle over the future of OpenAI\n\n\nnostr:naddr1qvzqqqr4gupzpsu8t27kvl3az3dyjrh38j2nxg7my6ukh8dqjxjqnypt7hhx98yuqqjrscmxx3nxyenz956rxdp3956xvwp394skgcnp95ergvfs8p3rjef3xpnxvx2wd7z ","created_at",1778018416,"authorNPub","markdown","Brockman said that [focusing on the game](/2018/8/28/17787610/openai-dota-2-bots-ai-lost-international-reinforcement-learning) was his idea, in a project worked on by several people, later explaining that DeepMind was working on something similar with a different game but “had nothing” yet, contributing to their decision not to open-source the technology.\n\nThe most notable part of the project, however, is how its development led to understanding that increasing the scale of their compute could rapidly advance the AI capabilities. “…the first _Dota_ bot Jakub Pachocki trained was on 16 CPU cores … every week they had 2x the CPU cores, and the AI got 2x better. There was no limit. We kept increasing the scale, thinking this would peter out, but it never did,” said Brockman.","freeMarkdown","kind1EventId","kind30023EventId","d87a6f1f83a5f9784d2604ea0fd9b245be35c37a2c7fcc72dbf6918803042aeb","kind30023",{"_27":385,"_361":388,"_141":389,"_83":390,"_375":374,"_280":404,"_377":405},"0ebc83de190bf9c5d890ccc7c2516a0ae53bcabfd0618fc00b0d29c6dcad3a48c21fd9791561ee27a83de077040c2f5d69130fd107750419f8e7cde725278036",30023,[391,394,395,398,399,401,402,403],[392,393],"r","https://foxvector.com/articles/8cf4fbfb-4341-4f81-adba-24108b9e10ff",[40,266],[396,397],"published_at","1777997283",[44,273],[400,261],"d",[302,267],[371,372],[144,374,51],"Brockman talks Dota 2.\nFocusing on the game Dota 2 was a deliberate project choice, with DeepMind working on a similar project without immediate results. A key finding was that increasing compute scale rapidly advanced AI capabilities, as AI performance doubled weekly with doubled CPU cores, showing no apparent limit.\n - The decision to focus on the game Dota 2 was a project idea, with DeepMind pursuing a similar but less advanced project.\n - Increasing compute scale was found to rapidly advance AI capabilities.\n - AI performance doubled weekly with doubled CPU cores, indicating no limit to scaling benefits.\nContinue reading https://www.theverge.com/ai-artificial-intelligence/924151/brockman-talks-dota-2",1777997283,"publishedToNostr","htmlContentKind30023","

","pubDate","2026-05-05T16:08:03.000Z","modifiedAt","2026-05-05T22:00:16.054Z","2026-05-05T22:01:41.005Z","storyId","likes","zaps","reposts","comments","tweets",[],"relatedArticles",[],"actionData","errors"]