[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8},"actionData","errors","root",{"_98":99},"routes/item",{"_9":10},"item",{"_11":12,"_13":14,"_15":16,"_17":18,"_19":20,"_21":-5,"_22":23,"_24":25,"_26":27,"_28":29,"_30":27,"_31":32,"_33":34,"_35":36,"_37":38,"_39":40,"_41":-5,"_40":42,"_43":44,"_45":46,"_47":-5,"_48":49,"_50":51,"_52":53,"_54":38,"_55":56,"_57":-5,"_58":38,"_59":60},"id","q918a7enryar5x0yvrrp1chfn","revision",3,"title","NVIDIA 编程智能体 AVO 满分通关 ARC-AGI-3","originalTitle","Holy: NVIDIA’s coding agent AVO scored 100% on ARC-AGI-3’s 25 public games, solving all 183 levels. The agent receives no rules or stated goals. It must learn by trying things, observing the results and correcting its mistakes. AVO succeeds by remembering what it learned and building on it over long periods instead of starting over when the model’s context resets. Powered by Claude Opus 5, the same system previously worked autonomously for seven days optimizing GPU code. Really really cool to see what NVIDIA achieved here","summary","NVIDIA 的编程智能体 AVO 在 ARC-AGI-3 的 25 个公开游戏中全部通关，解决了全部 183 个关卡。该智能体不接收任何规则或目标说明，只能通过不断尝试、观察结果并纠正错误来学习，靠长期记忆积累而非在上下文重置后从头开始。它由 Claude Opus 5 驱动，同一系统此前曾自主运行七天优化 GPU 代码。","reason","source",{"_11":90,"_91":92,"_73":93,"_94":38,"_95":-5,"_96":97},"links",{"_87":88,"_60":89},"publishedAt","2026-08-21T14:54:53.000Z","discoveredAt","2026-10-05T00:42:08.218Z","timelineAt","category","ai-models","tags",[83,84,85,86],"score",45,"selected",false,"channel","x","story",{"_65":46,"_66":67,"_68":-5,"_69":-5,"_70":71},"readingMode","full","author","@kimmonismus","language","body",{"_61":-5,"_60":62,"_63":-5,"_64":56},"outline",[],"relatedStories",[],"indexable","markdownAvailable",true,"group","hasTranslation","bodyLanguage","original","zh","

Holy: NVIDIA’s coding agent AVO scored 100% on ARC-AGI-3’s 25 public games, solving all 183 levels.

The agent receives no rules or stated goals. It must learn by trying things, observing the results and correcting its mistakes.

AVO succeeds by remembering what it learned and building on it over long periods instead of starting over when the model’s context resets.

Really really cool to see what NVIDIA achieved here

","zhKind","complete","authorName","handle","kimmonismus","avatarUrl","quoted","media",[72],{"_73":74,"_75":76,"_77":78,"_79":-5,"_80":-5,"_81":-5,"_82":-5},"kind","image","url","/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHQQRA1KXAAATmv6.jpg&mode=full&exp=1791504000&sig=6d8b8027460bdac8","srcSet","/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHQQRA1KXAAATmv6.jpg&mode=image-720&exp=1791504000&sig=bbd897ef5eb4ec6e 720w, /api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHQQRA1KXAAATmv6.jpg&mode=image-1200&exp=1791504000&sig=d42e6fccf83708f4 1200w, /api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHQQRA1KXAAATmv6.jpg&mode=image-1600&exp=1791504000&sig=ad5d25f0dc31add4 1600w","width","height","alt","poster","评测/基准","编码","推理","Anthropic","aihot","/items/q918a7enryar5x0yvrrp1chfn","https://x.com/kimmonismus/status/2090814903133098211","radar-x-kimmonismus","name","@kimmonismus · X","x_search","firstParty","iconUrl","href","/sources/x/kimmonismus","changelogVersion","2026-10-01T08:00"]