🐦 Twitter Post Details

Viewing enriched Twitter post

@pengrui_han

The human brain is strikingly modular, with distinct networks for language, formal reasoning, social reasoning, and physical reasoning. Is this a fundamental principle of how intelligent systems are built, or an accident of biological evolution? In our latest preprint, we find that a similar modular organization emerges in Large Language Models, another class of intelligent system. Brains and LLMs are shaped by entirely different kinds of optimization (biological evolution vs. gradient descent). That they arrive at the same modular design anyway suggests modularity may be a fundamental property of intelligent systems. 🌐 Web: https://t.co/ZKrnTSSuSf 📄 Paper: https://t.co/ZibBXz3PUy 💻 Code & data: https://t.co/uBo5iOYNjy Using circuit analyses across 46 tasks spanning four cognitive domains, we find: 1️⃣ Tasks that draw on the same network in humans recruit overlapping units in LLMs, while tasks drawing on different networks recruit distinct units. 2️⃣ These units are causally linked to model behavior. Ablating the units critical for one domain impairs performance in that domain (−26% accuracy) but barely touches the others (−2.5%). This project has been in the works for a while :) Huge thanks to my advisors @jacobandreas @ev_fedorenko @devarda_a, and to @Nancy_Kanwisher for valuable conceptual input and feedback throughout. #MIT

Media 2
View on Twitter

📊 Media Metadata

{
  "media": [
    {
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2072017872390344904/media_0.mp4",
      "media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2072017872390344904/media_0.mp4",
      "type": "video",
      "filename": "media_0.mp4"
    },
    {
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2072017872390344904/media_1.jpg",
      "media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2072017872390344904/media_1.jpg",
      "type": "photo",
      "filename": "media_1.jpg"
    }
  ],
  "processed_at": "2026-06-30T22:00:57.457165",
  "pipeline_version": "2.0"
}

🔧 Raw API Response

{
  "type": "tweet",
  "id": "2072017872390344904",
  "url": "https://x.com/pengrui_han/status/2072017872390344904",
  "twitterUrl": "https://twitter.com/pengrui_han/status/2072017872390344904",
  "text": "The human brain is strikingly modular, with distinct networks for language, formal reasoning, social reasoning, and physical reasoning. Is this a fundamental principle of how intelligent systems are built, or an accident of biological evolution?\n\nIn our latest preprint, we find that a similar modular organization emerges in Large Language Models, another class of intelligent system.\n\nBrains and LLMs are shaped by entirely different kinds of optimization (biological evolution vs. gradient descent). That they arrive at the same modular design anyway suggests modularity may be a fundamental property of intelligent systems.\n\n🌐 Web: https://t.co/ZKrnTSSuSf\n 📄 Paper: https://t.co/ZibBXz3PUy\n 💻 Code & data: https://t.co/uBo5iOYNjy\n\nUsing circuit analyses across 46 tasks spanning four cognitive domains, we find:\n\n1️⃣ Tasks that draw on the same network in humans recruit overlapping units in LLMs, while tasks drawing on different networks recruit distinct units.\n\n2️⃣ These units are causally linked to model behavior. Ablating the units critical for one domain impairs performance in that domain (−26% accuracy) but barely touches the others (−2.5%).\n\nThis project has been in the works for a while :) Huge thanks to my advisors @jacobandreas @ev_fedorenko @devarda_a, and to @Nancy_Kanwisher for valuable conceptual input and feedback throughout. #MIT",
  "source": "Twitter for iPhone",
  "retweetCount": 61,
  "replyCount": 18,
  "likeCount": 350,
  "quoteCount": 11,
  "viewCount": 22847,
  "createdAt": "Tue Jun 30 18:02:11 +0000 2026",
  "lang": "en",
  "bookmarkCount": 341,
  "isReply": false,
  "inReplyToId": null,
  "conversationId": "2072017872390344904",
  "displayTextRange": [
    0,
    278
  ],
  "inReplyToUserId": null,
  "inReplyToUsername": null,
  "author": {
    "type": "user",
    "userName": "pengrui_han",
    "url": "https://x.com/pengrui_han",
    "twitterUrl": "https://twitter.com/pengrui_han",
    "id": "1732540701521408000",
    "name": "Pengrui Han (Barry)",
    "isVerified": false,
    "isBlueVerified": true,
    "verifiedType": null,
    "profilePicture": "https://pbs.twimg.com/profile_images/1732540796719542272/THnjUob8_normal.jpg",
    "coverPicture": "https://pbs.twimg.com/profile_banners/1732540701521408000/1779828108",
    "description": "",
    "location": "",
    "followers": 268,
    "following": 214,
    "status": "",
    "canDm": false,
    "canMediaTag": true,
    "createdAt": "Wed Dec 06 23:21:29 +0000 2023",
    "entities": {
      "description": {
        "urls": []
      },
      "url": {}
    },
    "fastFollowersCount": 0,
    "favouritesCount": 4227,
    "hasCustomTimelines": true,
    "isTranslator": false,
    "mediaCount": 11,
    "statusesCount": 223,
    "withheldInCountries": [],
    "affiliatesHighlightedLabel": {},
    "possiblySensitive": false,
    "pinnedTweetIds": [
      "2072017872390344904"
    ],
    "profile_bio": {
      "description": "CS @UIUC |  Research @MIT | Prev. @Caltech \nunderstanding AI scientifically",
      "entities": {
        "description": {
          "user_mentions": [
            {
              "id_str": "",
              "indices": [
                3,
                8
              ],
              "name": "",
              "screen_name": "UIUC"
            },
            {
              "id_str": "",
              "indices": [
                21,
                25
              ],
              "name": "",
              "screen_name": "MIT"
            },
            {
              "id_str": "",
              "indices": [
                34,
                42
              ],
              "name": "",
              "screen_name": "Caltech"
            }
          ]
        },
        "url": {
          "urls": [
            {
              "display_url": "pengrui-han.github.io",
              "expanded_url": "https://pengrui-han.github.io/",
              "indices": [
                0,
                23
              ],
              "url": "https://t.co/4VlKcgBNAZ"
            }
          ]
        }
      }
    },
    "isAutomated": false,
    "automatedBy": null
  },
  "extendedEntities": {
    "media": [
      {
        "additional_media_info": {
          "monetizable": false
        },
        "allow_download_status": {
          "allow_download": true
        },
        "display_url": "pic.twitter.com/4CxlxRkpOg",
        "expanded_url": "https://twitter.com/pengrui_han/status/2072017872390344904/video/1",
        "ext_master_playlist_only": [],
        "ext_media_availability": {
          "status": "Available"
        },
        "ext_playlists": [],
        "id_str": "2072011649481936896",
        "indices": [
          279,
          302
        ],
        "media_key": "13_2072011649481936896",
        "media_results": {
          "id": "QXBpTWVkaWFSZXN1bHRzOgwABAoAARzBQ54B15AAAAA=",
          "result": {
            "__typename": "ApiMedia",
            "id": "QXBpTWVkaWE6DAAECgABHMFDngHXkAAAAA==",
            "media_key": "13_2072011649481936896"
          }
        },
        "media_url_https": "https://pbs.twimg.com/amplify_video_thumb/2072011649481936896/img/ChHlrxM0B1ybRbK7.jpg",
        "original_info": {
          "focus_rects": [],
          "height": 2160,
          "width": 2160
        },
        "sizes": {
          "large": {
            "h": 2048,
            "w": 2048
          }
        },
        "type": "video",
        "url": "https://t.co/4CxlxRkpOg",
        "video_info": {
          "aspect_ratio": [
            1,
            1
          ],
          "duration_millis": 28960,
          "variants": [
            {
              "content_type": "application/x-mpegURL",
              "url": "https://video.twimg.com/amplify_video/2072011649481936896/pl/-rAfeRCCSsEByQUC.m3u8?tag=28"
            },
            {
              "bitrate": 432000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/2072011649481936896/vid/avc1/320x320/JqvyU9zoJ7CKlUDw.mp4?tag=28"
            },
            {
              "bitrate": 832000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/2072011649481936896/vid/avc1/540x540/wFA3yaEJ8KXJ-SoD.mp4?tag=28"
            },
            {
              "bitrate": 1280000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/2072011649481936896/vid/avc1/720x720/DMaSDaW0FtvERKVK.mp4?tag=28"
            },
            {
              "bitrate": 8768000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/2072011649481936896/vid/avc1/1080x1080/-3kT_1CnGVnFhFgn.mp4?tag=28"
            },
            {
              "bitrate": 25128000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/2072011649481936896/vid/avc1/2160x2160/duPlLi31OmI4MKgY.mp4?tag=28"
            }
          ]
        }
      }
    ]
  },
  "card": null,
  "place": {},
  "entities": {
    "hashtags": [
      {
        "indices": [
          1354,
          1358
        ],
        "text": "MIT"
      }
    ],
    "symbols": [],
    "timestamps": [],
    "urls": [
      {
        "display_url": "pengrui-han.github.io/LLM_Modularity…",
        "expanded_url": "https://pengrui-han.github.io/LLM_Modularity_Page",
        "indices": [
          636,
          659
        ],
        "url": "https://t.co/ZKrnTSSuSf"
      },
      {
        "display_url": "pengrui-han.github.io/LLM_Modularity…",
        "expanded_url": "https://pengrui-han.github.io/LLM_Modularity_Page/assets/paper.pdf",
        "indices": [
          670,
          693
        ],
        "url": "https://t.co/ZibBXz3PUy"
      },
      {
        "display_url": "github.com/Pengrui-Han/LL…",
        "expanded_url": "https://github.com/Pengrui-Han/LLM_Modularity",
        "indices": [
          710,
          733
        ],
        "url": "https://t.co/uBo5iOYNjy"
      }
    ],
    "user_mentions": [
      {
        "id_str": "2210861",
        "indices": [
          1235,
          1248
        ],
        "name": "Jacob Andreas",
        "screen_name": "jacobandreas"
      },
      {
        "id_str": "1006149495279771648",
        "indices": [
          1249,
          1262
        ],
        "name": "Ev (like in 'evidence', not Eve) Fedorenko 🇺🇦",
        "screen_name": "ev_fedorenko"
      },
      {
        "id_str": "1506315259337973767",
        "indices": [
          1263,
          1273
        ],
        "name": "Andrea de Varda",
        "screen_name": "devarda_a"
      },
      {
        "id_str": "49389484",
        "indices": [
          1282,
          1298
        ],
        "name": "Nancy Kanwisher @NancyKanwisher@mas.to",
        "screen_name": "Nancy_Kanwisher"
      }
    ]
  },
  "quoted_tweet": null,
  "retweeted_tweet": null,
  "isLimitedReply": false,
  "communityInfo": null,
  "article": null
}