🐦 Twitter Post Details

Viewing enriched Twitter post

@dair_ai

// δ-mem: Efficient Online Memory for LLMs // One of the more elegant memory mechanisms I've seen this month. Most long-term memory work either inflates context or retrains the model. This paper shows a tiny external state, coupled directly into the attention computation, can do the work that bigger context windows fail at. It's cheap, modular, and frozen-model friendly. There is no fine-tuning, no backbone swap, and no context extension. δ-mem augments a frozen full-attention model with a compact online associative-memory state. The state is a fixed-size matrix updated by delta-rule learning, and its readout produces low-rank corrections to the backbone's attention during generation. Results: An 8×8 online memory state is enough to lift the frozen backbone's average score by 1.10x and beat the strongest non-δ-mem memory baseline by 1.15x. On memory-heavy benchmarks the gap widens (1.31x on MemoryAgentBench, 1.20x on LoCoMo) while general capabilities are largely preserved. Paper: https://t.co/peVmZ9reue Learn to build effective AI agents in our academy: https://t.co/LRnpZN7L4c

Media 1
View on Twitter

📊 Media Metadata

{
  "media": [
    {
      "type": "photo",
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2054600147020222630/media_0.jpg",
      "filename": "media_0.jpg"
    }
  ],
  "processed_at": "2026-05-13T17:31:00.869058",
  "pipeline_version": "2.0"
}

🔧 Raw API Response

{
  "type": "tweet",
  "id": "2054600147020222630",
  "url": "https://x.com/dair_ai/status/2054600147020222630",
  "twitterUrl": "https://twitter.com/dair_ai/status/2054600147020222630",
  "text": "// δ-mem: Efficient Online Memory for LLMs //\n\nOne of the more elegant memory mechanisms I've seen this month.\n\nMost long-term memory work either inflates context or retrains the model. This paper shows a tiny external state, coupled directly into the attention computation, can do the work that bigger context windows fail at.\n\nIt's cheap, modular, and frozen-model friendly.\n\nThere is no fine-tuning, no backbone swap, and no context extension.\n\nδ-mem augments a frozen full-attention model with a compact online associative-memory state. The state is a fixed-size matrix updated by delta-rule learning, and its readout produces low-rank corrections to the backbone's attention during generation.\n\nResults:\n\nAn 8×8 online memory state is enough to lift the frozen backbone's average score by 1.10x and beat the strongest non-δ-mem memory baseline by 1.15x. On memory-heavy benchmarks the gap widens (1.31x on MemoryAgentBench, 1.20x on LoCoMo) while general capabilities are largely preserved.\n\nPaper: https://t.co/peVmZ9reue\n\nLearn to build effective AI agents in our academy: https://t.co/LRnpZN7L4c",
  "source": "Twitter for iPhone",
  "retweetCount": 6,
  "replyCount": 0,
  "likeCount": 28,
  "quoteCount": 2,
  "viewCount": 1352,
  "createdAt": "Wed May 13 16:30:22 +0000 2026",
  "lang": "en",
  "bookmarkCount": 28,
  "isReply": false,
  "inReplyToId": null,
  "conversationId": "2054600147020222630",
  "displayTextRange": [
    0,
    278
  ],
  "inReplyToUserId": null,
  "inReplyToUsername": null,
  "author": {
    "type": "user",
    "userName": "dair_ai",
    "url": "https://x.com/dair_ai",
    "twitterUrl": "https://twitter.com/dair_ai",
    "id": "889050642903293953",
    "name": "DAIR.AI",
    "isVerified": false,
    "isBlueVerified": true,
    "verifiedType": null,
    "profilePicture": "https://pbs.twimg.com/profile_images/1643277398522187778/31dedbLo_normal.jpg",
    "coverPicture": "https://pbs.twimg.com/profile_banners/889050642903293953/1773242460",
    "description": "",
    "location": "",
    "followers": 112082,
    "following": 1,
    "status": "",
    "canDm": true,
    "canMediaTag": true,
    "createdAt": "Sun Jul 23 09:12:45 +0000 2017",
    "entities": {
      "description": {
        "urls": []
      },
      "url": {}
    },
    "fastFollowersCount": 0,
    "favouritesCount": 4445,
    "hasCustomTimelines": true,
    "isTranslator": false,
    "mediaCount": 221,
    "statusesCount": 3196,
    "withheldInCountries": [],
    "affiliatesHighlightedLabel": {},
    "possiblySensitive": false,
    "pinnedTweetIds": [
      "2054600147020222630"
    ],
    "profile_bio": {
      "description": "Democratizing AI research, education, and technologies. New AI learning portal: https://t.co/LRnpZN7L4c",
      "entities": {
        "description": {
          "urls": [
            {
              "display_url": "academy.dair.ai",
              "expanded_url": "https://academy.dair.ai/",
              "indices": [
                80,
                103
              ],
              "url": "https://t.co/LRnpZN7L4c"
            }
          ]
        },
        "url": {
          "urls": [
            {
              "display_url": "dair.ai",
              "expanded_url": "https://www.dair.ai/",
              "indices": [
                0,
                23
              ],
              "url": "https://t.co/lkqPZtMU5s"
            }
          ]
        }
      }
    },
    "isAutomated": false,
    "automatedBy": null
  },
  "extendedEntities": {
    "media": [
      {
        "display_url": "pic.twitter.com/2qVQ2t79Yl",
        "expanded_url": "https://twitter.com/dair_ai/status/2054600147020222630/photo/1",
        "ext_media_availability": {
          "status": "Available"
        },
        "features": {
          "large": {
            "faces": [
              {
                "h": 102,
                "w": 102,
                "x": 157,
                "y": 1393
              }
            ]
          },
          "orig": {
            "faces": [
              {
                "h": 102,
                "w": 102,
                "x": 157,
                "y": 1393
              }
            ]
          }
        },
        "id_str": "2054600143224414208",
        "indices": [
          279,
          302
        ],
        "media_key": "3_2054600143224414208",
        "media_results": {
          "id": "QXBpTWVkaWFSZXN1bHRzOgwAAQoAARyDZ/HMW6AACgACHINn8q6bEKYAAA==",
          "result": {
            "__typename": "ApiMedia",
            "id": "QXBpTWVkaWE6DAABCgABHINn8cxboAAKAAIcg2fyrpsQpgAA",
            "media_key": "3_2054600143224414208"
          }
        },
        "media_url_https": "https://pbs.twimg.com/media/HINn8cxboAAJtm3.jpg",
        "original_info": {
          "focus_rects": [
            {
              "h": 907,
              "w": 1620,
              "x": 0,
              "y": 0
            },
            {
              "h": 1620,
              "w": 1620,
              "x": 0,
              "y": 0
            },
            {
              "h": 1834,
              "w": 1609,
              "x": 0,
              "y": 0
            },
            {
              "h": 1834,
              "w": 917,
              "x": 320,
              "y": 0
            },
            {
              "h": 1834,
              "w": 1620,
              "x": 0,
              "y": 0
            }
          ],
          "height": 1834,
          "width": 1620
        },
        "sizes": {
          "large": {
            "h": 1834,
            "w": 1620
          }
        },
        "type": "photo",
        "url": "https://t.co/2qVQ2t79Yl"
      }
    ]
  },
  "card": null,
  "place": {},
  "entities": {
    "hashtags": [],
    "symbols": [],
    "urls": [
      {
        "display_url": "arxiv.org/abs/2605.12357",
        "expanded_url": "https://arxiv.org/abs/2605.12357",
        "indices": [
          1004,
          1027
        ],
        "url": "https://t.co/peVmZ9reue"
      },
      {
        "display_url": "academy.dair.ai",
        "expanded_url": "https://academy.dair.ai/",
        "indices": [
          1080,
          1103
        ],
        "url": "https://t.co/LRnpZN7L4c"
      }
    ],
    "user_mentions": []
  },
  "quoted_tweet": null,
  "retweeted_tweet": null,
  "isLimitedReply": false,
  "communityInfo": null,
  "article": null
}