A small tool to view real-world ActivityPub objects as JSON! Enter a URL
or username from Mastodon or a similar service below, and we'll send a
request with
the right
Accept
header
to the server to view the underlying object.
{
"@context": [
"https://www.w3.org/ns/activitystreams",
{
"ostatus": "http://ostatus.org#",
"atomUri": "ostatus:atomUri",
"inReplyToAtomUri": "ostatus:inReplyToAtomUri",
"conversation": "ostatus:conversation",
"sensitive": "as:sensitive",
"toot": "http://joinmastodon.org/ns#",
"votersCount": "toot:votersCount",
"litepub": "http://litepub.social/ns#",
"directMessage": "litepub:directMessage",
"blurhash": "toot:blurhash",
"focalPoint": {
"@container": "@list",
"@id": "toot:focalPoint"
},
"Hashtag": "as:Hashtag"
}
],
"id": "https://neuromatch.social/users/fabrice13/statuses/112355833572884870",
"type": "Note",
"summary": null,
"inReplyTo": null,
"published": "2024-04-29T18:06:04Z",
"url": "https://neuromatch.social/@fabrice13/112355833572884870",
"attributedTo": "https://neuromatch.social/users/fabrice13",
"to": [
"https://www.w3.org/ns/activitystreams#Public"
],
"cc": [
"https://neuromatch.social/users/fabrice13/followers"
],
"sensitive": false,
"atomUri": "https://neuromatch.social/users/fabrice13/statuses/112355833572884870",
"inReplyToAtomUri": null,
"conversation": "tag:neuromatch.social,2024-04-29:objectId=12580489:objectType=Conversation",
"content": "<p>I don't understand why this paper ends like this.<br />We can offer explanations or at least hypotheses as to why GLU variants improve Transformers: GLU variants allow multiplicative interactions of the 2nd order between (all) tokens. Once you can stack 2nd order interactions you might have virtually any higher order. A stack of MLPs is a stack of additive interactions, it can't generalize and approximate the same classes of functions as well.<br />This paper is from Google, as is the paper highlighting both the importance of Multiplicative Interactions and the fact that GLUs are Multiplicative Interactions.<br />Why are ML papers like this?😮💨</p><p><a href=\"https://neuromatch.social/tags/deeplearning\" class=\"mention hashtag\" rel=\"tag\">#<span>deeplearning</span></a> <a href=\"https://neuromatch.social/tags/transformers\" class=\"mention hashtag\" rel=\"tag\">#<span>transformers</span></a> <a href=\"https://neuromatch.social/tags/inductivebiases\" class=\"mention hashtag\" rel=\"tag\">#<span>inductivebiases</span></a></p>",
"contentMap": {
"it": "<p>I don't understand why this paper ends like this.<br />We can offer explanations or at least hypotheses as to why GLU variants improve Transformers: GLU variants allow multiplicative interactions of the 2nd order between (all) tokens. Once you can stack 2nd order interactions you might have virtually any higher order. A stack of MLPs is a stack of additive interactions, it can't generalize and approximate the same classes of functions as well.<br />This paper is from Google, as is the paper highlighting both the importance of Multiplicative Interactions and the fact that GLUs are Multiplicative Interactions.<br />Why are ML papers like this?😮💨</p><p><a href=\"https://neuromatch.social/tags/deeplearning\" class=\"mention hashtag\" rel=\"tag\">#<span>deeplearning</span></a> <a href=\"https://neuromatch.social/tags/transformers\" class=\"mention hashtag\" rel=\"tag\">#<span>transformers</span></a> <a href=\"https://neuromatch.social/tags/inductivebiases\" class=\"mention hashtag\" rel=\"tag\">#<span>inductivebiases</span></a></p>"
},
"attachment": [
{
"type": "Document",
"mediaType": "image/jpeg",
"url": "https://media.neuromatch.social/media_attachments/files/112/355/802/559/169/851/original/3390712235ebaee1.jpg",
"name": "PDF excerpt from the paper \"GLU Variants improve transformers\" that says in particular \"We offer no explanation as to why these architectures seem to work; we attribute their success, as all else, to divine benevolence\"",
"blurhash": "UDSF;LIot7ayayj@WBay~qj[RjfQxuM{j[f6",
"width": 1079,
"height": 344
}
],
"tag": [
{
"type": "Hashtag",
"href": "https://neuromatch.social/tags/deeplearning",
"name": "#deeplearning"
},
{
"type": "Hashtag",
"href": "https://neuromatch.social/tags/transformers",
"name": "#transformers"
},
{
"type": "Hashtag",
"href": "https://neuromatch.social/tags/inductivebiases",
"name": "#inductivebiases"
}
],
"replies": {
"id": "https://neuromatch.social/users/fabrice13/statuses/112355833572884870/replies",
"type": "Collection",
"first": {
"type": "CollectionPage",
"next": "https://neuromatch.social/users/fabrice13/statuses/112355833572884870/replies?min_id=112355851093497863&page=true",
"partOf": "https://neuromatch.social/users/fabrice13/statuses/112355833572884870/replies",
"items": [
"https://neuromatch.social/users/fabrice13/statuses/112355851093497863"
]
}
},
"likes": {
"id": "https://neuromatch.social/users/fabrice13/statuses/112355833572884870/likes",
"type": "Collection",
"totalItems": 1
},
"shares": {
"id": "https://neuromatch.social/users/fabrice13/statuses/112355833572884870/shares",
"type": "Collection",
"totalItems": 1
}
}