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"
}
],
"id": "https://rukii.net/users/tero/statuses/113172163313962708",
"type": "Note",
"summary": null,
"inReplyTo": "https://kolektiva.social/users/sleepfreeparent/statuses/113170726836983172",
"published": "2024-09-20T22:09:27Z",
"url": "https://rukii.net/@tero/113172163313962708",
"attributedTo": "https://rukii.net/users/tero",
"to": [
"https://www.w3.org/ns/activitystreams#Public"
],
"cc": [
"https://rukii.net/users/tero/followers",
"https://kolektiva.social/users/sleepfreeparent"
],
"sensitive": false,
"atomUri": "https://rukii.net/users/tero/statuses/113172163313962708",
"inReplyToAtomUri": "https://kolektiva.social/users/sleepfreeparent/statuses/113170726836983172",
"conversation": "tag:rukii.net,2024-09-20:objectId=317525645:objectType=Conversation",
"content": "<p><span class=\"h-card\" translate=\"no\"><a href=\"https://kolektiva.social/@sleepfreeparent\" class=\"u-url mention\">@<span>sleepfreeparent</span></a></span>, only supervised learning is about ground truth. We don't actually need ground truth for other learning rules such as reinforcement learning where we only need to define the rewards, or contrastive learning where we only need to know which task performance was better.</p><p>Luckily LLMs are great at evaluating which of their own alternative performances was better across arbitrary dimensions of quality, and even quality dimensions they themselves suggest for the situation at hand.</p><p>And they can also evaluate and rank their own performances of evaluation tasks.</p><p>All this allows us to use LLMs to refine their own training data to make them self-improve, and use the improved models to improve themselves even better, thus implementing recursive self-improvement which isn't bounded by human level or imitative objectives.</p><p>You are however raising a crucial point that the LLMs pondering in a dark room have limits on how much new knowledge they can derive from only the knowledge they already know. They will need raw data to adopt, refine and produce new knowledge. But this raw data doesn't need to be of a high quality, as we can automatically refine it. We can in effect extract all valuable information, knowledge and skills out of raw data which is of low quality, and apply compute to further derive all implications of everything that is known.</p><p>These implications also include learning new skills by applying known skills together. Not in sequence, but applying for example the skill of judging to the skill of judging, which allows the models to learn to judge and therefore improve their judgement skills.</p>",
"contentMap": {
"en": "<p><span class=\"h-card\" translate=\"no\"><a href=\"https://kolektiva.social/@sleepfreeparent\" class=\"u-url mention\">@<span>sleepfreeparent</span></a></span>, only supervised learning is about ground truth. We don't actually need ground truth for other learning rules such as reinforcement learning where we only need to define the rewards, or contrastive learning where we only need to know which task performance was better.</p><p>Luckily LLMs are great at evaluating which of their own alternative performances was better across arbitrary dimensions of quality, and even quality dimensions they themselves suggest for the situation at hand.</p><p>And they can also evaluate and rank their own performances of evaluation tasks.</p><p>All this allows us to use LLMs to refine their own training data to make them self-improve, and use the improved models to improve themselves even better, thus implementing recursive self-improvement which isn't bounded by human level or imitative objectives.</p><p>You are however raising a crucial point that the LLMs pondering in a dark room have limits on how much new knowledge they can derive from only the knowledge they already know. They will need raw data to adopt, refine and produce new knowledge. But this raw data doesn't need to be of a high quality, as we can automatically refine it. We can in effect extract all valuable information, knowledge and skills out of raw data which is of low quality, and apply compute to further derive all implications of everything that is known.</p><p>These implications also include learning new skills by applying known skills together. Not in sequence, but applying for example the skill of judging to the skill of judging, which allows the models to learn to judge and therefore improve their judgement skills.</p>"
},
"updated": "2024-09-20T22:14:26Z",
"attachment": [],
"tag": [
{
"type": "Mention",
"href": "https://kolektiva.social/users/sleepfreeparent",
"name": "@sleepfreeparent@kolektiva.social"
}
],
"replies": {
"id": "https://rukii.net/users/tero/statuses/113172163313962708/replies",
"type": "Collection",
"first": {
"type": "CollectionPage",
"next": "https://rukii.net/users/tero/statuses/113172163313962708/replies?only_other_accounts=true&page=true",
"partOf": "https://rukii.net/users/tero/statuses/113172163313962708/replies",
"items": []
}
}
}