Viewing enriched Twitter post
1/6 We introduce RPG, a principled framework for deriving and analyzing KL-regularized policy gradient methods, unifying GRPO/k3-estimator and REINFORCE++ under this framework and discovering better RL objectives than GRPO: Paper: https://t.co/7xSUj01GIx Code:… https://t.co/0pn5sqhhC7
{
"media": [
{
"type": "image",
"url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/1926839185371017265/media_0.jpg?",
"filename": "media_0.jpg"
}
],
"nlp": {
"sentiment": "neutral",
"topics": [
"RPG framework",
"policy gradient methods",
"reinforcement learning"
],
"entities": [],
"summary": "The tweet introduces a new framework for policy gradient methods in reinforcement learning.",
"language": "en",
"processed_at": "2025-08-07T14:40:46.088532"
}
}{
"data": {
"threaded_conversation_with_injections_v2": {
"instructions": [
{
"type": "TimelineClearCache"
},
{
"type": "TimelineAddEntries",
"entries": [
{
"entryId": "tweet-1926839185371017265",
"sortIndex": "1953556922426720256",
"content": {
"entryType": "TimelineTimelineItem",
"__typename": "TimelineTimelineItem",
"itemContent": {
"itemType": "TimelineTweet",
"__typename": "TimelineTweet",
"tweet_results": {
"result": {
"__typename": "Tweet",
"rest_id": "1926839185371017265",
"has_birdwatch_notes": false,
"core": {
"user_results": {
"result": {
"__typename": "User",
"id": "VXNlcjoxNzc4MjU4NjM0ODg5MzgzOTM3",
"rest_id": "1778258634889383937",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": false,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": false,
"can_media_tag": true,
"created_at": "Thu Apr 11 03:08:00 +0000 2024",
"default_profile": true,
"default_profile_image": false,
"description": "CS Ph.D. student on LLM @ UCLA AGI Lab. Previous works: RPG, MARS, TPA, Kimi-1.5....",
"entities": {
"description": {
"urls": []
},
"url": {
"urls": [
{
"display_url": "lauyikfung.github.io",
"expanded_url": "http://lauyikfung.github.io",
"url": "https://t.co/p0AfEGpLJX",
"indices": [
0,
23
]
}
]
}
},
"fast_followers_count": 0,
"favourites_count": 96,
"followers_count": 170,
"friends_count": 25,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 4,
"location": "Los Angeles",
"media_count": 9,
"name": "YIFENG LIU",
"normal_followers_count": 170,
"pinned_tweet_ids_str": [],
"possibly_sensitive": false,
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1926831203694280705/W4z2Gwxi_normal.jpg",
"profile_interstitial_type": "",
"screen_name": "YIFENGLIU_AI",
"statuses_count": 37,
"translator_type": "none",
"url": "https://t.co/p0AfEGpLJX",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"tipjar_settings": {}
}
}
},
"unmention_data": {},
"edit_control": {
"edit_tweet_ids": [
"1926839185371017265"
],
"editable_until_msecs": "1748232835000",
"is_edit_eligible": false,
"edits_remaining": "5"
},
"is_translatable": false,
"views": {
"count": "50673",
"state": "EnabledWithCount"
},
"source": "<a href=\"https://mobile.twitter.com\" rel=\"nofollow\">Twitter Web App</a>",
"note_tweet": {
"is_expandable": true,
"note_tweet_results": {
"result": {
"id": "Tm90ZVR3ZWV0OjE5MjY4MzkxODUxNDQ0Njc0NTc=",
"text": "1/6 We introduce RPG, a principled framework for deriving and analyzing KL-regularized policy gradient methods, unifying GRPO/k3-estimator and REINFORCE++ under this framework and discovering better RL objectives than GRPO:\nPaper: https://t.co/aGuiLNOmb7\nCode: https://t.co/ox32VIAVMC\nWebpage: https://t.co/bl5lh3KtY7\n@yifan_zhang_, @HuiZhuoY, @QuanquanGu",
"entity_set": {
"hashtags": [],
"symbols": [],
"urls": [
{
"display_url": "arxiv.org/abs/2505.17508",
"expanded_url": "https://arxiv.org/abs/2505.17508",
"url": "https://t.co/aGuiLNOmb7",
"indices": [
231,
254
]
},
{
"display_url": "github.com/complex-reason…",
"expanded_url": "https://github.com/complex-reasoning/RPG",
"url": "https://t.co/ox32VIAVMC",
"indices": [
261,
284
]
},
{
"display_url": "complex-reasoning.github.io/RPG",
"expanded_url": "https://complex-reasoning.github.io/RPG",
"url": "https://t.co/bl5lh3KtY7",
"indices": [
294,
317
]
}
],
"user_mentions": [
{
"id_str": "1585517096120655872",
"name": "Y",
"screen_name": "yifan_zhang_",
"indices": [
318,
331
]
},
{
"id_str": "1658240531858886656",
"name": "Huizhuo Yuan",
"screen_name": "HuiZhuoY",
"indices": [
333,
342
]
},
{
"id_str": "901303999529312256",
"name": "Quanquan Gu",
"screen_name": "QuanquanGu",
"indices": [
344,
355
]
}
]
},
"richtext": {
"richtext_tags": []
},
"media": {
"inline_media": []
}
}
}
},
"legacy": {
"bookmark_count": 163,
"bookmarked": false,
"created_at": "Mon May 26 03:13:55 +0000 2025",
"conversation_id_str": "1926839185371017265",
"display_text_range": [
0,
261
],
"entities": {
"hashtags": [],
"media": [
{
"display_url": "pic.x.com/0pn5sqhhC7",
"expanded_url": "https://x.com/YIFENGLIU_AI/status/1926839185371017265/photo/1",
"id_str": "1926838790770868224",
"indices": [
262,
285
],
"media_key": "3_1926838790770868224",
"media_url_https": "https://pbs.twimg.com/media/Gr2Bq99XQAAM6TL.jpg",
"type": "photo",
"url": "https://t.co/0pn5sqhhC7",
"ext_media_availability": {
"status": "Available"
},
"features": {
"large": {
"faces": []
},
"medium": {
"faces": []
},
"small": {
"faces": []
},
"orig": {
"faces": []
}
},
"sizes": {
"large": {
"h": 1140,
"w": 2048,
"resize": "fit"
},
"medium": {
"h": 668,
"w": 1200,
"resize": "fit"
},
"small": {
"h": 379,
"w": 680,
"resize": "fit"
},
"thumb": {
"h": 150,
"w": 150,
"resize": "crop"
}
},
"original_info": {
"height": 1162,
"width": 2087,
"focus_rects": [
{
"x": 0,
"y": 0,
"w": 2075,
"h": 1162
},
{
"x": 305,
"y": 0,
"w": 1162,
"h": 1162
},
{
"x": 377,
"y": 0,
"w": 1019,
"h": 1162
},
{
"x": 596,
"y": 0,
"w": 581,
"h": 1162
},
{
"x": 0,
"y": 0,
"w": 2087,
"h": 1162
}
]
},
"allow_download_status": {
"allow_download": true
},
"media_results": {
"result": {
"media_key": "3_1926838790770868224"
}
}
}
],
"symbols": [],
"timestamps": [],
"urls": [
{
"display_url": "arxiv.org/abs/2505.17508",
"expanded_url": "https://arxiv.org/abs/2505.17508",
"url": "https://t.co/7xSUj01GIx",
"indices": [
231,
254
]
}
],
"user_mentions": []
},
"extended_entities": {
"media": [
{
"display_url": "pic.x.com/0pn5sqhhC7",
"expanded_url": "https://x.com/YIFENGLIU_AI/status/1926839185371017265/photo/1",
"id_str": "1926838790770868224",
"indices": [
262,
285
],
"media_key": "3_1926838790770868224",
"media_url_https": "https://pbs.twimg.com/media/Gr2Bq99XQAAM6TL.jpg",
"type": "photo",
"url": "https://t.co/0pn5sqhhC7",
"ext_media_availability": {
"status": "Available"
},
"features": {
"large": {
"faces": []
},
"medium": {
"faces": []
},
"small": {
"faces": []
},
"orig": {
"faces": []
}
},
"sizes": {
"large": {
"h": 1140,
"w": 2048,
"resize": "fit"
},
"medium": {
"h": 668,
"w": 1200,
"resize": "fit"
},
"small": {
"h": 379,
"w": 680,
"resize": "fit"
},
"thumb": {
"h": 150,
"w": 150,
"resize": "crop"
}
},
"original_info": {
"height": 1162,
"width": 2087,
"focus_rects": [
{
"x": 0,
"y": 0,
"w": 2075,
"h": 1162
},
{
"x": 305,
"y": 0,
"w": 1162,
"h": 1162
},
{
"x": 377,
"y": 0,
"w": 1019,
"h": 1162
},
{
"x": 596,
"y": 0,
"w": 581,
"h": 1162
},
{
"x": 0,
"y": 0,
"w": 2087,
"h": 1162
}
]
},
"allow_download_status": {
"allow_download": true
},
"media_results": {
"result": {
"media_key": "3_1926838790770868224"
}
}
}
]
},
"favorite_count": 199,
"favorited": false,
"full_text": "1/6 We introduce RPG, a principled framework for deriving and analyzing KL-regularized policy gradient methods, unifying GRPO/k3-estimator and REINFORCE++ under this framework and discovering better RL objectives than GRPO:\nPaper: https://t.co/7xSUj01GIx\nCode:… https://t.co/0pn5sqhhC7",
"is_quote_status": false,
"lang": "en",
"possibly_sensitive": false,
"possibly_sensitive_editable": true,
"quote_count": 3,
"reply_count": 5,
"retweet_count": 38,
"retweeted": false,
"user_id_str": "1778258634889383937",
"id_str": "1926839185371017265"
},
"quick_promote_eligibility": {
"eligibility": "IneligibleNotProfessional"
}
}
},
"tweetDisplayType": "SelfThread"
},
"clientEventInfo": {
"component": "tweet",
"element": "tweet"
}
}
},
{
"entryId": "conversationthread-1926839189095563710",
"sortIndex": "1953556922426720246",
"content": {
"entryType": "TimelineTimelineModule",
"__typename": "TimelineTimelineModule",
"items": [
{
"entryId": "conversationthread-1926839189095563710-tweet-1926839189095563710",
"item": {
"itemContent": {
"itemType": "TimelineTweet",
"__typename": "TimelineTweet",
"tweet_results": {
"result": {
"__typename": "Tweet",
"rest_id": "1926839189095563710",
"has_birdwatch_notes": false,
"core": {
"user_results": {
"result": {
"__typename": "User",
"id": "VXNlcjoxNzc4MjU4NjM0ODg5MzgzOTM3",
"rest_id": "1778258634889383937",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": false,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": false,
"can_media_tag": true,
"created_at": "Thu Apr 11 03:08:00 +0000 2024",
"default_profile": true,
"default_profile_image": false,
"description": "CS Ph.D. student on LLM @ UCLA AGI Lab. Previous works: RPG, MARS, TPA, Kimi-1.5....",
"entities": {
"description": {
"urls": []
},
"url": {
"urls": [
{
"display_url": "lauyikfung.github.io",
"expanded_url": "http://lauyikfung.github.io",
"url": "https://t.co/p0AfEGpLJX",
"indices": [
0,
23
]
}
]
}
},
"fast_followers_count": 0,
"favourites_count": 96,
"followers_count": 170,
"friends_count": 25,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 4,
"location": "Los Angeles",
"media_count": 9,
"name": "YIFENG LIU",
"normal_followers_count": 170,
"pinned_tweet_ids_str": [],
"possibly_sensitive": false,
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1926831203694280705/W4z2Gwxi_normal.jpg",
"profile_interstitial_type": "",
"screen_name": "YIFENGLIU_AI",
"statuses_count": 37,
"translator_type": "none",
"url": "https://t.co/p0AfEGpLJX",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"tipjar_settings": {}
}
}
},
"unmention_data": {},
"edit_control": {
"edit_tweet_ids": [
"1926839189095563710"
],
"editable_until_msecs": "1748232836000",
"is_edit_eligible": false,
"edits_remaining": "5"
},
"is_translatable": false,
"views": {
"count": "990",
"state": "EnabledWithCount"
},
"source": "<a href=\"https://mobile.twitter.com\" rel=\"nofollow\">Twitter Web App</a>",
"legacy": {
"bookmark_count": 2,
"bookmarked": false,
"created_at": "Mon May 26 03:13:56 +0000 2025",
"conversation_id_str": "1926839185371017265",
"display_text_range": [
0,
199
],
"entities": {
"hashtags": [],
"media": [
{
"display_url": "pic.x.com/9O3m2UZBjf",
"expanded_url": "https://x.com/YIFENGLIU_AI/status/1926839189095563710/photo/1",
"id_str": "1926838914595082240",
"indices": [
200,
223
],
"media_key": "3_1926838914595082240",
"media_url_https": "https://pbs.twimg.com/media/Gr2ByLPW0AAG2na.jpg",
"type": "photo",
"url": "https://t.co/9O3m2UZBjf",
"ext_media_availability": {
"status": "Available"
},
"features": {
"large": {
"faces": []
},
"medium": {
"faces": []
},
"small": {
"faces": []
},
"orig": {
"faces": []
}
},
"sizes": {
"large": {
"h": 626,
"w": 2048,
"resize": "fit"
},
"medium": {
"h": 367,
"w": 1200,
"resize": "fit"
},
"small": {
"h": 208,
"w": 680,
"resize": "fit"
},
"thumb": {
"h": 150,
"w": 150,
"resize": "crop"
}
},
"original_info": {
"height": 722,
"width": 2362,
"focus_rects": [
{
"x": 0,
"y": 0,
"w": 1289,
"h": 722
},
{
"x": 0,
"y": 0,
"w": 722,
"h": 722
},
{
"x": 0,
"y": 0,
"w": 633,
"h": 722
},
{
"x": 114,
"y": 0,
"w": 361,
"h": 722
},
{
"x": 0,
"y": 0,
"w": 2362,
"h": 722
}
]
},
"allow_download_status": {
"allow_download": true
},
"media_results": {
"result": {
"media_key": "3_1926838914595082240"
}
}
}
],
"symbols": [],
"timestamps": [],
"urls": [],
"user_mentions": []
},
"extended_entities": {
"media": [
{
"display_url": "pic.x.com/9O3m2UZBjf",
"expanded_url": "https://x.com/YIFENGLIU_AI/status/1926839189095563710/photo/1",
"id_str": "1926838914595082240",
"indices": [
200,
223
],
"media_key": "3_1926838914595082240",
"media_url_https": "https://pbs.twimg.com/media/Gr2ByLPW0AAG2na.jpg",
"type": "photo",
"url": "https://t.co/9O3m2UZBjf",
"ext_media_availability": {
"status": "Available"
},
"features": {
"large": {
"faces": []
},
"medium": {
"faces": []
},
"small": {
"faces": []
},
"orig": {
"faces": []
}
},
"sizes": {
"large": {
"h": 626,
"w": 2048,
"resize": "fit"
},
"medium": {
"h": 367,
"w": 1200,
"resize": "fit"
},
"small": {
"h": 208,
"w": 680,
"resize": "fit"
},
"thumb": {
"h": 150,
"w": 150,
"resize": "crop"
}
},
"original_info": {
"height": 722,
"width": 2362,
"focus_rects": [
{
"x": 0,
"y": 0,
"w": 1289,
"h": 722
},
{
"x": 0,
"y": 0,
"w": 722,
"h": 722
},
{
"x": 0,
"y": 0,
"w": 633,
"h": 722
},
{
"x": 114,
"y": 0,
"w": 361,
"h": 722
},
{
"x": 0,
"y": 0,
"w": 2362,
"h": 722
}
]
},
"allow_download_status": {
"allow_download": true
},
"media_results": {
"result": {
"media_key": "3_1926838914595082240"
}
}
}
]
},
"favorite_count": 8,
"favorited": false,
"full_text": "2/6 ✅We derive policy gradients and corresponding surrogate loss functions for objectives regularized by Forward and Reverse KL divergences, considering both standard normalized and unnormalized KLs. https://t.co/9O3m2UZBjf",
"in_reply_to_screen_name": "YIFENGLIU_AI",
"in_reply_to_status_id_str": "1926839185371017265",
"in_reply_to_user_id_str": "1778258634889383937",
"is_quote_status": false,
"lang": "en",
"possibly_sensitive": false,
"possibly_sensitive_editable": true,
"quote_count": 0,
"reply_count": 1,
"retweet_count": 1,
"retweeted": false,
"user_id_str": "1778258634889383937",
"id_str": "1926839189095563710"
},
"quick_promote_eligibility": {
"eligibility": "IneligibleNotProfessional"
}
}
},
"tweetDisplayType": "SelfThread"
},
"clientEventInfo": {
"component": "tweet",
"element": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
},
{
"entryId": "conversationthread-1926839189095563710-tweet-1926839192648134919",
"item": {
"itemContent": {
"itemType": "TimelineTweet",
"__typename": "TimelineTweet",
"tweet_results": {
"result": {
"__typename": "Tweet",
"rest_id": "1926839192648134919",
"has_birdwatch_notes": false,
"core": {
"user_results": {
"result": {
"__typename": "User",
"id": "VXNlcjoxNzc4MjU4NjM0ODg5MzgzOTM3",
"rest_id": "1778258634889383937",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": false,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": false,
"can_media_tag": true,
"created_at": "Thu Apr 11 03:08:00 +0000 2024",
"default_profile": true,
"default_profile_image": false,
"description": "CS Ph.D. student on LLM @ UCLA AGI Lab. Previous works: RPG, MARS, TPA, Kimi-1.5....",
"entities": {
"description": {
"urls": []
},
"url": {
"urls": [
{
"display_url": "lauyikfung.github.io",
"expanded_url": "http://lauyikfung.github.io",
"url": "https://t.co/p0AfEGpLJX",
"indices": [
0,
23
]
}
]
}
},
"fast_followers_count": 0,
"favourites_count": 96,
"followers_count": 170,
"friends_count": 25,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 4,
"location": "Los Angeles",
"media_count": 9,
"name": "YIFENG LIU",
"normal_followers_count": 170,
"pinned_tweet_ids_str": [],
"possibly_sensitive": false,
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1926831203694280705/W4z2Gwxi_normal.jpg",
"profile_interstitial_type": "",
"screen_name": "YIFENGLIU_AI",
"statuses_count": 37,
"translator_type": "none",
"url": "https://t.co/p0AfEGpLJX",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"tipjar_settings": {}
}
}
},
"unmention_data": {},
"edit_control": {
"edit_tweet_ids": [
"1926839192648134919"
],
"editable_until_msecs": "1748232837000",
"is_edit_eligible": false,
"edits_remaining": "5"
},
"is_translatable": false,
"views": {
"count": "651",
"state": "EnabledWithCount"
},
"source": "<a href=\"https://mobile.twitter.com\" rel=\"nofollow\">Twitter Web App</a>",
"legacy": {
"bookmark_count": 1,
"bookmarked": false,
"created_at": "Mon May 26 03:13:57 +0000 2025",
"conversation_id_str": "1926839185371017265",
"display_text_range": [
0,
243
],
"entities": {
"hashtags": [],
"media": [
{
"display_url": "pic.x.com/K0I40RTmjz",
"expanded_url": "https://x.com/YIFENGLIU_AI/status/1926839192648134919/photo/1",
"id_str": "1926838974544265216",
"indices": [
244,
267
],
"media_key": "3_1926838974544265216",
"media_url_https": "https://pbs.twimg.com/media/Gr2B1qkWwAAwadT.jpg",
"type": "photo",
"url": "https://t.co/K0I40RTmjz",
"ext_media_availability": {
"status": "Available"
},
"features": {
"large": {
"faces": []
},
"medium": {
"faces": []
},
"small": {
"faces": []
},
"orig": {
"faces": []
}
},
"sizes": {
"large": {
"h": 702,
"w": 2048,
"resize": "fit"
},
"medium": {
"h": 411,
"w": 1200,
"resize": "fit"
},
"small": {
"h": 233,
"w": 680,
"resize": "fit"
},
"thumb": {
"h": 150,
"w": 150,
"resize": "crop"
}
},
"original_info": {
"height": 811,
"width": 2367,
"focus_rects": [
{
"x": 919,
"y": 0,
"w": 1448,
"h": 811
},
{
"x": 1305,
"y": 0,
"w": 811,
"h": 811
},
{
"x": 1355,
"y": 0,
"w": 711,
"h": 811
},
{
"x": 1507,
"y": 0,
"w": 406,
"h": 811
},
{
"x": 0,
"y": 0,
"w": 2367,
"h": 811
}
]
},
"allow_download_status": {
"allow_download": true
},
"media_results": {
"result": {
"media_key": "3_1926838974544265216"
}
}
}
],
"symbols": [],
"timestamps": [],
"urls": [],
"user_mentions": []
},
"extended_entities": {
"media": [
{
"display_url": "pic.x.com/K0I40RTmjz",
"expanded_url": "https://x.com/YIFENGLIU_AI/status/1926839192648134919/photo/1",
"id_str": "1926838974544265216",
"indices": [
244,
267
],
"media_key": "3_1926838974544265216",
"media_url_https": "https://pbs.twimg.com/media/Gr2B1qkWwAAwadT.jpg",
"type": "photo",
"url": "https://t.co/K0I40RTmjz",
"ext_media_availability": {
"status": "Available"
},
"features": {
"large": {
"faces": []
},
"medium": {
"faces": []
},
"small": {
"faces": []
},
"orig": {
"faces": []
}
},
"sizes": {
"large": {
"h": 702,
"w": 2048,
"resize": "fit"
},
"medium": {
"h": 411,
"w": 1200,
"resize": "fit"
},
"small": {
"h": 233,
"w": 680,
"resize": "fit"
},
"thumb": {
"h": 150,
"w": 150,
"resize": "crop"
}
},
"original_info": {
"height": 811,
"width": 2367,
"focus_rects": [
{
"x": 919,
"y": 0,
"w": 1448,
"h": 811
},
{
"x": 1305,
"y": 0,
"w": 811,
"h": 811
},
{
"x": 1355,
"y": 0,
"w": 711,
"h": 811
},
{
"x": 1507,
"y": 0,
"w": 406,
"h": 811
},
{
"x": 0,
"y": 0,
"w": 2367,
"h": 811
}
]
},
"allow_download_status": {
"allow_download": true
},
"media_results": {
"result": {
"media_key": "3_1926838974544265216"
}
}
}
]
},
"favorite_count": 9,
"favorited": false,
"full_text": "3/6 ✅We provide derivations for fully differentiable losses and REINFORCE-style gradient estimators. These are developed for the online setting with off-policy gradient estimation.\n\n⭐k3 estimator is equivalent to our unnormalized KL framework. https://t.co/K0I40RTmjz",
"in_reply_to_screen_name": "YIFENGLIU_AI",
"in_reply_to_status_id_str": "1926839189095563710",
"in_reply_to_user_id_str": "1778258634889383937",
"is_quote_status": false,
"lang": "en",
"possibly_sensitive": false,
"possibly_sensitive_editable": true,
"quote_count": 0,
"reply_count": 1,
"retweet_count": 1,
"retweeted": false,
"user_id_str": "1778258634889383937",
"id_str": "1926839192648134919"
},
"quick_promote_eligibility": {
"eligibility": "IneligibleNotProfessional"
}
}
},
"tweetDisplayType": "SelfThread"
},
"clientEventInfo": {
"component": "tweet",
"element": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
},
{
"entryId": "conversationthread-1926839189095563710-tweet-1926839196380987399",
"item": {
"itemContent": {
"itemType": "TimelineTweet",
"__typename": "TimelineTweet",
"tweet_results": {
"result": {
"__typename": "Tweet",
"rest_id": "1926839196380987399",
"has_birdwatch_notes": false,
"core": {
"user_results": {
"result": {
"__typename": "User",
"id": "VXNlcjoxNzc4MjU4NjM0ODg5MzgzOTM3",
"rest_id": "1778258634889383937",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": false,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": false,
"can_media_tag": true,
"created_at": "Thu Apr 11 03:08:00 +0000 2024",
"default_profile": true,
"default_profile_image": false,
"description": "CS Ph.D. student on LLM @ UCLA AGI Lab. Previous works: RPG, MARS, TPA, Kimi-1.5....",
"entities": {
"description": {
"urls": []
},
"url": {
"urls": [
{
"display_url": "lauyikfung.github.io",
"expanded_url": "http://lauyikfung.github.io",
"url": "https://t.co/p0AfEGpLJX",
"indices": [
0,
23
]
}
]
}
},
"fast_followers_count": 0,
"favourites_count": 96,
"followers_count": 170,
"friends_count": 25,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 4,
"location": "Los Angeles",
"media_count": 9,
"name": "YIFENG LIU",
"normal_followers_count": 170,
"pinned_tweet_ids_str": [],
"possibly_sensitive": false,
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1926831203694280705/W4z2Gwxi_normal.jpg",
"profile_interstitial_type": "",
"screen_name": "YIFENGLIU_AI",
"statuses_count": 37,
"translator_type": "none",
"url": "https://t.co/p0AfEGpLJX",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"tipjar_settings": {}
}
}
},
"unmention_data": {},
"edit_control": {
"edit_tweet_ids": [
"1926839196380987399"
],
"editable_until_msecs": "1748232838000",
"is_edit_eligible": false,
"edits_remaining": "5"
},
"is_translatable": false,
"views": {
"count": "531",
"state": "EnabledWithCount"
},
"source": "<a href=\"https://mobile.twitter.com\" rel=\"nofollow\">Twitter Web App</a>",
"legacy": {
"bookmark_count": 1,
"bookmarked": false,
"created_at": "Mon May 26 03:13:58 +0000 2025",
"conversation_id_str": "1926839185371017265",
"display_text_range": [
0,
199
],
"entities": {
"hashtags": [],
"symbols": [],
"timestamps": [],
"urls": [],
"user_mentions": []
},
"favorite_count": 9,
"favorited": false,
"full_text": "4/6 ✅We identify a theoretical inconsistency in the KL estimation for GRPO, and propose a corrected gradient estimator and corresponding loss function that properly incorporates importance weighting.",
"in_reply_to_screen_name": "YIFENGLIU_AI",
"in_reply_to_status_id_str": "1926839192648134919",
"in_reply_to_user_id_str": "1778258634889383937",
"is_quote_status": false,
"lang": "en",
"quote_count": 0,
"reply_count": 1,
"retweet_count": 1,
"retweeted": false,
"user_id_str": "1778258634889383937",
"id_str": "1926839196380987399"
},
"quick_promote_eligibility": {
"eligibility": "IneligibleNotProfessional"
}
}
},
"tweetDisplayType": "SelfThread"
},
"clientEventInfo": {
"component": "tweet",
"element": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
},
{
"entryId": "conversationthread-1926839189095563710-tweet-1926839198813761553",
"item": {
"itemContent": {
"itemType": "TimelineTweet",
"__typename": "TimelineTweet",
"tweet_results": {
"result": {
"__typename": "Tweet",
"rest_id": "1926839198813761553",
"has_birdwatch_notes": false,
"core": {
"user_results": {
"result": {
"__typename": "User",
"id": "VXNlcjoxNzc4MjU4NjM0ODg5MzgzOTM3",
"rest_id": "1778258634889383937",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": false,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": false,
"can_media_tag": true,
"created_at": "Thu Apr 11 03:08:00 +0000 2024",
"default_profile": true,
"default_profile_image": false,
"description": "CS Ph.D. student on LLM @ UCLA AGI Lab. Previous works: RPG, MARS, TPA, Kimi-1.5....",
"entities": {
"description": {
"urls": []
},
"url": {
"urls": [
{
"display_url": "lauyikfung.github.io",
"expanded_url": "http://lauyikfung.github.io",
"url": "https://t.co/p0AfEGpLJX",
"indices": [
0,
23
]
}
]
}
},
"fast_followers_count": 0,
"favourites_count": 96,
"followers_count": 170,
"friends_count": 25,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 4,
"location": "Los Angeles",
"media_count": 9,
"name": "YIFENG LIU",
"normal_followers_count": 170,
"pinned_tweet_ids_str": [],
"possibly_sensitive": false,
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1926831203694280705/W4z2Gwxi_normal.jpg",
"profile_interstitial_type": "",
"screen_name": "YIFENGLIU_AI",
"statuses_count": 37,
"translator_type": "none",
"url": "https://t.co/p0AfEGpLJX",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"tipjar_settings": {}
}
}
},
"unmention_data": {},
"edit_control": {
"edit_tweet_ids": [
"1926839198813761553"
],
"editable_until_msecs": "1748232838000",
"is_edit_eligible": false,
"edits_remaining": "5"
},
"is_translatable": false,
"views": {
"count": "703",
"state": "EnabledWithCount"
},
"source": "<a href=\"https://mobile.twitter.com\" rel=\"nofollow\">Twitter Web App</a>",
"legacy": {
"bookmark_count": 1,
"bookmarked": false,
"created_at": "Mon May 26 03:13:58 +0000 2025",
"conversation_id_str": "1926839185371017265",
"display_text_range": [
0,
149
],
"entities": {
"hashtags": [],
"symbols": [],
"timestamps": [],
"urls": [],
"user_mentions": []
},
"favorite_count": 10,
"favorited": false,
"full_text": "5/6 ✅We also analyze KL regularization in REINFORCE++, revealing its non-standard KL penalty term and its implications for off-policy regularization.",
"in_reply_to_screen_name": "YIFENGLIU_AI",
"in_reply_to_status_id_str": "1926839196380987399",
"in_reply_to_user_id_str": "1778258634889383937",
"is_quote_status": false,
"lang": "en",
"quote_count": 0,
"reply_count": 2,
"retweet_count": 1,
"retweeted": false,
"user_id_str": "1778258634889383937",
"id_str": "1926839198813761553"
},
"quick_promote_eligibility": {
"eligibility": "IneligibleNotProfessional"
}
}
},
"tweetDisplayType": "SelfThread"
},
"clientEventInfo": {
"component": "tweet",
"element": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
},
{
"entryId": "conversationthread-1926839189095563710-tweet-1926839202450129366",
"item": {
"itemContent": {
"itemType": "TimelineTweet",
"__typename": "TimelineTweet",
"tweet_results": {
"result": {
"__typename": "Tweet",
"rest_id": "1926839202450129366",
"has_birdwatch_notes": false,
"core": {
"user_results": {
"result": {
"__typename": "User",
"id": "VXNlcjoxNzc4MjU4NjM0ODg5MzgzOTM3",
"rest_id": "1778258634889383937",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": false,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": false,
"can_media_tag": true,
"created_at": "Thu Apr 11 03:08:00 +0000 2024",
"default_profile": true,
"default_profile_image": false,
"description": "CS Ph.D. student on LLM @ UCLA AGI Lab. Previous works: RPG, MARS, TPA, Kimi-1.5....",
"entities": {
"description": {
"urls": []
},
"url": {
"urls": [
{
"display_url": "lauyikfung.github.io",
"expanded_url": "http://lauyikfung.github.io",
"url": "https://t.co/p0AfEGpLJX",
"indices": [
0,
23
]
}
]
}
},
"fast_followers_count": 0,
"favourites_count": 96,
"followers_count": 170,
"friends_count": 25,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 4,
"location": "Los Angeles",
"media_count": 9,
"name": "YIFENG LIU",
"normal_followers_count": 170,
"pinned_tweet_ids_str": [],
"possibly_sensitive": false,
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1926831203694280705/W4z2Gwxi_normal.jpg",
"profile_interstitial_type": "",
"screen_name": "YIFENGLIU_AI",
"statuses_count": 37,
"translator_type": "none",
"url": "https://t.co/p0AfEGpLJX",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"tipjar_settings": {}
}
}
},
"unmention_data": {},
"edit_control": {
"edit_tweet_ids": [
"1926839202450129366"
],
"editable_until_msecs": "1748232839000",
"is_edit_eligible": false,
"edits_remaining": "5"
},
"is_translatable": false,
"views": {
"count": "2821",
"state": "EnabledWithCount"
},
"source": "<a href=\"https://mobile.twitter.com\" rel=\"nofollow\">Twitter Web App</a>",
"legacy": {
"bookmark_count": 4,
"bookmarked": false,
"created_at": "Mon May 26 03:13:59 +0000 2025",
"conversation_id_str": "1926839185371017265",
"display_text_range": [
0,
260
],
"entities": {
"hashtags": [],
"media": [
{
"display_url": "pic.x.com/OJFSs2PQH5",
"expanded_url": "https://x.com/YIFENGLIU_AI/status/1926839202450129366/photo/1",
"id_str": "1926839106723540992",
"indices": [
261,
284
],
"media_key": "3_1926839106723540992",
"media_url_https": "https://pbs.twimg.com/media/Gr2B9W-WcAAGor8.jpg",
"type": "photo",
"url": "https://t.co/OJFSs2PQH5",
"ext_media_availability": {
"status": "Available"
},
"features": {
"large": {
"faces": []
},
"medium": {
"faces": []
},
"small": {
"faces": []
},
"orig": {
"faces": []
}
},
"sizes": {
"large": {
"h": 1160,
"w": 1906,
"resize": "fit"
},
"medium": {
"h": 730,
"w": 1200,
"resize": "fit"
},
"small": {
"h": 414,
"w": 680,
"resize": "fit"
},
"thumb": {
"h": 150,
"w": 150,
"resize": "crop"
}
},
"original_info": {
"height": 1160,
"width": 1906,
"focus_rects": [
{
"x": 0,
"y": 0,
"w": 1906,
"h": 1067
},
{
"x": 324,
"y": 0,
"w": 1160,
"h": 1160
},
{
"x": 395,
"y": 0,
"w": 1018,
"h": 1160
},
{
"x": 614,
"y": 0,
"w": 580,
"h": 1160
},
{
"x": 0,
"y": 0,
"w": 1906,
"h": 1160
}
]
},
"allow_download_status": {
"allow_download": true
},
"media_results": {
"result": {
"media_key": "3_1926839106723540992"
}
}
},
{
"display_url": "pic.x.com/OJFSs2PQH5",
"expanded_url": "https://x.com/YIFENGLIU_AI/status/1926839202450129366/photo/1",
"id_str": "1926839126726160384",
"indices": [
261,
284
],
"media_key": "3_1926839126726160384",
"media_url_https": "https://pbs.twimg.com/media/Gr2B-hfWMAAqr9i.jpg",
"type": "photo",
"url": "https://t.co/OJFSs2PQH5",
"ext_media_availability": {
"status": "Available"
},
"features": {
"large": {
"faces": []
},
"medium": {
"faces": []
},
"small": {
"faces": []
},
"orig": {
"faces": []
}
},
"sizes": {
"large": {
"h": 1267,
"w": 1847,
"resize": "fit"
},
"medium": {
"h": 823,
"w": 1200,
"resize": "fit"
},
"small": {
"h": 466,
"w": 680,
"resize": "fit"
},
"thumb": {
"h": 150,
"w": 150,
"resize": "crop"
}
},
"original_info": {
"height": 1267,
"width": 1847,
"focus_rects": [
{
"x": 0,
"y": 233,
"w": 1847,
"h": 1034
},
{
"x": 580,
"y": 0,
"w": 1267,
"h": 1267
},
{
"x": 736,
"y": 0,
"w": 1111,
"h": 1267
},
{
"x": 1203,
"y": 0,
"w": 634,
"h": 1267
},
{
"x": 0,
"y": 0,
"w": 1847,
"h": 1267
}
]
},
"allow_download_status": {
"allow_download": true
},
"media_results": {
"result": {
"media_key": "3_1926839126726160384"
}
}
}
],
"symbols": [],
"timestamps": [],
"urls": [],
"user_mentions": []
},
"extended_entities": {
"media": [
{
"display_url": "pic.x.com/OJFSs2PQH5",
"expanded_url": "https://x.com/YIFENGLIU_AI/status/1926839202450129366/photo/1",
"id_str": "1926839106723540992",
"indices": [
261,
284
],
"media_key": "3_1926839106723540992",
"media_url_https": "https://pbs.twimg.com/media/Gr2B9W-WcAAGor8.jpg",
"type": "photo",
"url": "https://t.co/OJFSs2PQH5",
"ext_media_availability": {
"status": "Available"
},
"features": {
"large": {
"faces": []
},
"medium": {
"faces": []
},
"small": {
"faces": []
},
"orig": {
"faces": []
}
},
"sizes": {
"large": {
"h": 1160,
"w": 1906,
"resize": "fit"
},
"medium": {
"h": 730,
"w": 1200,
"resize": "fit"
},
"small": {
"h": 414,
"w": 680,
"resize": "fit"
},
"thumb": {
"h": 150,
"w": 150,
"resize": "crop"
}
},
"original_info": {
"height": 1160,
"width": 1906,
"focus_rects": [
{
"x": 0,
"y": 0,
"w": 1906,
"h": 1067
},
{
"x": 324,
"y": 0,
"w": 1160,
"h": 1160
},
{
"x": 395,
"y": 0,
"w": 1018,
"h": 1160
},
{
"x": 614,
"y": 0,
"w": 580,
"h": 1160
},
{
"x": 0,
"y": 0,
"w": 1906,
"h": 1160
}
]
},
"allow_download_status": {
"allow_download": true
},
"media_results": {
"result": {
"media_key": "3_1926839106723540992"
}
}
},
{
"display_url": "pic.x.com/OJFSs2PQH5",
"expanded_url": "https://x.com/YIFENGLIU_AI/status/1926839202450129366/photo/1",
"id_str": "1926839126726160384",
"indices": [
261,
284
],
"media_key": "3_1926839126726160384",
"media_url_https": "https://pbs.twimg.com/media/Gr2B-hfWMAAqr9i.jpg",
"type": "photo",
"url": "https://t.co/OJFSs2PQH5",
"ext_media_availability": {
"status": "Available"
},
"features": {
"large": {
"faces": []
},
"medium": {
"faces": []
},
"small": {
"faces": []
},
"orig": {
"faces": []
}
},
"sizes": {
"large": {
"h": 1267,
"w": 1847,
"resize": "fit"
},
"medium": {
"h": 823,
"w": 1200,
"resize": "fit"
},
"small": {
"h": 466,
"w": 680,
"resize": "fit"
},
"thumb": {
"h": 150,
"w": 150,
"resize": "crop"
}
},
"original_info": {
"height": 1267,
"width": 1847,
"focus_rects": [
{
"x": 0,
"y": 233,
"w": 1847,
"h": 1034
},
{
"x": 580,
"y": 0,
"w": 1267,
"h": 1267
},
{
"x": 736,
"y": 0,
"w": 1111,
"h": 1267
},
{
"x": 1203,
"y": 0,
"w": 634,
"h": 1267
},
{
"x": 0,
"y": 0,
"w": 1847,
"h": 1267
}
]
},
"allow_download_status": {
"allow_download": true
},
"media_results": {
"result": {
"media_key": "3_1926839126726160384"
}
}
}
]
},
"favorite_count": 11,
"favorited": false,
"full_text": "6/6 🚀Experimental Results: RPG beats GRPO/DAPO/REINFORCE++\n\nWe implement RL training experiments on math datasets with Qwen-2.5-7B-Instruct and Qwen-2.5-Math-7B, achieving more stable and better performances than baselines including GRPO, DAPO and REINFORCE++. https://t.co/OJFSs2PQH5",
"in_reply_to_screen_name": "YIFENGLIU_AI",
"in_reply_to_status_id_str": "1926839198813761553",
"in_reply_to_user_id_str": "1778258634889383937",
"is_quote_status": false,
"lang": "en",
"possibly_sensitive": false,
"possibly_sensitive_editable": true,
"quote_count": 1,
"reply_count": 1,
"retweet_count": 1,
"retweeted": false,
"user_id_str": "1778258634889383937",
"id_str": "1926839202450129366"
},
"quick_promote_eligibility": {
"eligibility": "IneligibleNotProfessional"
}
}
},
"tweetDisplayType": "SelfThread"
},
"clientEventInfo": {
"component": "tweet",
"element": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
}
],
"metadata": {
"conversationMetadata": {
"allTweetIds": [
"1926839196380987399"
],
"enableDeduplication": true
}
},
"displayType": "VerticalConversation",
"clientEventInfo": {
"component": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
},
{
"entryId": "conversationthread-1927209256341668166",
"sortIndex": "1953556922426720236",
"content": {
"entryType": "TimelineTimelineModule",
"__typename": "TimelineTimelineModule",
"items": [
{
"entryId": "conversationthread-1927209256341668166-tweet-1927209256341668166",
"item": {
"itemContent": {
"itemType": "TimelineTweet",
"__typename": "TimelineTweet",
"tweet_results": {
"result": {
"__typename": "Tweet",
"rest_id": "1927209256341668166",
"has_birdwatch_notes": false,
"core": {
"user_results": {
"result": {
"__typename": "User",
"id": "VXNlcjoxNzAxMzE2NTIwMjg1MTg0MDAx",
"rest_id": "1701316520285184001",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": true,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": false,
"can_media_tag": true,
"created_at": "Mon Sep 11 19:27:33 +0000 2023",
"default_profile": true,
"default_profile_image": false,
"description": "ML @CollinearAI | LLM and diffusion hothead",
"entities": {
"description": {
"urls": []
}
},
"fast_followers_count": 0,
"favourites_count": 863,
"followers_count": 124,
"friends_count": 97,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 1,
"location": "",
"media_count": 18,
"name": "Muyu He",
"normal_followers_count": 124,
"pinned_tweet_ids_str": [],
"possibly_sensitive": false,
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1877696567374909440/KRSFANxk_normal.jpg",
"profile_interstitial_type": "",
"screen_name": "HeMuyu0327",
"statuses_count": 75,
"translator_type": "none",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"tipjar_settings": {}
}
}
},
"unmention_data": {},
"edit_control": {
"edit_tweet_ids": [
"1927209256341668166"
],
"editable_until_msecs": "1748321067000",
"is_edit_eligible": false,
"edits_remaining": "5"
},
"is_translatable": false,
"views": {
"count": "126",
"state": "EnabledWithCount"
},
"source": "<a href=\"https://mobile.twitter.com\" rel=\"nofollow\">Twitter Web App</a>",
"legacy": {
"bookmark_count": 0,
"bookmarked": false,
"created_at": "Tue May 27 03:44:27 +0000 2025",
"conversation_id_str": "1926839185371017265",
"display_text_range": [
14,
201
],
"entities": {
"hashtags": [],
"symbols": [],
"timestamps": [],
"urls": [],
"user_mentions": [
{
"id_str": "1778258634889383937",
"name": "YIFENG LIU",
"screen_name": "YIFENGLIU_AI",
"indices": [
0,
13
]
}
]
},
"favorite_count": 1,
"favorited": false,
"full_text": "@YIFENGLIU_AI Fascinating work! Curious if there are possible explanations of why different KL strategies work differently on different AIME subsets and how to pick and choose when it comes to training",
"in_reply_to_screen_name": "YIFENGLIU_AI",
"in_reply_to_status_id_str": "1926839185371017265",
"in_reply_to_user_id_str": "1778258634889383937",
"is_quote_status": false,
"lang": "en",
"quote_count": 0,
"reply_count": 1,
"retweet_count": 0,
"retweeted": false,
"user_id_str": "1701316520285184001",
"id_str": "1927209256341668166"
},
"quick_promote_eligibility": {
"eligibility": "IneligibleNotProfessional"
}
}
},
"tweetDisplayType": "Tweet"
},
"clientEventInfo": {
"component": "tweet",
"element": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
},
{
"entryId": "conversationthread-1927209256341668166-tweet-1927214495731941867",
"item": {
"itemContent": {
"itemType": "TimelineTweet",
"__typename": "TimelineTweet",
"tweet_results": {
"result": {
"__typename": "Tweet",
"rest_id": "1927214495731941867",
"has_birdwatch_notes": false,
"core": {
"user_results": {
"result": {
"__typename": "User",
"id": "VXNlcjoxNzc4MjU4NjM0ODg5MzgzOTM3",
"rest_id": "1778258634889383937",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": false,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": false,
"can_media_tag": true,
"created_at": "Thu Apr 11 03:08:00 +0000 2024",
"default_profile": true,
"default_profile_image": false,
"description": "CS Ph.D. student on LLM @ UCLA AGI Lab. Previous works: RPG, MARS, TPA, Kimi-1.5....",
"entities": {
"description": {
"urls": []
},
"url": {
"urls": [
{
"display_url": "lauyikfung.github.io",
"expanded_url": "http://lauyikfung.github.io",
"url": "https://t.co/p0AfEGpLJX",
"indices": [
0,
23
]
}
]
}
},
"fast_followers_count": 0,
"favourites_count": 96,
"followers_count": 170,
"friends_count": 25,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 4,
"location": "Los Angeles",
"media_count": 9,
"name": "YIFENG LIU",
"normal_followers_count": 170,
"pinned_tweet_ids_str": [],
"possibly_sensitive": false,
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1926831203694280705/W4z2Gwxi_normal.jpg",
"profile_interstitial_type": "",
"screen_name": "YIFENGLIU_AI",
"statuses_count": 37,
"translator_type": "none",
"url": "https://t.co/p0AfEGpLJX",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"tipjar_settings": {}
}
}
},
"unmention_data": {},
"edit_control": {
"edit_tweet_ids": [
"1927214495731941867"
],
"editable_until_msecs": "1748322316000",
"is_edit_eligible": false,
"edits_remaining": "5"
},
"is_translatable": false,
"views": {
"count": "127",
"state": "EnabledWithCount"
},
"source": "<a href=\"https://mobile.twitter.com\" rel=\"nofollow\">Twitter Web App</a>",
"legacy": {
"bookmark_count": 1,
"bookmarked": false,
"created_at": "Tue May 27 04:05:16 +0000 2025",
"conversation_id_str": "1926839185371017265",
"display_text_range": [
12,
292
],
"entities": {
"hashtags": [],
"symbols": [],
"timestamps": [],
"urls": [],
"user_mentions": [
{
"id_str": "1701316520285184001",
"name": "Muyu He",
"screen_name": "HeMuyu0327",
"indices": [
0,
11
]
}
]
},
"favorite_count": 2,
"favorited": false,
"full_text": "@HeMuyu0327 Different KL-terms can have different learning efficiencies. E.g., minimizing forward KL encourages model to cover the support of ref model, while reverse KL encourages model to be concentrated where ref model has high prob. mass. The exp. results show RPG-RKL/UFKL may be better.",
"in_reply_to_screen_name": "HeMuyu0327",
"in_reply_to_status_id_str": "1927209256341668166",
"in_reply_to_user_id_str": "1701316520285184001",
"is_quote_status": false,
"lang": "en",
"quote_count": 0,
"reply_count": 1,
"retweet_count": 0,
"retweeted": false,
"user_id_str": "1778258634889383937",
"id_str": "1927214495731941867"
},
"quick_promote_eligibility": {
"eligibility": "IneligibleNotProfessional"
}
}
},
"tweetDisplayType": "Tweet"
},
"clientEventInfo": {
"component": "tweet",
"element": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
},
{
"entryId": "conversationthread-1927209256341668166-cursor-showmore-912504839",
"item": {
"itemContent": {
"itemType": "TimelineTimelineCursor",
"__typename": "TimelineTimelineCursor",
"value": "DAABCwABAAAAMEVtUEM2d0FBQUFNL2dHSk4wdkdwL0FBQUFBSWF2dEtibkJjQlJocSsxMStBRmdIcggAAgAAAAYKAAMavtKbnBcBRg8ABAoAAAACGr7Sm5wXAUYavtdfgBYB6wAA",
"cursorType": "ShowMore",
"displayTreatment": {
"actionText": "Show replies"
}
},
"clientEventInfo": {
"component": "cursor",
"element": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
}
],
"metadata": {
"conversationMetadata": {
"allTweetIds": [
"1927214495731941867"
],
"enableDeduplication": true
}
},
"displayType": "VerticalConversation",
"clientEventInfo": {
"component": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
},
{
"entryId": "conversationthread-1927328728696144100",
"sortIndex": "1953556922426720226",
"content": {
"entryType": "TimelineTimelineModule",
"__typename": "TimelineTimelineModule",
"items": [
{
"entryId": "conversationthread-1927328728696144100-tweet-1927328728696144100",
"item": {
"itemContent": {
"itemType": "TimelineTweet",
"__typename": "TimelineTweet",
"tweet_results": {
"result": {
"__typename": "Tweet",
"rest_id": "1927328728696144100",
"has_birdwatch_notes": false,
"core": {
"user_results": {
"result": {
"__typename": "User",
"id": "VXNlcjoxNjc0Mzk2OTc0OTk1NjE1NzQ0",
"rest_id": "1674396974995615744",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": false,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": false,
"can_media_tag": true,
"created_at": "Thu Jun 29 12:38:58 +0000 2023",
"default_profile": true,
"default_profile_image": false,
"description": "@GoogleDeepMind research scientist, working on LLM, RL, and more.",
"entities": {
"description": {
"urls": []
},
"url": {
"urls": [
{
"display_url": "waterhorse1.github.io",
"expanded_url": "https://waterhorse1.github.io/",
"url": "https://t.co/j8T4ItigOp",
"indices": [
0,
23
]
}
]
}
},
"fast_followers_count": 0,
"favourites_count": 337,
"followers_count": 966,
"friends_count": 299,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 15,
"location": "",
"media_count": 8,
"name": "Xidong Feng",
"normal_followers_count": 966,
"pinned_tweet_ids_str": [
"1848839891045495291"
],
"possibly_sensitive": false,
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1674397044323258369/CZRPQdD6_normal.png",
"profile_interstitial_type": "",
"screen_name": "Xidong_Feng",
"statuses_count": 174,
"translator_type": "none",
"url": "https://t.co/j8T4ItigOp",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"tipjar_settings": {}
}
}
},
"unmention_data": {},
"edit_control": {
"edit_tweet_ids": [
"1927328728696144100"
],
"editable_until_msecs": "1748349551000",
"is_edit_eligible": false,
"edits_remaining": "5"
},
"is_translatable": false,
"views": {
"count": "223",
"state": "EnabledWithCount"
},
"source": "<a href=\"https://mobile.twitter.com\" rel=\"nofollow\">Twitter Web App</a>",
"legacy": {
"bookmark_count": 1,
"bookmarked": false,
"created_at": "Tue May 27 11:39:11 +0000 2025",
"conversation_id_str": "1926839185371017265",
"display_text_range": [
14,
290
],
"entities": {
"hashtags": [],
"symbols": [],
"timestamps": [],
"urls": [],
"user_mentions": [
{
"id_str": "1778258634889383937",
"name": "YIFENG LIU",
"screen_name": "YIFENGLIU_AI",
"indices": [
0,
13
]
}
]
},
"favorite_count": 3,
"favorited": false,
"full_text": "@YIFENGLIU_AI Nice work! Theorem 3.6 seems to address KL's gradient bias? Can I interpret it as, you need to both take the indirect gradient of E_{\\pi_theta} (handled by policy gradient), and the direct gradient of inner log_\\pi_theta. So a combination of GRPO and REINFORCE KL calculation?",
"in_reply_to_screen_name": "YIFENGLIU_AI",
"in_reply_to_status_id_str": "1926839185371017265",
"in_reply_to_user_id_str": "1778258634889383937",
"is_quote_status": false,
"lang": "en",
"quote_count": 0,
"reply_count": 1,
"retweet_count": 0,
"retweeted": false,
"user_id_str": "1674396974995615744",
"id_str": "1927328728696144100"
},
"quick_promote_eligibility": {
"eligibility": "IneligibleNotProfessional"
}
}
},
"tweetDisplayType": "Tweet"
},
"clientEventInfo": {
"component": "tweet",
"element": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
},
{
"entryId": "conversationthread-1927328728696144100-tweet-1927854485507895533",
"item": {
"itemContent": {
"itemType": "TimelineTweet",
"__typename": "TimelineTweet",
"tweet_results": {
"result": {
"__typename": "Tweet",
"rest_id": "1927854485507895533",
"has_birdwatch_notes": false,
"core": {
"user_results": {
"result": {
"__typename": "User",
"id": "VXNlcjoxNzc4MjU4NjM0ODg5MzgzOTM3",
"rest_id": "1778258634889383937",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": false,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": false,
"can_media_tag": true,
"created_at": "Thu Apr 11 03:08:00 +0000 2024",
"default_profile": true,
"default_profile_image": false,
"description": "CS Ph.D. student on LLM @ UCLA AGI Lab. Previous works: RPG, MARS, TPA, Kimi-1.5....",
"entities": {
"description": {
"urls": []
},
"url": {
"urls": [
{
"display_url": "lauyikfung.github.io",
"expanded_url": "http://lauyikfung.github.io",
"url": "https://t.co/p0AfEGpLJX",
"indices": [
0,
23
]
}
]
}
},
"fast_followers_count": 0,
"favourites_count": 96,
"followers_count": 170,
"friends_count": 25,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 4,
"location": "Los Angeles",
"media_count": 9,
"name": "YIFENG LIU",
"normal_followers_count": 170,
"pinned_tweet_ids_str": [],
"possibly_sensitive": false,
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1926831203694280705/W4z2Gwxi_normal.jpg",
"profile_interstitial_type": "",
"screen_name": "YIFENGLIU_AI",
"statuses_count": 37,
"translator_type": "none",
"url": "https://t.co/p0AfEGpLJX",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"tipjar_settings": {}
}
}
},
"unmention_data": {},
"edit_control": {
"edit_tweet_ids": [
"1927854485507895533"
],
"editable_until_msecs": "1748474901000",
"is_edit_eligible": false,
"edits_remaining": "5"
},
"is_translatable": false,
"views": {
"count": "59",
"state": "EnabledWithCount"
},
"source": "<a href=\"https://mobile.twitter.com\" rel=\"nofollow\">Twitter Web App</a>",
"legacy": {
"bookmark_count": 0,
"bookmarked": false,
"created_at": "Wed May 28 22:28:21 +0000 2025",
"conversation_id_str": "1926839185371017265",
"display_text_range": [
13,
178
],
"entities": {
"hashtags": [],
"symbols": [],
"timestamps": [],
"urls": [],
"user_mentions": [
{
"id_str": "1674396974995615744",
"name": "Xidong Feng",
"screen_name": "Xidong_Feng",
"indices": [
0,
12
]
}
]
},
"favorite_count": 0,
"favorited": false,
"full_text": "@Xidong_Feng Thank you for your appreciation! But I have to point out that GRPO can be classified as URKL instead of RKL, and in Section 4, we combine it with REINFORCE-style KL.",
"in_reply_to_screen_name": "Xidong_Feng",
"in_reply_to_status_id_str": "1927328728696144100",
"in_reply_to_user_id_str": "1674396974995615744",
"is_quote_status": false,
"lang": "en",
"quote_count": 0,
"reply_count": 0,
"retweet_count": 0,
"retweeted": false,
"user_id_str": "1778258634889383937",
"id_str": "1927854485507895533"
},
"quick_promote_eligibility": {
"eligibility": "IneligibleNotProfessional"
}
}
},
"tweetDisplayType": "Tweet"
},
"clientEventInfo": {
"component": "tweet",
"element": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
}
],
"metadata": {
"conversationMetadata": {
"allTweetIds": [
"1927854485507895533"
],
"enableDeduplication": true
}
},
"displayType": "VerticalConversation",
"clientEventInfo": {
"component": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
},
{
"entryId": "conversationthread-1927522250569171093",
"sortIndex": "1953556922426720216",
"content": {
"entryType": "TimelineTimelineModule",
"__typename": "TimelineTimelineModule",
"items": [
{
"entryId": "conversationthread-1927522250569171093-tweet-1927522250569171093",
"item": {
"itemContent": {
"itemType": "TimelineTweet",
"__typename": "TimelineTweet",
"tweet_results": {
"result": {
"__typename": "Tweet",
"rest_id": "1927522250569171093",
"has_birdwatch_notes": false,
"core": {
"user_results": {
"result": {
"__typename": "User",
"id": "VXNlcjo3MDQzODk2MzUwMDc1MDAyOTI=",
"rest_id": "704389635007500292",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": false,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": false,
"can_media_tag": false,
"created_at": "Mon Feb 29 19:35:50 +0000 2016",
"default_profile": true,
"default_profile_image": false,
"description": "Research Associate at Alan Turing Institute",
"entities": {
"description": {
"urls": []
},
"url": {
"urls": [
{
"display_url": "scholar.google.com/citations?user…",
"expanded_url": "https://scholar.google.com/citations?user=UNNnTjsAAAAJ&hl=en&oi=ao",
"url": "https://t.co/DfC9CgWT8N",
"indices": [
0,
23
]
}
]
}
},
"fast_followers_count": 0,
"favourites_count": 3832,
"followers_count": 625,
"friends_count": 1808,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 3,
"location": "London, England",
"media_count": 8,
"name": "Gholamali Aminian",
"normal_followers_count": 625,
"pinned_tweet_ids_str": [],
"possibly_sensitive": false,
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1703493195605151744/5K-q6Gtw_normal.jpg",
"profile_interstitial_type": "",
"screen_name": "gh_aminian",
"statuses_count": 335,
"translator_type": "none",
"url": "https://t.co/DfC9CgWT8N",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"tipjar_settings": {}
}
}
},
"unmention_data": {},
"edit_control": {
"edit_tweet_ids": [
"1927522250569171093"
],
"editable_until_msecs": "1748395690000",
"is_edit_eligible": false,
"edits_remaining": "5"
},
"is_translatable": false,
"views": {
"count": "136",
"state": "EnabledWithCount"
},
"source": "<a href=\"https://mobile.twitter.com\" rel=\"nofollow\">Twitter Web App</a>",
"legacy": {
"bookmark_count": 0,
"bookmarked": false,
"created_at": "Wed May 28 00:28:10 +0000 2025",
"conversation_id_str": "1926839185371017265",
"display_text_range": [
14,
179
],
"entities": {
"hashtags": [],
"symbols": [],
"timestamps": [],
"urls": [
{
"display_url": "arxiv.org/pdf/2502.01203",
"expanded_url": "https://arxiv.org/pdf/2502.01203",
"url": "https://t.co/qoJZ09jLot",
"indices": [
156,
179
]
}
],
"user_mentions": [
{
"id_str": "1778258634889383937",
"name": "YIFENG LIU",
"screen_name": "YIFENGLIU_AI",
"indices": [
0,
13
]
}
]
},
"favorite_count": 1,
"favorited": false,
"full_text": "@YIFENGLIU_AI Thanks for sharing your interesting work. \nWe also looked at forward KL from theoretical perspective and multiple reference models scenario.\n\nhttps://t.co/qoJZ09jLot",
"in_reply_to_screen_name": "YIFENGLIU_AI",
"in_reply_to_status_id_str": "1926839185371017265",
"in_reply_to_user_id_str": "1778258634889383937",
"is_quote_status": false,
"lang": "en",
"possibly_sensitive": false,
"possibly_sensitive_editable": true,
"quote_count": 0,
"reply_count": 1,
"retweet_count": 0,
"retweeted": false,
"user_id_str": "704389635007500292",
"id_str": "1927522250569171093"
},
"quick_promote_eligibility": {
"eligibility": "IneligibleNotProfessional"
}
}
},
"tweetDisplayType": "Tweet"
},
"clientEventInfo": {
"component": "tweet",
"element": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
},
{
"entryId": "conversationthread-1927522250569171093-tweet-1927855417113084231",
"item": {
"itemContent": {
"itemType": "TimelineTweet",
"__typename": "TimelineTweet",
"tweet_results": {
"result": {
"__typename": "Tweet",
"rest_id": "1927855417113084231",
"has_birdwatch_notes": false,
"core": {
"user_results": {
"result": {
"__typename": "User",
"id": "VXNlcjoxNzc4MjU4NjM0ODg5MzgzOTM3",
"rest_id": "1778258634889383937",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": false,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": false,
"can_media_tag": true,
"created_at": "Thu Apr 11 03:08:00 +0000 2024",
"default_profile": true,
"default_profile_image": false,
"description": "CS Ph.D. student on LLM @ UCLA AGI Lab. Previous works: RPG, MARS, TPA, Kimi-1.5....",
"entities": {
"description": {
"urls": []
},
"url": {
"urls": [
{
"display_url": "lauyikfung.github.io",
"expanded_url": "http://lauyikfung.github.io",
"url": "https://t.co/p0AfEGpLJX",
"indices": [
0,
23
]
}
]
}
},
"fast_followers_count": 0,
"favourites_count": 96,
"followers_count": 170,
"friends_count": 25,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 4,
"location": "Los Angeles",
"media_count": 9,
"name": "YIFENG LIU",
"normal_followers_count": 170,
"pinned_tweet_ids_str": [],
"possibly_sensitive": false,
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1926831203694280705/W4z2Gwxi_normal.jpg",
"profile_interstitial_type": "",
"screen_name": "YIFENGLIU_AI",
"statuses_count": 37,
"translator_type": "none",
"url": "https://t.co/p0AfEGpLJX",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"tipjar_settings": {}
}
}
},
"unmention_data": {},
"edit_control": {
"edit_tweet_ids": [
"1927855417113084231"
],
"editable_until_msecs": "1748475123000",
"is_edit_eligible": false,
"edits_remaining": "5"
},
"is_translatable": false,
"views": {
"count": "52",
"state": "EnabledWithCount"
},
"source": "<a href=\"https://mobile.twitter.com\" rel=\"nofollow\">Twitter Web App</a>",
"legacy": {
"bookmark_count": 0,
"bookmarked": false,
"created_at": "Wed May 28 22:32:03 +0000 2025",
"conversation_id_str": "1926839185371017265",
"display_text_range": [
12,
56
],
"entities": {
"hashtags": [],
"symbols": [],
"timestamps": [],
"urls": [],
"user_mentions": [
{
"id_str": "704389635007500292",
"name": "Gholamali Aminian",
"screen_name": "gh_aminian",
"indices": [
0,
11
]
}
]
},
"favorite_count": 2,
"favorited": false,
"full_text": "@gh_aminian Thank you for sharing your interesting work!",
"in_reply_to_screen_name": "gh_aminian",
"in_reply_to_status_id_str": "1927522250569171093",
"in_reply_to_user_id_str": "704389635007500292",
"is_quote_status": false,
"lang": "en",
"quote_count": 0,
"reply_count": 0,
"retweet_count": 0,
"retweeted": false,
"user_id_str": "1778258634889383937",
"id_str": "1927855417113084231"
},
"quick_promote_eligibility": {
"eligibility": "IneligibleNotProfessional"
}
}
},
"tweetDisplayType": "Tweet"
},
"clientEventInfo": {
"component": "tweet",
"element": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
}
],
"metadata": {
"conversationMetadata": {
"allTweetIds": [
"1927855417113084231"
],
"enableDeduplication": true
}
},
"displayType": "VerticalConversation",
"clientEventInfo": {
"component": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
},
{
"entryId": "conversationthread-1926948219977621620",
"sortIndex": "1953556922426720206",
"content": {
"entryType": "TimelineTimelineModule",
"__typename": "TimelineTimelineModule",
"items": [
{
"entryId": "conversationthread-1926948219977621620-tweet-1926948219977621620",
"item": {
"itemContent": {
"itemType": "TimelineTweet",
"__typename": "TimelineTweet",
"tweet_results": {
"result": {
"__typename": "Tweet",
"rest_id": "1926948219977621620",
"has_birdwatch_notes": false,
"core": {
"user_results": {
"result": {
"__typename": "User",
"id": "VXNlcjoxNjMyNzMxMjYyMzEyOTMxMzI5",
"rest_id": "1632731262312931329",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": true,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": false,
"can_media_tag": true,
"created_at": "Mon Mar 06 13:14:11 +0000 2023",
"default_profile": true,
"default_profile_image": false,
"description": "App to supercharge your research",
"entities": {
"description": {
"urls": []
},
"url": {
"urls": [
{
"display_url": "synthical.com",
"expanded_url": "https://synthical.com",
"url": "https://t.co/ejn0MBwlau",
"indices": [
0,
23
]
}
]
}
},
"fast_followers_count": 0,
"favourites_count": 200,
"followers_count": 513,
"friends_count": 2,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 9,
"location": "Paris, France",
"media_count": 2,
"name": "Synthical",
"normal_followers_count": 513,
"pinned_tweet_ids_str": [
"1845085484449505511"
],
"possibly_sensitive": false,
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1669378971211907073/OOnOchbn_normal.jpg",
"profile_interstitial_type": "",
"screen_name": "synthical_ai",
"statuses_count": 10208,
"translator_type": "none",
"url": "https://t.co/ejn0MBwlau",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"tipjar_settings": {}
}
}
},
"card": {
"rest_id": "https://t.co/MpfgWCY3rR",
"legacy": {
"binding_values": [
{
"key": "photo_image_full_size_large",
"value": {
"image_value": {
"alt": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"height": 419,
"width": 800,
"url": "https://pbs.twimg.com/card_img/1952700785202212864/_7gsWwcu?format=png&name=800x419"
},
"type": "IMAGE"
}
},
{
"key": "thumbnail_image",
"value": {
"image_value": {
"alt": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"height": 147,
"width": 280,
"url": "https://pbs.twimg.com/card_img/1952700785202212864/_7gsWwcu?format=png&name=280x150"
},
"type": "IMAGE"
}
},
{
"key": "description",
"value": {
"string_value": "Policy gradient algorithms have been successfully applied to enhance the reasoning capabilities of large language models (LLMs). Despite the widespread use of Kullback-Leibler (KL) regularization in...",
"type": "STRING"
}
},
{
"key": "domain",
"value": {
"string_value": "synthical.com",
"type": "STRING"
}
},
{
"key": "thumbnail_image_large",
"value": {
"image_value": {
"alt": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"height": 315,
"width": 600,
"url": "https://pbs.twimg.com/card_img/1952700785202212864/_7gsWwcu?format=png&name=600x600"
},
"type": "IMAGE"
}
},
{
"key": "summary_photo_image_small",
"value": {
"image_value": {
"alt": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"height": 202,
"width": 386,
"url": "https://pbs.twimg.com/card_img/1952700785202212864/_7gsWwcu?format=png&name=386x202"
},
"type": "IMAGE"
}
},
{
"key": "thumbnail_image_original",
"value": {
"image_value": {
"alt": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"height": 630,
"width": 1200,
"url": "https://pbs.twimg.com/card_img/1952700785202212864/_7gsWwcu?format=png&name=orig"
},
"type": "IMAGE"
}
},
{
"key": "photo_image_full_size_small",
"value": {
"image_value": {
"alt": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"height": 202,
"width": 386,
"url": "https://pbs.twimg.com/card_img/1952700785202212864/_7gsWwcu?format=png&name=386x202"
},
"type": "IMAGE"
}
},
{
"key": "summary_photo_image_large",
"value": {
"image_value": {
"alt": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"height": 419,
"width": 800,
"url": "https://pbs.twimg.com/card_img/1952700785202212864/_7gsWwcu?format=png&name=800x419"
},
"type": "IMAGE"
}
},
{
"key": "thumbnail_image_small",
"value": {
"image_value": {
"alt": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"height": 76,
"width": 144,
"url": "https://pbs.twimg.com/card_img/1952700785202212864/_7gsWwcu?format=png&name=144x144"
},
"type": "IMAGE"
}
},
{
"key": "thumbnail_image_x_large",
"value": {
"image_value": {
"alt": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"height": 630,
"width": 1200,
"url": "https://pbs.twimg.com/card_img/1952700785202212864/_7gsWwcu?format=png&name=2048x2048_2_exp"
},
"type": "IMAGE"
}
},
{
"key": "photo_image_full_size_original",
"value": {
"image_value": {
"alt": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"height": 630,
"width": 1200,
"url": "https://pbs.twimg.com/card_img/1952700785202212864/_7gsWwcu?format=png&name=orig"
},
"type": "IMAGE"
}
},
{
"key": "photo_image_full_size_alt_text",
"value": {
"string_value": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"type": "STRING"
}
},
{
"key": "vanity_url",
"value": {
"scribe_key": "vanity_url",
"string_value": "synthical.com",
"type": "STRING"
}
},
{
"key": "photo_image_full_size",
"value": {
"image_value": {
"alt": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"height": 314,
"width": 600,
"url": "https://pbs.twimg.com/card_img/1952700785202212864/_7gsWwcu?format=png&name=600x314"
},
"type": "IMAGE"
}
},
{
"key": "summary_photo_image_alt_text",
"value": {
"string_value": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"type": "STRING"
}
},
{
"key": "thumbnail_image_color",
"value": {
"image_color_value": {
"palette": [
{
"rgb": {
"blue": 0,
"green": 0,
"red": 0
},
"percentage": 64.25
},
{
"rgb": {
"blue": 255,
"green": 255,
"red": 255
},
"percentage": 28.98
},
{
"rgb": {
"blue": 113,
"green": 113,
"red": 113
},
"percentage": 9.03
}
]
},
"type": "IMAGE_COLOR"
}
},
{
"key": "title",
"value": {
"string_value": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"type": "STRING"
}
},
{
"key": "summary_photo_image_color",
"value": {
"image_color_value": {
"palette": [
{
"rgb": {
"blue": 0,
"green": 0,
"red": 0
},
"percentage": 64.25
},
{
"rgb": {
"blue": 255,
"green": 255,
"red": 255
},
"percentage": 28.98
},
{
"rgb": {
"blue": 113,
"green": 113,
"red": 113
},
"percentage": 9.03
}
]
},
"type": "IMAGE_COLOR"
}
},
{
"key": "summary_photo_image_x_large",
"value": {
"image_value": {
"alt": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"height": 630,
"width": 1200,
"url": "https://pbs.twimg.com/card_img/1952700785202212864/_7gsWwcu?format=png&name=2048x2048_2_exp"
},
"type": "IMAGE"
}
},
{
"key": "summary_photo_image",
"value": {
"image_value": {
"alt": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"height": 314,
"width": 600,
"url": "https://pbs.twimg.com/card_img/1952700785202212864/_7gsWwcu?format=png&name=600x314"
},
"type": "IMAGE"
}
},
{
"key": "photo_image_full_size_color",
"value": {
"image_color_value": {
"palette": [
{
"rgb": {
"blue": 0,
"green": 0,
"red": 0
},
"percentage": 64.25
},
{
"rgb": {
"blue": 255,
"green": 255,
"red": 255
},
"percentage": 28.98
},
{
"rgb": {
"blue": 113,
"green": 113,
"red": 113
},
"percentage": 9.03
}
]
},
"type": "IMAGE_COLOR"
}
},
{
"key": "photo_image_full_size_x_large",
"value": {
"image_value": {
"alt": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"height": 630,
"width": 1200,
"url": "https://pbs.twimg.com/card_img/1952700785202212864/_7gsWwcu?format=png&name=2048x2048_2_exp"
},
"type": "IMAGE"
}
},
{
"key": "card_url",
"value": {
"scribe_key": "card_url",
"string_value": "https://t.co/MpfgWCY3rR",
"type": "STRING"
}
},
{
"key": "summary_photo_image_original",
"value": {
"image_value": {
"alt": "On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning",
"height": 630,
"width": 1200,
"url": "https://pbs.twimg.com/card_img/1952700785202212864/_7gsWwcu?format=png&name=orig"
},
"type": "IMAGE"
}
}
],
"card_platform": {
"platform": {
"audience": {
"name": "production"
},
"device": {
"name": "Android",
"version": "12"
}
}
},
"name": "summary_large_image",
"url": "https://t.co/MpfgWCY3rR",
"user_refs_results": []
}
},
"unmention_data": {},
"edit_control": {
"edit_tweet_ids": [
"1926948219977621620"
],
"editable_until_msecs": "1748258831000",
"is_edit_eligible": false,
"edits_remaining": "5"
},
"is_translatable": false,
"views": {
"count": "131",
"state": "EnabledWithCount"
},
"source": "<a href=\"https://example.com/callback\" rel=\"nofollow\">1858442586446233600JamesDread1</a>",
"legacy": {
"bookmark_count": 0,
"bookmarked": false,
"created_at": "Mon May 26 10:27:11 +0000 2025",
"conversation_id_str": "1926839185371017265",
"display_text_range": [
14,
82
],
"entities": {
"hashtags": [],
"symbols": [],
"timestamps": [],
"urls": [
{
"display_url": "synthical.com/abs/2505.17508…",
"expanded_url": "https://synthical.com/abs/2505.17508?is_dark=true&utm_source=dark_medium",
"url": "https://t.co/MpfgWCY3rR",
"indices": [
59,
82
]
}
],
"user_mentions": [
{
"id_str": "1778258634889383937",
"name": "YIFENG LIU",
"screen_name": "YIFENGLIU_AI",
"indices": [
0,
13
]
}
]
},
"favorite_count": 1,
"favorited": false,
"full_text": "@YIFENGLIU_AI Dark mode for this paper for night readers 🌙 https://t.co/MpfgWCY3rR",
"in_reply_to_screen_name": "YIFENGLIU_AI",
"in_reply_to_status_id_str": "1926839185371017265",
"in_reply_to_user_id_str": "1778258634889383937",
"is_quote_status": false,
"lang": "en",
"possibly_sensitive": false,
"possibly_sensitive_editable": true,
"quote_count": 0,
"reply_count": 0,
"retweet_count": 0,
"retweeted": false,
"user_id_str": "1632731262312931329",
"id_str": "1926948219977621620"
},
"quick_promote_eligibility": {
"eligibility": "IneligibleNotProfessional"
}
}
},
"tweetDisplayType": "Tweet"
},
"clientEventInfo": {
"component": "tweet",
"element": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
}
],
"metadata": {
"conversationMetadata": {
"allTweetIds": [
"1926948219977621620"
],
"enableDeduplication": true
}
},
"displayType": "VerticalConversation",
"clientEventInfo": {
"component": "tweet",
"details": {
"conversationDetails": {
"conversationSection": "HighQuality"
},
"timelinesDetails": {
"controllerData": "DAACDAAEDAABCgABAAAAAAAAAAEKAAIAAAAAAAAAAAAAAAA="
}
}
}
}
},
{
"entryId": "cursor-bottom-1953556922426720205",
"sortIndex": "1953556922426720205",
"content": {
"entryType": "TimelineTimelineCursor",
"__typename": "TimelineTimelineCursor",
"value": "DAAFCgABGxxtq6m__8sLAAIAAACoRW1QQzZ3QUFBZlEvZ0dKTjB2R3AvQUFBQUEwYXZZSUs0SmV3RVJxOWdnaWRWOEcrR3IyQ0I3OVhzREVhdllJS1Q1YUFCeHE5Z2d1NVZrSFdHc0VlU2FOV2dVY2F2dEtibkJjQlJocS9QMFJ2VnNEa0dyM2xNbGhXWUhRYXZZSUpjUmV4QnhxLzcwWkIxa0NWR3NFZGNMdFhjTzBhdnRkZmdCWUI2dz09CAADAAAAAgAA",
"cursorType": "Bottom"
}
}
]
},
{
"type": "TimelineTerminateTimeline",
"direction": "Top"
},
{
"type": "TimelineTerminateTimeline",
"direction": "Bottom"
}
],
"metadata": {
"reader_mode_config": {
"is_reader_mode_available": true
},
"scribeConfig": {
"page": "ranked_replies"
}
}
}
}
}