@arankomatsuzaki
- Analogous to registers for LM - Notable improvements on some tasks (e.g. SQuAD, NaturalQA, CommonSenseQA) - It doesn't work on every task and it's not a replacement of CoT https://t.co/ZZ6t6d331a
Viewing enriched Twitter post
- Analogous to registers for LM - Notable improvements on some tasks (e.g. SQuAD, NaturalQA, CommonSenseQA) - It doesn't work on every task and it's not a replacement of CoT https://t.co/ZZ6t6d331a
{
"user": {
"created_at": "2016-11-04T06:57:37.000Z",
"default_profile_image": false,
"description": "ML PhD @ GaTech, @TheDuckAI (https://t.co/TLZw9TIidr), EleutherAI, LAION",
"fast_followers_count": 0,
"favourites_count": 8669,
"followers_count": 68235,
"friends_count": 76,
"has_custom_timelines": true,
"is_translator": false,
"listed_count": 998,
"location": "",
"media_count": 1472,
"name": "Aran Komatsuzaki",
"normal_followers_count": 68235,
"possibly_sensitive": false,
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1561220982328754176/JOYS5kab_normal.jpg",
"screen_name": "arankomatsuzaki",
"statuses_count": 3836,
"translator_type": "none",
"url": "https://t.co/aZGCShnLYq",
"verified": false,
"withheld_in_countries": [],
"id_str": "794433401591693312"
},
"id": "1709377922358755822",
"conversation_id": "1709377922358755822",
"full_text": "- Analogous to registers for LM\n- Notable improvements on some tasks (e.g. SQuAD, NaturalQA, CommonSenseQA)\n- It doesn't work on every task and it's not a replacement of CoT https://t.co/ZZ6t6d331a",
"reply_count": 0,
"retweet_count": 5,
"favorite_count": 46,
"hashtags": [],
"symbols": [],
"user_mentions": [],
"urls": [],
"media": [
{
"media_url": "https://pbs.twimg.com/media/F7jrax0X0AARewX.jpg",
"type": "photo"
}
],
"url": "https://twitter.com/arankomatsuzaki/status/1709377922358755822",
"created_at": "2023-10-04T01:20:30.000Z",
"#sort_index": "1709377922358755822",
"view_count": 7504,
"quote_count": 0,
"is_quote_tweet": true,
"is_retweet": false,
"is_pinned": false,
"is_truncated": false,
"quoted_tweet": {
"user": {
"created_at": "2016-11-04T06:57:37.000Z",
"default_profile_image": false,
"description": "ML PhD @ GaTech, @TheDuckAI (https://t.co/TLZw9TIidr), EleutherAI, LAION",
"fast_followers_count": 0,
"favourites_count": 8669,
"followers_count": 68235,
"friends_count": 76,
"has_custom_timelines": true,
"is_translator": false,
"listed_count": 998,
"location": "",
"media_count": 1472,
"name": "Aran Komatsuzaki",
"normal_followers_count": 68235,
"possibly_sensitive": false,
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1561220982328754176/JOYS5kab_normal.jpg",
"screen_name": "arankomatsuzaki",
"statuses_count": 3836,
"translator_type": "none",
"url": "https://t.co/aZGCShnLYq",
"verified": false,
"withheld_in_countries": [],
"id_str": "794433401591693312"
},
"id": "1709372124891070915",
"conversation_id": "1709372124891070915",
"full_text": "Think before you speak: Training Language Models With Pause Tokens\n\n- Performing training and inference on LMs with a learnable pause token appended to the input prefix\n- Gains on 8 tasks, e,g, +18% on SQuAD\n\nhttps://t.co/snkfjFZhhZ https://t.co/wUhZspVtSj",
"reply_count": 18,
"retweet_count": 169,
"favorite_count": 917,
"hashtags": [],
"symbols": [],
"user_mentions": [],
"urls": [
{
"url": "https://t.co/snkfjFZhhZ",
"expanded_url": "https://arxiv.org/abs/2310.02226",
"display_url": "arxiv.org/abs/2310.02226"
}
],
"media": [
{
"media_url": "https://pbs.twimg.com/media/F7joVYLXYAACeL0.png",
"type": "photo"
}
],
"url": "https://twitter.com/arankomatsuzaki/status/1709372124891070915",
"created_at": "2023-10-04T00:57:27.000Z",
"#sort_index": "1709377922358755800",
"view_count": 343658,
"quote_count": 41,
"is_quote_tweet": false,
"is_retweet": false,
"is_pinned": false,
"is_truncated": false
},
"startUrl": "https://twitter.com/arankomatsuzaki/status/1709377922358755822"
}