Home
Language
English
Türkçe
Bahasa Indonesia
About
Privacy Policy
Terms of Service
Pricing
Sign In
Download All
Share
Phil
@philying
Geek/Mac lover/full stack/body builder
shanghai
Joined February 2009
230
Following
122
Followers
1.4K
Posts
Phil
@philying
9 months ago
@i3486
这是和美团点评死磕到底的意思么
Phil
@philying
9 months ago
@i3486
一个人打几十个那段太离谱
Phil
@philying
about 1 year ago
@frxiaobei
@tinyfool
@readwise
save thread
Phil
@philying
about 1 year ago
@frxiaobei
@tinyfool
@Readwiseio
save thread
Who to follow
Tabio
@Destiny2Mars
A dreamer working for blockchain, site reliability, data platform. A founder and invest in cryptocurrency, advertising. Rare SRE & OCP practicer. e/acc.
Lukin
@iLukin
Wild dev, AI-coded vibes. Built Telegram bots used by 100M+. English learning https://t.co/c7S0Y4VOz8, AI https://t.co/wm8nhNE8cj
Zhang Erning
@erning
Phil
@philying
about 1 year ago
@frxiaobei
@tinyfool
@Readwiseio
save
philying
retweeted
xwh
@xuwenhao
over 1 year ago
其实 R1 的训练过程和人的学习成长特别像。SFT Memorize,RL Generalize。所以打工别打太长时间,学会了基本格式就尽量去真实世界RL就好了。 模型需要非常好的 Pre-Train,并且数据质量要高。对人来说,就是要好的通识角度,博览群书,啥科目都要看看,而且要多看教科书,少看低质量短视频和网络帖子。 然后 SFT 帮助你能够强化记忆和能力,基本上大家专门学习特定的专业,以及去大厂打工都是这个逻辑。在某一个领域还是需要监督学习,做题也好,在公司里收到立刻短时过程反馈。 但是 SFT 有个缺点,就是 Overfit,一旦遇到分布外的数据表现急剧下降。所以如果一门心思在大厂打工,习惯了短时的公司内反馈,一旦遇到现在这种行业大变化,离开大厂就会非常痛苦,实际解决真实世界问题的能力并没有那么好。 但是 RL 有很强的泛化能力,所谓人教人不会,事儿教人一教就会。融资创业也好,做个小生意也好,被“稀疏奖励”的现实世界抽两个大嘴巴子立刻就成长了。 但是完全从 RL 学习又很难,所以 RL-Zero 那种一毕业就创业还是很难的。最好是先简单 SFT 一下,拿一点“格式奖励”,学会做事情的基本套路,再跑出来让现实世界对自己打脸。不过千万别长期SFT下去。 Pre-Train => 简单SFT => 然后依赖RL 可以获得最好的泛化能力。
See More
Phil
@philying
over 1 year ago
@i3486
现在这么强了?
Phil
@philying
over 1 year ago
@i3486
退休了?
Phil
@philying
over 1 year ago
@i3486
@xuwenhao
公司问员工借钱发工资?
philying
retweeted
iPaul
@iPaulCanada
over 2 years ago
澳网观众,一辈子的创伤。
Phil
@philying
over 2 years ago
@i3486
最美周芷若
Phil
@philying
over 2 years ago
@caijingshujuku
@readwise
Phil
@philying
almost 3 years ago
@i3486
发现个bug,链接居然不是https
Phil
@philying
about 3 years ago
@FinanceYF5
@readwise
save thread
Phil
@philying
about 3 years ago
@FinanceYF5
@readwise
save thread
Phil
@philying
about 3 years ago
看过足球教练 第一季 https://t.co/XK1PRp5BZl
Phil
@philying
over 3 years ago
@thingnotok
@readwise
save thread
Phil
@philying
over 3 years ago
@xuwenhao
@readwise
save thread
Phil
@philying
over 3 years ago
@0xthefool
@readwise
save thread
Phil
@philying
over 3 years ago
@starzq
@readwise
save thread
Last Seen Users on Sotwe
nepal123
Seen from
Netherlands
Dawn of a New Gay
Seen from
Italy
مايكل عزيز
Seen from
Netherlands
chubby chaser
Seen from
Nigeria
simion.03
✨ 𝐓𝐨𝐥𝐞𝐞𝐧 𓃠♥︎
Seen from
France
คู่อ้วน ชอบเย็ด สาวอ้วนหีใหญ่ สนใจinbox ช.31ญ24
Seen from
Thailand
🇮🇶🏴☠️ إبن ديالى
Nam
Seen from
Vietnam
Old Goes Young
Seen from
Qatar
Trends for you
1
Knicks
Under 10K tweets
2
Wemby
Under 10K tweets
3
Josh Hart
Under 10K tweets
4
ORM VEGGIE POPUP ICONSIAM
Under 10K tweets
5
Kenzie
Under 10K tweets
6
#AEWDynamite
Under 10K tweets
7
Scott Foster
Under 10K tweets
8
#82and0
Under 10K tweets
9
Hop Sing
Under 10K tweets
10
Betrayed
Under 10K tweets
Most Popular Users
1
Elon Musk
@elonmusk
240.1M followers
2
Barack Obama
@barackobama
119.3M followers
3
Donald J. Trump
@realdonaldtrump
111.6M followers
4
Cristiano Ronaldo
@cristiano
108.8M followers
5
Narendra Modi
@narendramodi
106.9M followers
6
Rihanna
@rihanna
97.2M followers
7
NASA
@nasa
92.1M followers
8
Justin Bieber
@justinbieber
90.5M followers
9
KATY PERRY
@katyperry
86.7M followers
10
Taylor Swift
@taylorswift13
80.5M followers
11
Lady Gaga
@ladygaga
72.1M followers
12
Kim Kardashian
@kimkardashian
69.3M followers
13
YouTube
@youtube
68.6M followers
14
Virat Kohli
@imvkohli
68.4M followers
15
Bill Gates
@billgates
63.4M followers
16
The Ellen Show
@theellenshow
62.5M followers
17
CNN
@cnn
61.9M followers
18
Neymar Jr
@neymarjr
60.9M followers
19
X
@x
60.9M followers
20
CNN Breaking News
@cnnbrk
59.9M followers
Olivia
Online
✨
⭐
💫