Search results for REI
People
Not Found
Tweets including REI
Reindeer Cyclones are a phenomenon where a herd of anxious reindeer forms a swirling mass, making it difficult to single out individual animals.
The fawns remain protected at the centre of this circular stampede.
https://t.co/DHHxnfrT22
Show more
0
0
21
101
17
Kieran Reilly just crushed it with two world-first BMX Freestyle tricks in one night
https://t.co/ugt3BR9qaO
0
0
3
13
5
Truly Grandma's reincarnate! https://t.co/Y5qBhxCaK2
0
0
120
5.2K
921
THIS IS FOR #REI#
4️⃣ Listen [THIS IS FOR] 4️⃣
https://t.co/g1w24qqi6D
#TWICE# #트와이스# #THISISFOR#
#SANA# #사나#
#IVE# #아이브# #레이# https://t.co/BuSXgkGMJB
0
0
42
21.8K
4.5K
When I reincarnate as a dog, I won’t say anything, but there will be signs. https://t.co/lmtyW2lNxT
0
0
16
42
11
0
0
0
1
0
We’re delighted to reinforce our commitment to the South Island and Canterbury with the newly-named One New Zealand Stadium! https://t.co/AKqURci15T
0
0
1
20
8
🧙♂️💜 It's time to Reinvent Bitcoin 🧡!
What if holding BTC isn’t the endgame — but just the beginning?
@MerlinLayer2 has been building since DAY ONE — from inscriptions & BTCfi to now AI-powered tools. With Merlin 2.0, we are sharpening our focus on a singular vision: Reinventing Bitcoin – Hold, Earn, Invest.
By harnessing advanced AI and chain abstraction, we aim to unlock new liquidity scenarios and enhance Bitcoin’s utility as a global currency.
👇 Here’s how @MerlinLayer2 is building the future of Bitcoin:
HOLD. EARN. INVEST.
🧵
Show more
0
0
26
55
25
David Silver在DeepMind负责Reinforcement Learning（RL，强化学习），是AlphaGo团队的核心。他最近有个分享，有些启发：
学习的三个阶段：
1. 基于现有知识的学习
2. 基于人类已有经验的强化学习
3. 忘记人类经验的自我学习
每个更高阶的学习，都在不断丢掉旧的知识，旧的经验。
更具体而言：
1. 当前LLM主流还是Data驱动。问题在于1）Data会成为一个天花板，2）模型会和当前的Data尽量拟合而自身的学习能力被弱化。3）LLM默认会证明自己是对的，所以经常有幻觉的产生。而强化学习是让AI在实际使用中获得数据，反馈，它不试图证明自己是对的，而是不断地在真实世界中试错来验证。
2. 如果没有强化学习，那么AlphaGo只会学会人类的技巧，而没有它第二局超越人类经验的第37手。经验比知识更重要。
3. 而AlphaZero是AlphaGo的下一代，证明了忘记人类经验学习的价值。AlphaZero从完全随机的行为开始，从系统自身生成的经验中学习的方式，使得 AlphaZero 从零开始达到了并超越了人类在围棋和国际象棋等领域的最高水平。
4. 反馈有两种，基于人类经验的反馈（RLHF）和真实世界的反馈。前者是人类事先根据经验设定的标准，但它在真实世界仍然可能是错的。而后者是真正通过行动在真实世界得到的反馈。这是AlphaZero超越AlphaGo的背后原理。
5. David Silver提出“Reward is enough”。我们训练AI的职责主要是把目标和我们的价值观对齐融合在设定的奖励标准中。比如我们设定一个标准，奖励AI让我们变得更健康的策略。这个符合我们的需求，而且这里并不是具体的目标，而是模糊的高层目标——如何变得健康。AI系统自己去优化具体的目标（心率，BMI等），并根据真实的反馈调整具体目标的组合。
6. Anthropic以及一些团队证明了基础模型仍然有提升的空间，因此，Data驱动和强化学习驱动会是两个持续演进的双螺旋，而强化学习会越来越重要。
进一步：
1. 从哲学上讲，在时间之矢方向上的Randomness和熵增是确定的，因而我们无法完全从既有的数据和经验学习，我们只有不停地探索，无论是AI还是我们自身。
2. 我们几乎所有在AI训练中的启示都适用于我们对下一代的教育。
https://t.co/YbjAcqZF4U
https://t.co/tG9cFMe0t0
https://t.co/jIR5twS3gj
Show more
0
0
2
6
3
An update on COVID-19 reinstatements. https://t.co/o8GVFAlcJN
0
0
2.5K
22.9K
5.3K
Happy Birthday to the incredibly talented Sam Reid! We are so happy you are our Lestat 🩷
We love you, Sam! https://t.co/jvBuFc1Iyd
0
0
24
6.1K
1.1K
happy birthday to the incredibly talented sam reid! so blessed to be able to witness his passion & his nerdiness, the dedication to his craft, and his beautiful, magnetic, endearing soul. wishing him so much love & joy, and many more years of success 💫💕 https://t.co/K1TylIGPmX
Show more
0
0
0
738
141
The UK made a shock decision to reinstate the need for visas for Colombian travellers this week.
There will be a short transition period, but many people's travel plans are in chaos.
Find out more here:
https://t.co/MfImdrPnhq
Show more
0
0
2
1
1
America is now the 4th Reich. Where is our "Churchill"?
0
0
0
1
0
Happy birthday Milady 3 Year Total Internet Reign 1000 Years Remilia Empire Complete Ecosystem Domination Catalyzation of Emergent Futures Quantum-Grade Memetic Warfare Containment Unit Euphoric Obedience Grid Viral Reality Distortion Cheese https://t.co/oxWVO9Omeu
Show more
0
0
15
210
31
6 young adult novels that reimagine and retell Greek mythology. https://t.co/cxy5MXjwQR
0
0
9
19
8
