Pico | (✱,✱)

3.4K posts

Pico | (✱,✱) banner
Pico | (✱,✱)

Pico | (✱,✱)

@_Picokel

Beta Testeur | Based Contributor | Defi Explorer. 🎒 Contributing in little ways I can. 🧡 Tried proving my love to @SuccinctLabs and @themaitrixai.

web3 Katılım Şubat 2024
489 Takip Edilen292 Takipçiler
Sabitlenmiş Tweet
Pico | (✱,✱)
Pico | (✱,✱)@_Picokel·
Highlights of today's @axisrobotics VC session • Axis is looking to collect high quality and high scale data. • There would be a badge system soon • Axis doesn't deal with making robotics themselves now, only data collection for training robots
Pico | (✱,✱) tweet media
English
2
0
8
330
ryzzu
ryzzu@0xRyzzu·
big day tomorrow, gaxis
English
2
0
4
62
mysti ✨
mysti ✨@yagirl·
Good morning :)) Any tickers today?
English
98
4
227
14.1K
Zack Brenner
Zack Brenner@zjbrenner·
There is nothing more powerful than community feedback! 🫶 Every day, we onboard more people to the @opensea Mobile beta and they directly shape what’s coming next in the product 🙏
The Ghost.HL🇦🇱@theghost_alb

Had a call with @zjbrenner and got early access to the @opensea app He walked me through the whole experience and honestly… I was impressed. The app is super easy to use, very clean visually and extremely fast. Everything feels smooth. What I like the most is that it’s not just about NFTs You can trade tokens across multiple chains, bridge assets, track everything in one place and explore pretty much the whole Web3 ecosystem from the app. Tested it today and it will definitely make things much easier for me when browsing and buying NFTs. Much respect to Zack for the work he’s putting into this

English
26
7
77
4.5K
Zack Brenner
Zack Brenner@zjbrenner·
Yay or Nay? 📍 Kopitiam, NYC
Zack Brenner tweet media
HT
73
1
131
6.9K
Pico | (✱,✱)
Pico | (✱,✱)@_Picokel·
New tasks are coming to the @axisrobotics web-based simulation tomorrow (Monday). 🖥️ Here's what you should know: • Task participation history would recorded onchain, on Base Network mainnet. • Tasks would have a set participation cap. After the participation cap is reached, you can't complete that task. 🪧 • You'll get multiple tries at completing the tasks. Ensure to participate before participation cap is reached. • About 5-10 new tasks drop every 1-2 days.⏳ • Badges would be given when certain requirements are met.🎖️ Goodluck
Pico | (✱,✱) tweet media
English
0
0
5
62
Tony O. Elumelu, CFR
Tony O. Elumelu, CFR@TonyOElumelu·
Today, I turn a year older. And every year on this day, I reflect on something far bigger than me. For a long time, I believed luck was something that just happens to you. Then I realised, luck can be engineered. Opportunity can be democratised. Hope is not just a feeling; it is a system we can build.
Tony O. Elumelu, CFR tweet mediaTony O. Elumelu, CFR tweet media
English
5.7K
3.9K
34.9K
664.4K
Pico | (✱,✱)
Pico | (✱,✱)@_Picokel·
@0xzagen The end point is that robotics need simulation+ real data to scale, or refining & optimization of simulation data for it to be high quality
English
0
0
0
11
Pico | (✱,✱) retweetledi
zagen扎根(✱,✱)
机器人到底需要什么数据?⬇️ 银河通用机器人宣布在三月份完成25亿新一轮融资,投资方包括多家国家队选手:国家人工智能产业投资基金、中国石化等等,这代表着来自更高层次的对于机器人仿真数据赛道的认可。 在机器人数据赛道,仿真数据与真机数据的地位一直是一个争议巨大的话题,代表性的关键意见人物就是银河通用的王鹤和星海图的高继扬,分别对应仿真合成数据和真机数据。 综合两位的观点,核心就是: 仿真数据: 可规模化,边际成本趋零;同时包含多模态信息,适配端到端VLA模型训练;场景覆盖度随着任务训练场景可以无限制扩展; 真机数据: 采集成本“高”,但是综合成本低;针对特定领域,最好的训练方式就是使用该领域的数据而非大数据集的迁移;仿真数据和物理世界的差距(sim2real gap)一直没被攻克,仿真中表现很好的模型在真实世界很容易失效; 星海图的真实运营成本计算下来: 1小时高质量真实操作数据的成本为200-250元 这意味着10万小时的真实数据(相当于一个人从出生到18岁清醒状态下与物理世界的总交互时长数量级)成本仅需2500万,低于3000万-5000万的千亿级别参数大模型的单次与训练成本 高认为,真实数据必须在真实场景中、以众包的方式采集;数据获取成本和模型训练成本的比例,是1:5-1:10,低质量的训练数据本质是捡了芝麻丢了西瓜,让训练环节的成本大幅提升; 银河通用的训练比例是99%的仿真+1%的真实数据; 星海图是真实数据为主导; 英伟达的通用具身基础模型是大约60%的仿真数据; 波士顿动力的atlas以真机数据为核心; 特斯拉的optimus采用仿真与训练+真机遥操数据微调; 但是共识是: 1. 数据是具身智能当前的核心问题 2. 低质量的仿真数据缺乏价值 3. 仿真合成数据+真机数据混合训练是必然,具体的比例各家有各家的判断,且需要根据实践的场景(通用 or 垂类)在实际训练中调整recipe 4. 真实数据规模化难度高,需要厂商在软硬件端都有深厚的积累 5. 无论仿真数据还是真实数据,都要形成数据-训练-验证-迭代的循环才能确保数据质量,不懂模型训练的人做不出好的数据集 @axisrobotics 端到端的从数据采集到数据处理到模型训练到物理世界部署,已经完成了可行性验证。
zagen扎根(✱,✱) tweet media
Axis Robotics@axisrobotics

Today's robots can be programmed to perform specific tasks, but they fundamentally lack the autonomy to make decisions in dynamic physical environments. Axis Robotics is changing that. We are building the intelligence foundation for robots by bridging the physical AI data gap. Through a browser-based simulation platform, Axis Robotics lets anyone train robots virtually through intuitive interactions. Every contribution of data is transparently tracked, verified, and rewarded on blockchain—creating a scalable, distributed data engine for the future of Physical AI.

中文
8
2
7
1.2K
Pico | (✱,✱) retweetledi
Forte Protocol
Forte Protocol@ForteProtocol·
1/ We’re building a perpetual protocol around a simple idea: outcomes should come from market direction and position management, not from hard-to-model exchange mechanics.
English
251
8.9K
9.9K
69.9K
ER1CK || W3B 👑
ER1CK || W3B 👑@erick_crypt·
@Abyomiii Still trying to understand but I can see the potential. Better get on it sooner
English
2
0
2
22
Pico | (✱,✱)
Pico | (✱,✱)@_Picokel·
@Abyomiii In our language, what Axis is building x.com/i/status/20348…
Pico | (✱,✱)@_Picokel

Using regional language (Pidgin English) Wetin Axis Robotics deh Build? Wetin @axisrobotics deh build na simulated environment for teleoperation where anybody go fit get beta high-quality data for robotics training / improvement. ⚙️ By simulated environment, I mean something wey be like game environment, e deh "virtual". 🖥️ You don play mini militia na? You go gree say things deh happen, like characters deh move around and bullets deh fly around with real physics(motion, gravity, etc), but no be real, na virtual thing. 🖥️⚛️ Na the same thing for this simulated environment, but na robotics deh move here with real physics based on how you take control am. Axis robotics don build, and still deh refine their simulated environment with real physics mirroring real life scenarios for teleoperation. (Physical gravity & motion laws still deh respected; so Objects no go hang for air. 😅) For this simulated environment, you go fit teleoperate (guide & control) robotics to complete tasks like "Arrange objects on table", etc. From controlling this robotic in simulated environment, them go record as the robotic take move (en manipulation trajectories) to complete tasks, refine the manipulation trajectory data so e go deh scalable and usable for different real-life scenarios, then save am. 👨🏽‍🔬 From this data now, If any robotics company want source data to train their robotic models, Axis robotics go fit provide this very high-quality data to them at a cost, and revenue go go back to people weh contribute the data. 🔁

English
0
0
1
24