Harry_Filet

探索Web3开发者，Filecoin开发者

[Harry_Filet @HarryM0507]

@blackanger 大佬太强，学习

[Harry_Filet @HarryM0507]

@chunxiangai 给个邀请码耍耍

[Harry_Filet @HarryM0507]

@Pluvio9yte 只要bill不开，过期了谷歌云是不会收费的，到90天，或者3个月，哪个条件先到就结束，这一点谷歌还是很好的

[Harry_Filet @HarryM0507]

@op7418 大佬，感谢来个码

[Harry_Filet @HarryM0507]

@tualatrix 祝大卖

[Harry_Filet @HarryM0507]

@LuoSays 高级黑😂

[Harry_Filet @HarryM0507]

@dotey @YanyuRensheng 听劝，入���o1 pro

[Jim Fan @DrJimFan]

We are living in a timeline where a non-US company is keeping the original mission of OpenAI alive - truly open, frontier research that empowers all. It makes no sense. The most entertaining outcome is the most likely.

DeepSeek-R1 not only open-sources a barrage of models but also spills all the training secrets. They are perhaps the first OSS project that shows major, sustained growth of an RL flywheel.

Impact can be done by "ASI achieved internally" or mythical names like "Project Strawberry".
Impact can also be done by simply dumping the raw algorithms and matplotlib learning curves.

I'm reading the paper:

> Purely driven by RL, no SFT at all ("cold start"). Reminiscent of AlphaZero - master Go, Shogi, and Chess from scratch, without imitating human grandmaster moves first. This is the most significant takeaway from the paper.
> Use groundtruth rewards computed by hardcoded rules. Avoid any learned reward models that RL can easily hack against.
> Thinking time of the model steadily increases as training proceeds - this is not pre-programmed, but an emergent property!
> Emergence of self-reflection and exploration behaviors.
> GRPO instead of PPO: it removes the critic net from PPO and uses the average reward of multiple samples instead. Simple method to reduce memory use. Note that GRPO was also invented by DeepSeek in Feb 2024 ... what a cracked team.

[Harry_Filet @HarryM0507]

@phyrexni 这样搞，刚上任就遭唾沫

[Harry_Filet @HarryM0507]

@op7418 @poe_platform 还有windsurf

[Harry_Filet @HarryM0507]

@TeddyAC01 生日快乐

[Harry_Filet @HarryM0507]

@phyrexni P大，那之前很多人说sol ETF可能性，您怎么看

[Harry_Filet @HarryM0507]

@mrbear1024 想问下大佬，目前是否是买房的时机

[Harry_Filet @HarryM0507]

@dotey 这是咋做出来的

[Harry_Filet @HarryM0507]

@witunix @imsingee @ftium4 你是什么硬盘？这夸张了吧，

[Harry_Filet @HarryM0507]

@follow_clues 大佬sol值不值得加上

[AI Will @FinanceYF5]

我们正在构建最先进的 Apple Vision Pro -> 人形机器人堆栈。 他的学生奠定了基础。看看他们的作品吧！

https://t.co/vJh1guNDvW

[AI Will @FinanceYF5]

MimicGen：我们的生成动作框架。该代码对于机器人手臂是开源的，但我们将为人形和 5 指手提供另一个版本。
https://t.co/T8s4GsotmM

[AI Will @FinanceYF5]

RoboCasa：我们的生成式仿真框架。它是完全开源的！

https://t.co/wOgrAh4OIv
https://t.co/yqo1Qre40r

[AI Will @FinanceYF5]

NVIDIA科学家Jim Fan发布了关于GR00T项目的令人兴奋的更新：

我们发现了一种系统性的方式来扩大机器人数据，解决了机器人学中最痛苦的痛点。这个想法很简单：人类在真实的机器人上收集演示数据，我们在模拟中将这些数据增加1000倍或更多。

原文：