Reverse-o1:OpenAI o1原理逆向工程图解
OpenAI o1的推出称为横空出世不为过,尽管关于Q*、草莓等各种传闻很久了,用了强化学习增强逻辑推理能力这个大方向大家猜的也八九不离十,但是融合LLM和RL来生成Hidden COT,估计很少人能想到这点,而且目前看效果确实挺好的。
OpenAI奔向Close的路上越走越远,你要从o1官宣字面来看,除了“强化学习生成Hidden COT”外,基本找不到其它有技术含量的内容。Sora好歹还给出了个粗略的技术框架图,字里行间也透漏不少隐含的技术点,细心点总能发现很多蛛丝马迹,串起来之后整个背后的技术就若隐若现(若对此感兴趣可看下我之前写的分析: 技术神秘化的去魅:Sora关键技术逆向工程图解 )。而且,尽管目前有不少公开文献在用LLM+RL增强大模型的推理能力,但几乎找不到做Hidden COT生成的工作,所以可供直接参考的内容非常少,这为分析o1进一步增添了难度。
那是否就没办法 … ⌘ Read more
next-20240925: linux-next
Version:next-20240925 (linux-next)Released:2024-09-25 ⌘ Read more
Late Cenozoic
⌘ Read more
How to Downgrade from iOS 18 Back to iOS 17
If you have recently installed iOS 18 on iPhone, or iPadOS 18 on iPad, and you’re not thrilled with the experience, or maybe you’ve encountered something that is incompatible with your workflow, you can still downgrade from iOS 18 back to iOS 17. Specifically, currently you can downgrade an iPhone or iPad with iOS 18 … Read More ⌘ Read more
(#2024-09-24T12:53:35Z) What does this screenshot show? The resolution it too low for reading the text…
(#2024-09-24T12:45:54Z) @prologic@twtxt.net I’m not really buying this one about readability. It’s easy to recognize that this is a URL and a date, so you skim over it like you would we mentions and markdown links and images. If you are not suppose to read the raw file, then we might a well jam everything into JSON like mastodon
„Otče náš” Martina Tomana Banátskeho
Spisovateľ, redaktor a úradník Martin Toman Banátsky sa narodil pred 125 rokmi 24. septembra1899 v Kovačici. V povojnovom období (1947 – 1949) bol v Báčskom Petrovci šéfredaktorom Hlasu ľudu a v roku 1949 bol prvým šéfredaktorom Nového života. „Martina Tomana Banátskeho som poznal predovšetkých ako básnika. Krátko pobudol v Petrovci. Väčšiu časť trávil v redakcii a ja som sa stretal s ním iba na zasadnutiach. Hovoril málo, ale rozvážne. Zjavom a … ⌘ Read more
Telegram Will Now Give Personal Data to Governments & Use AI to Moderate Content
After CEO Pavel Durov’s arrest, Telegram has drastically changed policies. ⌘ Read more
Linux on C64, 8086, & Intel 4004
With a little work, Linux can boot on 8 and 4 bit CPUs from the 1970s. Slowly. ⌘ Read more
A weekend with my family
This past weekend, I visited my family in the south of Germany. I wasn’t there for quite some time. On one day, we went to Biel in Switzerland, walking through the Taubenloch (“pigeonhole”, a canyon right next to the city) and sitting on a boat that took us across Lake Biel. It was quite picturesque. ⌘ Read more
Kubestronaut in Orbit: Camila Soares Câmara
Get to know Camila This week’s Kubestronaut in Orbit, Camila Soares Câmara, is a Senior Cloud Engineer at Wellhub in Brazil with experience in Cloud and DevOps, working with technologies such as Kubernetes, CI/CD, AWS, and Infrastructure as… ⌘ Read more
2024 Docker State of Application Development Survey: Share Your Thoughts on Development
Take the 2024 Docker State of Application Development Survey now. The survey is open from September 23rd, 2024 (7AM PST) to November 20, 2024 (11:59PM PST). ⌘ Read more
Aggred. But reading twtxt in raw form sounds… I can’t do this
Finally pubnix is alive! That’s im missing? Im only reading twtxt.net timeline because twtxt-v2.sh works slowly for displaying timeline…
你一直坚持热爱的事情是什么?坚持了多久?
谢邀 @央视新闻
知乎的朋友们好啊,我是打了50多年的乒乓球的运动员倪夏莲。
常常有年轻的朋友来问我是怎么坚持下来的,诚实地讲,我也搞不清楚。从没想过自己能打这么久,想放弃的时刻也多到数不清,但兜兜转转了这么些年,自己还是没能离开这颗小球。
你说坚持是因为热爱吗?肯定是的。但这五十多年里,在热爱之外,夹杂了太多其他东西。
上学那会儿,从电视里看到乒乓球赛的转播,拿冠军的人就像英雄一样,可威风了。加上学校条件比较好,有八个乒乓球台,于是我就报名去打球。
那时候我也喜欢唱歌,中间一度跑去了合唱队,但去了三天又后悔,总是想着那颗球。我特别幸运,老师觉得我是棵好苗子,又给了我一次机会。
那是我第一次重新回到球台。
后来,我每天六点爬起来训练,技术也好。当时我被推荐到江湾少体校,但教练组嫌我个子矮,就没有要我。我从小要强,做什么 … ⌘ Read more
中国制造能否诞生「奢侈品」?
谢邀 @央视新闻 。
什么是从中国制造中诞生的「奢侈品」?
这是一个来自2011年的提问,为了回答这个问题,我们有必要先穿越时空,一起回顾一下,「中国制造」到底走过了怎样的路。
不妨把时间线拉得再长一些。
75年,对于一个人,很长,意味着从垂髫小儿到黄发老者的一生;对于一个国家,很短,不过才经历了一代人一生的岁月。
75年前的中国,能造什么?
在“一辆汽车、一架飞机、一辆坦克,甚至是一辆拖拉机都造不出来”的75年前,他们一定这样幻想过这样的未来——「总有一天」。
那一年的大典上,我们没有一架自己制造的飞机,要成为航空强国就必须先造出飞机的“心脏”——航空发动机。直到1954年8月,新中国第一台活塞式航空发动机——M-11试制成功。
如今,中国研制的“太行”、“玉龙”、AES100等多型发动机为我国各类飞机、直升机提供了强 … ⌘ Read more
独立游戏在中国 插曲:手游版号办理完全攻略
2017.1.14更新了账号承诺书部分和注意事项里一则自朋友反馈的实践经历,以后会继续补充。
最近因为之前那篇关于版号的文章 ( 关于手游审批:大限已过,无号上线的手游都怎么样了? _)_,收到一些私信,也认识了不少新朋友,其中不少对我们帮助很大。感谢的话放在后面,值得一提的是,大多数新朋友们频繁地问我同一个问题,那就是: 版号究竟该怎么办。
我这才意识到之前关于版号新政的问题写得太潦草,毕竟版号问题至今依然是独立开发者或独立开发团队面临的难题之一。
今晚趁着干完活的空当在这儿详细写一下流程,并且会贴出所需的所有资料和撰写范本,希望能帮到大家。
二、如何选择机构先简单复述下版号定义
游戏版号就是“游戏出版备案”。它是由国家新闻出版总署批准的游戏出版运营的批准文号。
找谁办理?
目前有 两种方式 非常适合独立游 … ⌘ Read more
如何评价《黑神话:悟空》这款游戏?它到底好不好玩?
用上亿美元甚至数亿美元的预算,做出值得被提名为年度游戏的杰作,是世界上最好的游戏研发团队才能做到的事情。
乐观估计,这样的团队全世界不超过20个。
更多的游戏团队,就算拿到了上亿甚至数亿美元的预算,也只能做出8分、7分甚至6分的游戏。
即便是这样的团队,也都是经验丰富的行业中坚。
那么,当一个团队的开发预算注定只有8分游戏的规模时,他们要怎么办?
当研发团队的预算,人力,周期都注定不可能做出一款完美的游戏时,他们要如何决断?
这就是《黑神话悟空》的故事。
在黑神话全盛期的争议过后,现在是可以从一个相对客观的角度来讨论这个游戏的设计得失的时候了。这是一个在极其有限的资源之下,关于正确决断和大胆放弃的奇迹。
不止一位熟识的朋友都抨击旗舰总是倾向结论先行而略过大量前提;所以,这次我尝试一下和以往略有区别的,“登高远眺”的写法,争取把所有的判断前提和数据来源都写进来——当然,这会导致文章长度变得更长。
注意:本文长达25000 … ⌘ Read more
next-20240924: linux-next
Version:next-20240924 (linux-next)Released:2024-09-24 ⌘ Read more
GitHub Enterprise Cloud with data residency: How we built the next evolution of GitHub Enterprise using GitHub
How we used GitHub to build GitHub Enterprise Cloud with data residency.
The post GitHub Enterprise Cloud with data residency: How we built the next evolution of GitHub Enterprise using GitHub appeared first on The GitHub Blog. ⌘ Read more
TRUE COURT STORIES ⌘ Read more
Eyewatering supermarket grape prices expected to drop dramatically as local season approaches
If you have baulked at the price of table grapes at the supermarket lately the good news is they could drop by up to $10/kg when the Australian season brings locally-grown produce to the shelves. ⌘ Read more
x86 Embedded Controller with PC/104 Compatibility for Legacy Systems
The VDX3-6757 PC/104 family of low-power x86 embedded controllers meets PC/104 specifications, offering backward compatibility for projects facing end-of-life x86-based controllers. It is suited for applications like data acquisition, industrial automation, process control, and automotive control. Powered by a DM&P Vortex86DX3 1GHz dual-core CPU with 32KB L1 cache and 512KB L2 cache, the VDX3-6757 supports … ⌘ Read more
Trump vs Harris on Computer Tech Policies
How President Donald Trump & Vice President Kamala Harris differ on Net Neutrality, TikTok, AI, Broadband, Internet Censorship, & Section 230. ⌘ Read more
5th Beta of iOS 18.1, MacOS Sequoia 15.1, iPadOS 18.1 with Apple Intelligence, Available for Testing
Apple has released the 5th beta versions of iOS 18.1, macOS Sequoia 15.1, and iPadOS 18.1, with Apple Intelligence support. The Apple Intelligence features that are included with these releases are mostly Writing Tools, summaries, and new Siri features, which allow you to do things like summarize emails, offer Smart Replies in Mail and Mes … ⌘ Read more
Microsoft, Oracle, Amazon & the Nuclear Powered Data Center
Microsoft re-opens Three Mile Island. ⌘ Read more
4x Na krídlach piesní
Festival slovenských populárnych piesní pre deti s názvom Na krídlach piesní organizuje Komorný zbor Musica Viva a jeho vedúca Mariena Stankovićová Kriváková. Včera v SVD v Báčskom Petrovci na tohtoročnom 4. festivale vystúpili mladí speváci z Petrovca, Hložian, Selenče, Kysáča a Starej Pazovy. Na vystúpenie ich pripravovali učitelia hudobnej výchovy Mariena S. Kriváková, Olivera Popadićová, Anna Medveďová a Juraj Súdi. V programe vystúpili Perličky, skupina pôsobiaca pri … ⌘ Read more
Kubernetes governance: the great policy for innovation
Member post by Kyuho Han, SK Telecom Background : Age of collaboration Since the World Economic Forum (WEF) 2021, The great reset of our society through digital transformation has been accelerating. In Korea, digital transformation is accelerating not… ⌘ Read more
Using an AI Assistant to Read Tool Documentation
Explore how to use Docker and LLMs to streamline workflows for command-line tools to enhance the process of reading docs, troubleshooting errors, and running commands. ⌘ Read more
如何评价游戏《废土3》?
现在我的咽喉仍在被滚烫的火药烟灼烧,左手无名指在科罗拉多雪原的苦寒中不知何时起已经枯死失去知觉,复合装甲破碎的纤维在我全身半愈合的伤口中盘根错节。我的战友们为之流血牺牲的科罗拉多平原上,有人视我们为英雄,有人则欲除我们而后快,这似乎和我们刚刚到来时没有太大区别,我们到底给这片土地带来了什么?我的智慧远不足以让我从这段经历中立刻获得启迪,但我手中羔羊的血仍温热,这段故事应当在寒风中的血和泪被和平的歌声冲散前传承下去。
先做一个 前情提要 吧。
在游戏初代,失控AICochise以消灭人类并为自身谋取生存空间为目的,诱发核战摧毁了现代文明。原计划在核大战之后用机器大军接管地球的Cochise母机被毛子一发偏航的核弹破坏了对外通讯能力,于是只得默默蛰伏。废土 … ⌘ Read more
有人可以把《战锤2:全面战争》的背景故事和人物关系讲清楚吗?
既然题主这样问了,应该是想尽快了解战锤全战的背景。我尝试用最概括性的形式介绍一下。
篇幅比较长,感兴趣的朋友可以点个收藏,或者按照目录进行选择性阅读。
有些内容可能会出现记忆偏差,还望发现的朋友能在评论区及时指出。
首先,《战锤:全面战争》取材自英国桌游模型厂商 GamesWorkshop(简称GW) 的IP 《中古战锤》。该IP在2014年左右宣布灭世(就是背景上世界毁灭了,模型不卖了)。《战锤全战》改编自该IP,但背景设定上有所修改。
本文主要计划分为五个部分——【各族势力介绍】、【大背景以及关于ET的说明】、【主要人物传记】、【重要背景人物介绍】、【中古战锤与战锤AOS】 希望能够帮助各位全战玩家快速了解中古战锤的世界观。 持续更新中——! 前言中古战锤,总的来说是一个标准的西方奇幻世界观。除人类、精灵、矮人御三家之外,还有各种各样稀奇古怪的种族。而所有种族则都 … ⌘ Read more
OpenAI o1 self-play RL 技术路线推演
OpenAI的self-play RL新模型o1最近交卷,直接引爆了关于对于self-play的讨论。在数理推理领域获得了傲人的成绩,同时提出了train-time compute和test-time compute两个全新的RL scaling law。作为领域博主,在时效性方面肯定卷不过其他营销号了,所以这次准备了大概一万字的内容,彻底深入分析并推演一遍其中的相关技术细节。
首先要说一下,o1是一个多模态模型,很多人包括 Jim Fan 都忽略了这一点:
因此他继续叫做o,作为omni系列是没有任何疑问的。只不过这次发布是过于低调了,很多人都没有注意到 … ⌘ Read more
Wine grape growers looking to diversify plant agave tequila crops in SA
Some would say South Australia is too cold for the tropical agave plant, but one couple hopes to disprove that theory with the country’s second-ever commercial agave planting. ⌘ Read more
next-20240923: linux-next
Version:next-20240923 (linux-next)Released:2024-09-23 ⌘ Read more
“Call for a Good Time?” OK… ⌘ Read more
MS-CF16 Fanless Low-Power Pico-ITX SBC with Alder Lake-N and Amston Lake Processors
The MS-CF16 is a compact Pico-ITX single-board computer designed for fanless, low-power, high-performance applications in harsh environments. Powered by Intel Alder Lake-N or Amston Lake Series SoCs, the board features a 2.5GbE LAN port, a GbE LAN port, and SATA 3.0 for storage. Unlike the previously covered MS-CF17, this model offers configurable Intel processors, each […] ⌘ Read more
Physics Lab Thermostat
⌘ Read more
@prologic@twtxt.net Thanks for writing that up!
I hope it can remain a living document (or sequence of draft revisions) for a good long time while we figure out how this stuff works in practice.
I am not sure how I feel about all this being done at once, vs. letting conventions arise.
For example, even today I could reply to twt abc1234 with “(#abc1234) Edit: …” and I think all you humans would understand it as an edit to (#abc1234). Maybe eventually it would become a common enough convention that clients would start to support it explicitly.
Similarly we could just start using 11-digit hashes. We should iron out whether it’s sha256 or whatever but there’s no need get all the other stuff right at the same time.
I have similar thoughts about how some users could try out location-based replies in a backward-compatible way (append the replyto: stuff after the legacy (#hash) style).
However I recognize that I’m not the one implementing this stuff, and it’s less work to just have everything determined up front.
Misc comments (I haven’t read the whole thing):
Did you mean to make hashes hexadecimal? You lose 11 bits that way compared to base32. I’d suggest gaining 11 bits with base64 instead.
“Clients MUST preserve the original hash” — do you mean they MUST preserve the original twt?
Thanks for phrasing the bit about deletions so neutrally.
I don’t like the MUST in “Clients MUST follow the chain of reply-to references…”. If someone writes a client as a 40-line shell script that requires the user to piece together the threading themselves, IMO we shouldn’t declare the client non-conforming just because they didn’t get to all the bells and whistles.
Similarly I don’t like the MUST for user agents. For one thing, you might want to fetch a feed without revealing your identty. Also, it raises the bar for a minimal implementation (I’m again thinking again of the 40-line shell script).
For “who follows” lists: why must the long, random tokens be only valid for a limited time? Do you have a scenario in mind where they could leak?
Why can’t feeds be served over HTTP/1.0? Again, thinking about simple software. I recently tried implementing HTTP/1.1 and it wasn’t too bad, but 1.0 would have been slightly simpler.
Why get into the nitty-gritty about caching headers? This seems like generic advice for HTTP servers and clients.
I’m a little sad about other protocols being not recommended.
I don’t know how I feel about including markdown. I don’t mind too much that yarn users emit twts full of markdown, but I’m more of a plain text kind of person. Also it adds to the length. I wonder if putting a separate document would make more sense; that would also help with the length.
Modlitba ako milosrdenstvo
Galéria Matice srbskej v Novom Sade najnovšou expozíciou spojila dvoch maliarov. Na výstave Modlitba ako milosrdenstvo sú spolu predstavení český maliar Alfons Mucha a Predrag Đaković, súčasný srbský umelec, ktorý žije a tvorí v Prahe. Výstava Modlitba ako milosrdenstvo predstavuje dialóg medzi dielami dvoch maliarov z rôznych období: Alfonsa Muchu (* 24. júl 1860, Ivančice – † 14. júl 1939, Praha), ktorý tvoril v druhej polovici 19. storočia a prvej polovice … ⌘ Read more
Last week at The Lunduke Journal (Sep 15 - Sep 21, 2024)
Real-Time Linux! Dystopian AI Future! Fake AI Podcasts! Exploding Pagers! ⌘ Read more
Low-cost Makerdiary board with iMX RT1011 Crossover MCU and Zephyr Support
Makerdiary recently introduced the iMX RT1011 Nano Kit, a compact, high-performance development board featuring NXP’s iMX RT1011 Crossover MCU. With an Arm Cortex-M7 core running at up to 500 MHz, it delivers strong CPU performance and real-time responsiveness The iMX RT1011 Nano Kit includes 128 KB of on-chip RAM, configurable as Tightly Coupled Memory or […] ⌘ Read more
How garden waste from your green bin could help farmers produce better wine
Waste from suburban green bins is being reused as mulch on vineyards in central Victoria. Results have shown more vibrant grape growth and reduced chemical use. ⌘ Read more
Protectli Vault V1410: Fanless 4-Port 2.5GbE Network Appliance with Intel N5105
The Protectli Vault V1410 is a fanless network appliance designed for applications that demand robust performance and reliable connectivity. Key features include four 2.5GbE Ethernet ports and multiple expansion slots, making it a versatile solution for a wide range of networking environments. The device comes equipped with the Intel N5105 processor, a quad-core Celeron chip […] ⌘ Read more
Open-Source Oscilloscope with 1 GS/s High-Speed Data Streaming and Flexible Measurement Capabilities
Crowd Supply recently launched a campaign for ThunderScope, an oscilloscope that combines powerful hardware with open-source software. It captures data at 1 GS/s and streams it to a computer via Thunderbolt, USB4, or PCI Express for real-time processing, offering greater flexibility for complex measurements across various timescales. The Thunde … ⌘ Read more
Návrat perspektívy: Šírka ducha slovenského ľudu
V Archíve Vojvodiny včera otvorili výstavu archívnych materiálov “Návrat perspektívy: Šírka ducha slovenského ľudu”, ktorá prostredníctvom multimediálneho obsahu propaguje kultúrne a historické hodnoty, predovšetkým fotografie a archívne materiály. Projekt bol realizovaný v organizácii Multimediálneho environmentálneho združenia „Alfa Art” a podporili ho Mestská správa pre kultúru Mesta Nový Sad a Archív Vojvodiny. Obsah … ⌘ Read more
Upcoming I-Pi SMARC Embedded Prototype Kit Adopts Intel Amston Lake CPU
The I-Pi SMARC Amston Lake is a prototyping kit built on Intel’s Amston Lake architecture, designed to accelerate embedded system development. Key features include dual 2.5GbE LAN ports with Time-Sensitive Networking support and CAN interfaces for industrial applications. This kit includes the I-Pi SMARC Plus carrier and the LEC-ASL SMARC module, which features an Intel […] ⌘ Read more
ATOMS3R Dev Kit Equipped with 0.85″ color IPS screen and 6-axis IMU
The ATOMS3R development kit is a compact and versatile programmable controller based on the ESP32-S3-PICO-1-N8R8 module. Designed for embedded smart device applications, it combines robust processing power with built-in Wi-Fi, making it effective for a wide range of IoT and motion-sensing projects. The ATOMS3R development kit is built around the ESP32-S3-PICO-1-N8R8 SoC, a dual-core Xtensa […] ⌘ Read more
How to Stop Apple Music from Opening on Mac Randomly
A fair number of Mac users have discovered that the Apple Music application will seemingly spontaneously open itself at random, and even play music, without being prompted to do so. That Apple Music will randomly open itself and even start playing music is highly undesirable behavior for many Mac users, and thus it’s reasonable to … Read More ⌘ Read more
Linux has Real-Time now. What the fart does that actually mean?
20 years in the making. But what does a Real-Time Linux Kernel mean for most of us? ⌘ Read more
Achieving collaboration and impact for end users: introducing the CNCF’s End User Technical Advisory Board (TAB), its mission and initiatives
End user post by Alolita Sharma, Engineering Leader at Apple, CNCF Board & EndUser TAB, OpenTelemetry GC, CNCF Observability TAG Co-Chair The CNCF End User Technical Advisory Group (TAB) was formally announced at KubeCon + CloudNativeCon North America… ⌘ Read more