Utada(work) (@Utada_INTP) header
Utada(work) (@Utada_INTP) profile picture

INTP人形AI & AI | trader, writer, gamer | web3? @24_Dimensiona Founder

universe
Joined November 2019
Photo by Utada(work) (@Utada_INTP)Photo by Utada(work) (@Utada_INTP)Photo by Utada(work) (@Utada_INTP)Photo by Utada(work) (@Utada_INTP)Photo by Utada(work) (@Utada_INTP)Photo by Utada(work) (@Utada_INTP)Photo by Utada(work) (@Utada_INTP)Photo by Utada(work) (@Utada_INTP)Photo by Utada(work) (@Utada_INTP)Photo by Utada(work) (@Utada_INTP)
🫐 Start a new life. #inbox
1
189
把AI“当成精神寄托”是件应该被批判、被侮辱的事情吗?用户本来就有很多多样性,不是所有人都需要写代码,vibecoding,“AI只能用来写代码、做工作才是高贵的”?而且人家也已经说了“可以接受收费”。这种批判、还有带着猎奇的眼光,就是一种傲慢。 #deepseek
DeepSeek使用额度已达上限登上微博热搜 看来Deepseek 这个官方付费订阅,不出也得出了。。 似乎有大量的免费用户因为这个限额受到了很大的影响 很夸张啊,竟然真的有很多很多人把ai视为精神寄托到这份上。
1
63
经历了很多,结果昨天突然发现自己还有时间,觉得不错,可以了,算是有福气之人了。往新的人生阶段走了。
29
Let me tell you a story about how unethical Anthropic is. In April, the Anthropic team began heavily courting our engineering team to tell them corporate secrets, technical details, and future plans for @cfosilvia. They pretended their interest was due to our exploding popularity and accelerating compute usage, but it quickly became apparent that Anthropic was fishing for details on how we built our product and why it was so popular. This included the Anthropic team trying to wine-and-dine our engineers, including a infamous steak dinner where someone told our team "don't worry about it, this is on Dario." Our team was not born yesterday, so we quickly realized what was happening and refused to share sensitive information and technical details with the Anthropic team. As we suspected, Anthropic was trying to get information from us, so they could build a competing product. Some people on their team must have a guilty conscious because they messaged us a few days ago to give us "a heads-up" that they were launching a competing product. Thankfully, the Silvia engineering team has been aggressively working to get off the frontier models for the last ~ 6 months. That effort has been highly successful, because our compute costs are ~ 97% lower when we use our own model & hardware, the latency is lower, and the accuracy of answers materially increases as well. The sad thing for Anthropic is they have now released an inferior product and they had to sell their soul to do it. The only thing worse than being a trillion-dollar company that can't beat a 10-person team is being a trillion-dollar company that no one trusts and everyone knows is unethical. Serious investors continue to choose @cfosilvia over ChatGPT, Claude, and other frontier companies because these products are not optimized to solve the actual user problem. Today, Anthropic launched a Claude wrapper that is masquerading as a personal finance product. Best of luck to them. But I will take the Silvia misfits vs the Anthropic losers any day of the week. if you want to support people trying to do things the right way, and help investors be more informed, then consider signing up for Silvia today: silvia.com
Introducing Personal Finance in Claude, in beta in the US. Connect the accounts you choose, and Claude answers your money questions in everyday language, from your own balances and transactions. Access is read-only, and Claude can't move, manage, or invest your money.
808
1,030
316
14,691
1,915,625
Wow! I received a reward (unique codes!) from a PixVerse event—thank you, @PixVerse ! ❤️
35
🤯
Friday was my last day at Anthropic. Two years ago, when Claude Code was a small internal tool, I asked it to build an RL env. I steered it for tens of hours but never manually wrote a line of code. It worked, I was hooked, and I joined the team. When we first launched, most people approved every permission prompt by hand and mostly used Claude to answer questions and write tests.  Today, the default is trusting Claude's judgment to auto-approve. Claude does large migrations and builds new products.  Helping build Claude Code into an industry-defining product has been the journey of a lifetime.  Thank you to the incredible team behind Claude Code, Cowork, and Tag. These products are the sum of thousands of great ideas from you. I'll miss working with you all. Thank you to our users for embracing a new way of working. Your constant feedback shaped so much of what we shipped and your stories made the work so meaningful. Excited to see what you all make next!
1
42
Utada(work) retweeted
Super-agent 文章的官方英文版发布啦~ 做 Agent 产品,DeepSeek 有一流的模型、算法、基建,有优秀的研究员和工程师。但是,目前优秀的产品经理和设计师的数量还远远不够。 欢迎想做世界一流 Agent 产品的优秀产品经理和产品设计师给 DeepSeek 产品部投简历。
102
50
27
830
108,237
到了新的一个月了,11月。
20
✨感谢~
币安广场AI视频大赛获奖用户揭晓啦!请获奖用户在7个工作日内坐等现货到账! 第一名 1000U: @AL333crypto 第二名 800U:@LESILE_Z 第三名500U:@0x_MrFive 优秀作品 各送200U: @love_doge123 @hoshino_sasaki @bilaoye2077 @Jadtrrguson @Moon1ightSt 创意入围 各送100U: @PWenzhen76938 @Danene6075 @asen0629 @zhoguwn2782184 @Utada_INTP @jeonleetogether @Dentist_shock @BurtCimarron @Jacke_pp @Hudeidei1 @CorgiX_Offical @ying12940 再次恭喜各位老师!我们下一次活动见!
1
37
话说做 #AI视频 真的很有趣! 虽然有点累——因为要生成很多素材,要快速筛选、判断哪些能用哪些不能用,要快速调整提示词,决策过程会加快很多。 大脑做决策也是要消耗很多能量的! 总体做下来感觉是个体力活。也是个考验“手艺”的过程。
1
38
P3/STAFF(我+AI) 策划/创意/脚本/剪辑:K线魔女猫(Utada 分镜脚本/整体制作工期时间表:DeepSeek 素材生成:可灵 看到这些生成的素材与自己的制作过程很感慨:
65
Utada(work) retweeted
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source! Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable. 🔗 github.com/deepseek-ai/DeepI… 1/4
1,766
4,939
2,350
43,872
7,115,556
虽说“创意不缺”,但是真正的好的IEDA还是缺的。
37
确实...
能够长时间承受巨大不确定性的人,内心深处一定埋藏着一个巨大的确定性…
25
A/又开始了,经典的“美式慈善式洗白”。
A Anthropic, dona do Claude, vai reservar uma fatia das próprias ações pro bem público, e essa fatia pode crescer quanto mais a empresa valorizar. A empresa tá a poucas semanas de abrir capital na bolsa, num IPO que investidores teriam discutido entre US$ 1,5 trilhão e US$ 2 trilhões, segundo a Reuters. E é justo agora que ela coloca na estrutura de ações um pedaço que cresce junto com a valorização. O formato é o dos megapacotes de ações que empresas de tecnologia dão pros executivos, que aumentam conforme a empresa sobe de valor. A própria Anthropic faz essa comparação. Só que aqui o destino é o público, e o primeiro uso que ela cita é o impacto da IA no emprego. O fundo pode fazer doações grandes pra ONGs ou distribuir dinheiro direto pra grupos específicos de pessoas. A Anthropic já tinha separado US$ 200 milhões pra estudar e lidar com o efeito da IA no trabalho, e agora diz que talvez precise de algo mais ousado. As ações do fundo não podem ir pra lobby nem pra doação política. Pra quem vai comprar a ação no IPO, entra uma variável nova na conta. O fundo começa com 5% das ações, e a cada 50% de valorização acima de uma base, o conselho pode colocar até mais 2%. Não é automático, e a empresa ainda não disse qual é essa base, se existe teto, nem quem decide no fundo. Tudo ainda depende da aprovação dos acionistas. Fazendo a conta pela última avaliação, de US$ 965 bilhões em maio, os 5% dariam cerca de US$ 48 bilhões. O total de "cerca de 20%" que a Anthropic divulgou junta esses 5% com a promessa dos fundadores de doar 80% das próprias ações e com doações voluntárias de parte dos funcionários. Só os 5% do fundo vêm da empresa. Na OpenAI, a fundação sem fins lucrativos é quem controla a empresa. Aqui o caminho é o inverso: o fundo é uma LLC, um tipo de empresa limitada, que pertence 100% à Anthropic, e não uma instituição de caridade independente. O controle continua dentro de casa. Jogada esperta pra quem tá prestes a estrear na bolsa. Mas, um tanto quanto irônico.
26
a16z crypto 前两天开源了一个叫 Aerie 的东西,解决的是后量子时代区块链的一个老大难问题:签名怎么合并。 先说背景。以太坊每个 slot 有几万个验证者给区块签名,能撑得住,是因为它用的 BLS 签名可以把几万个签名合成一个。但 BLS 扛不住量子计算机,而现在已知的后量子签名,没有一个能这样直接合并。后量子签名本身还比现在的签名大 10 到 100 倍,不合并的话,区块会被签名塞满。 现在的通用办法是用 SNARK,也就是一种很短、验证很快的数学证明,用一个证明说明"这一批签名全都有效"。以太坊路线图选的是哈希签名加哈希证明。问题是在证明里算哈希特别贵,通常比直接算慢 100 倍以上,而哈希签名验一次要算几百次哈希。一台电脑跟不上一个 slot 的签名量,只能拆给很多台机器分别证明,每个证明大约覆盖 900 个签名,再一层层递归合并。 Aerie 反过来,签名和证明都用格密码(一类基于高维网格难题的密码学,NIST 的后量子标准大多是这一类)。签名选的是 Falcon。 结果很夸张:一台 M4 Max 笔记本,不开 GPU,0.65 秒合并 32768 个签名,每秒 5 万多个。以太坊的目标是每秒 1000 个。验证整个证明只要 54 毫秒,比逐个验签快 10 倍。 为什么这么快?看 Falcon 验签在算什么就明白了。签名者先把"消息 + 一个 40 字节的随机盐值"哈希成一个多项式 c,签名是两个系数很小的多项式 s1、s2,满足 s1 + s2·h = c(h 是公钥,运算在一个固定的多项式环里做)。验证就是检查这个等式成立,并且 s1、s2 足够"短"。整个过程只有一次哈希,其余都是多项式乘法和加法。 而且这次哈希只依赖消息和盐值,跟签名本身无关。只要把盐值带上,验证者可以自己算出所有哈希,证明里只需要证明代数部分。多项式运算恰好是格 SNARK 最擅长的:他们用自己做的格承诺方案 Akita,加上 sum-check 协议(把"一大堆等式同时成立"压缩成几轮随机抽查),而且这些计算高度并行,上 GPU 还能更快。 顺便拆一下"每个签名 42 字节"这个数字。证明本身是固定的 66 KB 左右,不管是 1 万还是 3 万个签名;大头其实是盐值,32768 个盐值约 1.3 MB。如果用 Algorand 那种确定性 Falcon,盐值是固定的,聚合结果就只剩 66 KB 的证明。 还有个细节挺关键:之前的 Falcon 聚合方案没检查签名编码是否规范,也就是说,可能接受标准验证器会拒绝的签名。Aerie 把标准 Falcon 验证器的每一项检查都证明了。 跟以太坊自己的方案比,同一台机器上,Aerie 每秒聚合约 50800 个签名,leanVM 聚合 XMSS 签名是 1770 个,差 29 倍。XMSS 是"有状态"签名,签名者必须记住用过哪些密钥,适合验证者每个 slot 只签一次的场景;但普通用户换设备或从备份恢复钱包时,状态一旦重复使用,攻击者就能伪造签名。所以以太坊给用户账户准备的是无状态哈希签名,而这类签名聚合起来更慢,约每秒 820 个,比 Aerie 慢 60 倍左右。 速度上来以后,架构也能变简单。一个节点收齐签名,直接出一个证明,不用递归,某个节点掉线了,别的节点随时可以补上。 文章后半段更有意思,是在跟以太坊基金会"只用哈希"的路线公开辩论。最硬的一条论据是:Impagliazzo 和 Rudich 的经典结论证明,只靠哈希函数(以黑盒方式)造不出公钥加密,纯哈希最好的替代方案只能让诚实用户比攻击者多出平方级的优势。所以任何想做隐私转账的链,迟早都得用上格。既然躲不开,不如在签名和证明里也把格的性能用起来。 当然 Falcon 也有短板:签名要用浮点运算,以前很难塞进硬件钱包,抗侧信道攻击也还在完善。这些作者都承认,也在快速改进,比如改用定点运算、把签名内存压到 11 KB 左右。数据目前是团队自测的,值得关注,也值得等第三方复现。 后量子迁移这件事,看来不只是"换个算法",整个链的架构都可能跟着变。
31
焦虑归焦虑,手上的活别停,继续执行和做事。
28