最新资讯
reddit.com OpenAI 称 Astra 仅满足 10% 安全任务标准 dev.to FreeToken:在自有 GPU 上运行前沿 MoE 模型及其内存需求 dev.to 当 AI Agent 说“完成”时,究竟完成了什么? dev.to 部署 Underdog Bark 3.0 Pro:逐字转写、自定义后端与分块设置 dev.to 为艰难对话构建 AI:五个 Agent、一条安全正则与一位盲评监督者 techcrunch.com Palo Alto Networks 洽谈以 5 亿美元收购 AI Agent 安全初创公司 Clutch theinformation.com 黄仁勋被要求就涉华间谍指控向国会作简报 www.infoq.com Edera 推出 Object Capability Bounded Authority,试图将凭证与 AI Agent 工作负载隔离 wechat 这波杀疯了!旗舰级模型不限量、1个月视频额度免费领 bloomberg.com AI 抛售蔓延至信用市场,软银领跌亚洲美元债 dev.to 如何在专用 GPU 服务器上用 OpenAI 兼容 API 部署 LLM dev.to 审计 153 份 Claude Code 与 Codex 会话记录:User-Agent 与 URL 中的邮箱泄露
多元动态
登录 注册
行业动态 hackernoon.com · @Arshavir Blackwell

创业创始人访谈:YourVoiceCraft创始人Arshavir Blackwell回答

原文 我的收藏

YourVoiceCraft创始人Arshavir Blackwell介绍其产品,通过重训小型私有模型保留用户写作风格,而非模仿前沿模型。强调硬件控制,认为威胁不是OpenAI而是作家接受同质化。

AI推荐理由 创始人访谈,介绍YourVoiceCraft产品,通过小型私有模型保留个人写作风格,有具体技术思路,但为推广性质。

65

正文

新故事 创业公司创始人访谈:YourVoiceCraft创始人阿尔沙维尔·布莱克韦尔作答 作者 阿尔沙维尔·布莱克韦尔 2026年9月7日 TLDR 翻译 ENTRDEESJADAURPLLVBGHTTATH 您的浏览器不支持音频元素。 作者:阿尔沙维尔·布莱克韦尔@arshavirblackwell 作者:阿尔沙维尔·布莱克韦尔|@arshavirblackwellYourVoiceCraft创始人 认知科学家兼AI研究员。YourVoiceCraft创始人——基于您的写作训练、听起来像您的AI。 故事可信度 欢迎来到HackerNoon的写作提示!您想尝试回答其中一些问题吗?模板链接在此。 1. 用2-5个词描述您的公司是什么? 像您一样写作的AI。 2. 为什么现在是贵公司存在的时机? 因为每个人都在用AI写作,一切都开始听起来千篇一律。ChatGPT和Claude很出色,但它们把每个声音都压平成一个流畅、讨喜的调子——没有任何提示词能解决这个问题,因为提示词描述你的风格,却不改变模型本身。作家们一直用同样的话告诉我同样的事:输出感觉不真实。我们已经过了那个拐点,人们知道自己无论喜不喜欢都会用AI。现在的问题是,他们用的时候能不能听起来像自己。 3. 您喜欢团队的哪些方面,为什么你们是解决这个问题的人? 团队就是我——我构建了全部,从模型训练到用户界面。我的资格在于近四十年的AI和认知科学经验。我在1990年代就研究神经网络,那时没人关心,而且我有幸师从该领域的两位先驱:伊丽莎白·贝茨和杰夫·埃尔曼,他们在人类和机器如何处理语言方面做了基础性工作。声音——什么让你的语言成为你的——是我整个职业生涯的研究问题。这个产品就是那个问题变成了软件。 4. 如果您没有创建自己的创业公司,您会做什么? 写关于AI的文章。我出版《黑盒内部》,一个关于大型语言模型内部实际运作的Substack——把这些系统真正做的事与炒作声称它们做的事区分开来。老实说,我可能也还在构建什么东西;我从调优引擎和迭代界面中获得的快乐不亚于研究本身。创业公司只是把所有我最喜欢的问题集中在一个地方的版本。 5. 目前,您如何衡量成功?您的指标是什么? 目前成功以诚实的反馈和激活来衡量:有多少受邀测试者注册、上传他们的语料库、并训练出他们的声音——以及过程中什么让他们困惑或出错。我最喜欢的近期指标:一位测试者在收到邀请的一天内注册并上传了超过13,000字。那种“我立刻把所有写的东西都喂给它”的反应是我正在优化的信号。 6. 用几句话,您向谁提供什么? YourVoiceCraft是一个面向所有在线写作人士的网页应用——Substack作者、博主、营销人员、教练、顾问。我们在您自己的写作上训练一个私有模型——真正的权重再训练,不是提示——这样您就能获得一个用您的声音写作的ChatGPT式体验。因为它读过了您的全部,它还能向您展示您的文体指纹,逐段将任何草稿与您的声音评分,当新作品与您六个月前写的内容矛盾时发出警告,映射您主题覆盖的空白,并在您的风格漂移时提醒您。一切都运行在我们自己的硬件上:您的文字永远不会接触OpenAI、Anthropic或任何第三方,而且是一个固定会员费,没有按字计费。 7. 迄今为止,最让您兴奋的进展是什么? 参与进来的人的质量。在几周安静的、仅限邀请的外联中,我们从零到有活跃的测试者训练声音——包括一位在获得访问权限当天就上传了大量作品的人。我还从专业作家和编辑那里获得了实质性的参与,他们正是这个产品必须赢得的怀疑受众。当一个整个职业生涯都围绕声音的人说“我很着迷”时,那比一个虚荣的注册数字更重要。 8. 您认为明年您的增长会在哪里? Substack优先。那里有现成的严肃作家受众,他们正在积极辩论是否以及如何使用AI——而且产品自带Substack阅读器:粘贴您的存档链接,几分钟后它就抓取了您的作品并获得了您的声音。从那里,营销团队是自然的第二市场,因为品牌声音一致性是同一个问题穿着西装。而对于那些文字不能离开组织的机构,我们在他们控制的基础设施上构建完全私有的、物理隔离的部署。 9. 告诉我们您的第一个付费客户和明年的收入预期。 还没有第一个付费客户——我们处于收入前阶段。模式是固定月度会员费,更高层级用于更多训练声音和更深入的分析工具。没有代币,没有按字计费:我希望人们重写一段话四十次而不看出租车计价器跳动,因为实验的自由是好写作发生的地方。 明年我预计收入来自那些发布频率足够高、声音一致性是实际工作问题而非好奇心的作家——先是Substack作者,然后是营销团队。我宁愿有少数几个真正会想念它的人,也不愿有大量只注册一次的人。 10. 您最大的威胁是什么? 显而易见的答案是前沿实验室决定关心这个利基市场。但诚实的答案是扁平化本身变得正常——一代作家决定听起来像别人是速度的可接受代价。如果人们不再重视自己的声音,就没有保护它的市场。我构建的一切都是一个赌注,赌他们不会停止——赌AI生成的同质化内容充斥互联网越多,一个独特、可验证的 人类声音就越成为作家拥有的最稀缺资产。 这个创业公司创始人访谈模板基于HackerNoon创始人兼CEO大卫·斯穆克的十个创业公司创始人问题。 您想尝试回答其中一些问题吗?模板链接在此。 关于作者 作者:阿尔沙维尔·布莱克韦尔|@arshavirblackwell YourVoiceCraft创始人 认知科学家兼AI研究员。YourVoiceCraft创始人——基于您的写作训练、听起来像您的AI。 阅读我的故事了解更多 主题 ai-and-ml #startup-founder-interview#founders#startup-media-coverage#startup#ai#writing-voice#writing#llms
查看原文(中文正文为 AI 翻译)
New Story Startup Founder Interview: YourVoiceCraft Founder, Arshavir Blackwell Answers by Arshavir Blackwell September 7th, 2026 TLDR Translations ENTRDEESJADAURPLLVBGHTTATH Your browser does not support the audio element. byArshavir Blackwell@arshavirblackwell byArshavir Blackwell|@arshavirblackwellFounder at YourVoiceCraft Cognitive scientist & AI researcher. Founder of YourVoiceCraft — AI trained on your writing that sounds like you. Story's Credibility Welcome to HackerNoon’s Writing Prompts! Would you like to take a stab at answering some of these questions? The link for the template is HERE. 1. What is your company in 2–5 words? AI that writes like you. 2. Why is now the time for your company to exist? Because everyone is using AI to write, and everything is starting to sound the same. ChatGPT and Claude are remarkable, but they flatten every voice into one smooth, agreeable register — and no prompt fixes that, because prompts describe your style without changing what the model is. Writers keep telling me the same thing in the same words: the output doesn’t feel authentic. We’ve passed the inflection point where people know they’ll be using AI whether they like it or not. The question now is whether they get to sound like themselves while doing it. 3. What do you love about your team, and why are you the ones to solve this problem? The team is me — I built the whole thing, from the model training to the UI. What qualifies me is nearly forty years in AI and cognitive science. I was working with neural networks in the 1990s, back when nobody cared, and I was lucky enough to train under two pioneers of the field, Elizabeth Bates and Jeff Elman, who did foundational work on how humans and machines process language. Voice — what makes your language yours — has been my research question my entire career. This product is that question turned into software. 4. If you weren’t building your startup, what would you be doing? Writing about AI. I publish Inside the Black Box, a Substack about what’s actually going on inside large language models — separating what these systems really do from what the hype says they do. Honestly, I’d probably also still be building something; I get as much joy from tuning the engine and iterating on the interface as I do from the research. The startup is just the version where all my favorite problems live in one place. 5. At the moment, how do you measure success? What are your metrics? Right now success is measured in honest feedback and activation: how many invited testers sign up, upload their corpus, and get their voice trained — and what breaks or confuses them along the way. My favorite recent metric: a beta tester signed up and uploaded 13,000+ words within a day of getting his invite. That “I immediately fed it everything I’ve written” reaction is the signal I’m optimizing for. 6. In a few sentences, what do you offer to whom? YourVoiceCraft is a web app for anyone who writes online — Substackers, bloggers, marketers, coaches, consultants. We train a private model on your own writing — actual weight retraining, not prompting — so you get a ChatGPT-style experience that writes in your voice. Because it has read all of you, it can also show you your stylometric fingerprint, score any draft against your voice paragraph by paragraph, flag when a new piece contradicts something you wrote six months ago, map the gaps in your topic coverage, and warn you when your style is drifting. Everything runs on our own hardware: your words never touch OpenAI, Anthropic, or any third party, and it’s one flat membership with no per-word meter. 7. What’s most exciting about your traction to date? The quality of the people leaning in. In a few weeks of quiet, invitation-only outreach, we’ve gone from zero to active beta users training voices — including one who uploaded a large body of work the day he got access. I’m also getting substantive engagement from professional writers and editors who are exactly the skeptical audience this has to win over. When someone whose whole career is voice says “I’m fascinated,” that matters more to me than a vanity signup number. 8. Where do you think your growth will be next year? Substack-first. It’s a built-in audience of serious writers having an active debate about whether and how to use AI — and the product ships with a Substack reader: paste your archive link, and a few minutes later it has crawled your work and acquired your voice. From there, marketing teams are the natural second market, because brand-voice consistency is the same problem wearing a suit. And for organizations whose words can’t leave the building, we build fully private, air-gapped deployments on infrastructure they control. 9. Tell us about your first paying customer and revenue expectations over the next year. No first paying customer yet — we’re pre-revenue. The model is a flat monthly membership, with higher tiers for more trained voices and the deeper analysis tools. No tokens, no per-word meter: I want people to rewrite a paragraph forty times without watching a taxi meter tick, because the freedom to experiment is where the good writing happens. Over the next year I expect revenue to come from writers who publish often enough that voice consistency is a working problem rather than a curiosity — Substackers first, then marketing teams. I’d rather have a small number of people who’d genuinely miss it than a large number who signed up once. 10. What’s your biggest threat? The obvious answer is the frontier labs deciding to care about this niche. But the honest answer is the flattening itself becoming normal — a generation of writers deciding that sounding like everyone else is an acceptable price for speed. If people stop valuing their own voice, there’s no market for protecting it. Everything I’m building is a bet that they won’t stop — that the more AI-generated sameness floods the internet, the more a distinct, verifiable human voice becomes the scarcest asset a writer has. This startup founder interview template is based on HackerNoon Founder & CEO David Smooke’s ten questions for startup founders. Would you like to take a stab at answering some of these questions? The link for the template is HERE. About Author byArshavir Blackwell|@arshavirblackwell Founder at YourVoiceCraft Cognitive scientist & AI researcher. Founder of YourVoiceCraft — AI trained on your writing that sounds like you. Read my storiesLearn More TOPICS ai-and-ml #startup-founder-interview#founders#startup-media-coverage#startup#ai#writing-voice#writing#llms

来源报道 · 1 篇 · 以下链接均前往原始媒体

← 返回精选动态 原文

更多内容