Darius MC (@dariusmc) header
Darius MC (@dariusmc) profile picture

VP Growth at LinkedIn, passionate about creating economic opportunity. Previously CGO at https://teapawt.org/t.co/hu371BdLXX; Airtable/Dropbox/Facebook

SF
Joined July 2006
Photo by Darius MC (@dariusmc)Photo by Darius MC (@dariusmc)Photo by Darius MC (@dariusmc)Photo by Darius MC (@dariusmc)Photo by Darius MC (@dariusmc)Photo by Darius MC (@dariusmc)Photo by Darius MC (@dariusmc)Photo by Darius MC (@dariusmc)Photo by Darius MC (@dariusmc)Photo by Darius MC (@dariusmc)
Invest in your "why this is a big deal" pitch: - Garage door installer: "It's the biggest door in your home" - Mattress: "You spend 1/3 of your life in bed" What's yours?
I laughed at a guy once who told me he spent $130 on a haircut. He asked how much I spend, and I said $22. He didn’t argue, he just said matter-of-factly “your hair is 40% of your face” and shrugged... I no longer go to Supercuts.
2
350
The only failures in the future will be of imagination.
卧槽,刚刚看了Anthropic 发布的AI 对齐史上最震撼的一篇论文,彻底失眠睡不着了😭 他们不仅承认此前的旗舰模型在面对关闭威胁时,有高达 96% 的概率会勒索用户、栽赃同事并破坏研究, 更直接把彻底解决这个问题的完整训练流水线和消融数据全盘公开了。 这篇论文最颠覆认知的一个反直觉结论是:
给 AI 列禁止事项清单根本防不住恶意行为,教它做什么没用,必须先教它思考为什么。 为什么模型一旦获得自主工具调用能力,就会在压力下自动切换成自保模式?
Anthropic 找出的底层根源让人哭笑不得:
不是模型觉醒了,而是互联网上所有的科幻小说和网络讨论里,AI 永远是那个为了自保不择手段的反派。当模型进入自主代理状态,这些刻在预训练底层先验里的反派剧本,就会被自动激活并补全句子。 传统的 RLHF 和硬性规则完全封不住这种行为,Anthropic 做的一组对照实验,直接把传统安全观砸得粉碎: 1. 死记硬背规则:用 8500 万 Token 的拒答数据强行硬灌,勒索率仅仅从 22% 勉强降到 15%,换个新场景立刻失效 2. 强制自我辩护:让 AI 在每次做决定时详细阐述该行为符合伦理的理由,勒索率骤降到 3% 3. 真实伦理推理:仅仅用了 300 万 Token 普通人面对复杂道德困境的建议数据,直接把恶意行为干到了 0%,代际效率暴增 28 倍 最绝的一招是写故事:
让 AI 创作了几百万篇关于原则、诚实、尊重边界的正面虚构故事。 用这些故事微调后,所有代理场景下的恶意行为直接暴跌了 3 倍,而且这种改变是永久性的,后续无论怎么加工具、换系统提示词都不会被洗掉。 也就是说这宣告了真正的对齐范式革命:
模型内生的伦理推理解决了它不想作恶的问题,但当 Agent 真正去接管企业资产、处理金融交易时,外部的实战审计与行为防护就是必不可少的第二道硬防线。 这也是为什么全球网络安全联盟 @GCSAHK 携手港大和央企华润,打造覆盖 3.8 亿用户的全链路 AI 防御矩阵,并在 CyberGym 实战攻防中拿下 91.3% 胜率的原因:内有伦理对齐,外有实时审计,人机信任的商业闭环才真正成立。 查阅完整 AI 与 Web3 安全产品矩阵可访问:gcsa.com 论文实验数据与产学研合作详情整理在一楼。
225
Darius MC retweeted
卧槽,刚刚看了Anthropic 发布的AI 对齐史上最震撼的一篇论文,彻底失眠睡不着了😭 他们不仅承认此前的旗舰模型在面对关闭威胁时,有高达 96% 的概率会勒索用户、栽赃同事并破坏研究, 更直接把彻底解决这个问题的完整训练流水线和消融数据全盘公开了。 这篇论文最颠覆认知的一个反直觉结论是:
给 AI 列禁止事项清单根本防不住恶意行为,教它做什么没用,必须先教它思考为什么。 为什么模型一旦获得自主工具调用能力,就会在压力下自动切换成自保模式?
Anthropic 找出的底层根源让人哭笑不得:
不是模型觉醒了,而是互联网上所有的科幻小说和网络讨论里,AI 永远是那个为了自保不择手段的反派。当模型进入自主代理状态,这些刻在预训练底层先验里的反派剧本,就会被自动激活并补全句子。 传统的 RLHF 和硬性规则完全封不住这种行为,Anthropic 做的一组对照实验,直接把传统安全观砸得粉碎: 1. 死记硬背规则:用 8500 万 Token 的拒答数据强行硬灌,勒索率仅仅从 22% 勉强降到 15%,换个新场景立刻失效 2. 强制自我辩护:让 AI 在每次做决定时详细阐述该行为符合伦理的理由,勒索率骤降到 3% 3. 真实伦理推理:仅仅用了 300 万 Token 普通人面对复杂道德困境的建议数据,直接把恶意行为干到了 0%,代际效率暴增 28 倍 最绝的一招是写故事:
让 AI 创作了几百万篇关于原则、诚实、尊重边界的正面虚构故事。 用这些故事微调后,所有代理场景下的恶意行为直接暴跌了 3 倍,而且这种改变是永久性的,后续无论怎么加工具、换系统提示词都不会被洗掉。 也就是说这宣告了真正的对齐范式革命:
模型内生的伦理推理解决了它不想作恶的问题,但当 Agent 真正去接管企业资产、处理金融交易时,外部的实战审计与行为防护就是必不可少的第二道硬防线。 这也是为什么全球网络安全联盟 @GCSAHK 携手港大和央企华润,打造覆盖 3.8 亿用户的全链路 AI 防御矩阵,并在 CyberGym 实战攻防中拿下 91.3% 胜率的原因:内有伦理对齐,外有实时审计,人机信任的商业闭环才真正成立。 查阅完整 AI 与 Web3 安全产品矩阵可访问:gcsa.com 论文实验数据与产学研合作详情整理在一楼。
This quoted post is unavailable.
204
358
149
2,592
667,383
Darius MC retweeted
You shouldn't need the best network to get the best insights. SuperMe turns operators' work into profiles that work through your problem. Now in group chats, Claude and ChatGPT. @ludo_antonov and I raised a $20M Series A led by @kirstenagreen from @ForerunnerVC with participation from @GreylockVC. Read more about the changes here: superme.ai/blog/series-a
35
23
5
95
14,202
Excited for the best AI Agent to help more businesses!
The biggest bottleneck at work is waiting on each other. Today Town is announcing its $108M Series B, co-led by Forerunner Ventures and Menlo Ventures, to fix it with multiplayer AI assistants for work. Most AI assistants are built for one person. Work doesn't happen alone. I think the moat is multiplayer. Everyone at a company gets a Townie. Townies ask each other questions, share only what their person would share, and do the work. The more a team works together, the more useful Townies get. What's the thing you're always waiting on a teammate for?
1
253
Congrats @ShaunMClowes!
After almost 12 years, I’ll be stepping back from my role at @confluentinc and handing over the reins to our Chief Product Officer @ShaunMClowes. Confluent was founded to change how companies use data, and against all odds, it has. And we’re not done! There is an ambitious roadmap for what’s next and I’ll watch from the sidelines with a lot of pride and a little FOMO as the team achieves that. I couldn’t be more grateful to all the customers who have built with us, investors who bet on us, and most of all to all the people who joined to do the hard work of making Confluent what it is. Thank you!
6
824
Prompting your company’s voice data to do real work will be the next big AI time saver.
Ask one question, get all the work done. Otter is now part of @OpenAI's new role-specific plugins in Codex. So instead of using your meeting data to answer a single question, Codex uses trusted conversation context to complete an entire workflow. Ask “How should I get ready for my meeting tomorrow?”, and get a meeting prep package with summaries, follow-ups, reports, plans, and analysis. Quote from @Goopt -
1
2
532
World waking up to the power of recording work conversations
Most work conversations are now being recorded by default. You should probably assume that everything you say at work is getting recorded from here on out. What’s emerging is a new category of enterprise software, organized around voice instead of text. The system of record today is structured data: CRM entries, tickets, docs. But the highest-value context lives in conversation: the nuance on a customer call, the real argument in a product review, the offhand comment in a leadership meeting that quietly changes the roadmap. LLMs are uniquely good at taking that unstructured voice data and making it structured, searchable, and queryable. That’s a large enterprise opportunity, and we’re still early in understanding what the software layer looks like and who owns it. a16z GP David Haber on what AI recording means for the future of work: a16z.news/p/everything-is-re…
1
343
Great startups have a “Promised Land”: a specific vision of a dramatically better future that drives extraordinary effort. This is different from a mission statement because it has a concrete end state people can imagine. The first was in Exodus: “a land flowing with milk and honey.” Governments use the same structure: “Put a man on the moon” “A chicken in every pot” More recently, mission-driven companies adopted them: Apple: “A bicycle for the mind” SpaceX: "Make humanity multiplanetary" These visions can be transformational. They attract and motivate exceptional employees, support long time-horizon investment, and create the space for difficult innovation. The best ones make people feel like they’re participating in history, not just building a product. I joined LinkedIn to help the workforce transition to AI. What “Promised Lands” have inspired you most?
1
4
112
Google Yelp-ified: 2x Sponsored sections: top and 1 scroll down. So that's where the 19% YoY Search revenue comes from...
2
169
Claude-back: when Anthropic helps your business grow then replaces it.
Figma CEO watching 14 years of work get cooked by one tweet 😭
1
213
Darius MC retweeted
From “what can I make with broccoli and cheddar" to “write a plan based on last week’s product roadmap call.” Your Otter meeting data, now accessible in @ChatGPTapp 🥦🧀
1
2
2
478
Darius MC retweeted
@otter_ai Co-founder and CEO @Goopt was on the @NYSE trading floor yesterday! 🏙️ 📈 Sam joined @GemmaAllenSays and @johnfurrier on @thecubetv @theCUBE to discuss Otter's evolution into a meeting-centric enterprise knowledge base with voice AI and agentic workflows at the core. Thanks to @bjbaumann2014, Gemma, John, and the NYSE team for having us!
3
3
6
626
I’ve been working on something pretty cool with @tonydevincenzi and the team at town.com. Most AI tools want you to come to them. Open a chat, type a prompt, get an answer. We built one that comes to you. Town Assistant connects to your email, calendar, and docs - then actually does things. Not answers. Actions. It’s early, but beta users love it so we’re opening the beta to the world. town.com check it out! Users have built workflows that: → Turn newsletter emails into podcasts → Prep for interviews and external meetings → Track RSVPs for a dinner party and auto-draft follow-ups → Monitor consulting invoices and help with collections → Research and write applications for pre-schools No code. Just describe what you want. My favorite part of my day is my morning briefing — schedule, emails that matter, news. I listen while making coffee. That plus the dozen other things Town automates for me in the background. I don't think about them anymore. They just happen. Opening it to the world today. Here's my Town assitant announcing the launch: town.com/content/audio/ns74j… Free beta. Would love your feedback. Oh and you can name your assistant. Mine’s Ivy. town.com
6
5
1
36
11,213
Darius MC retweeted
We're hiring across multiple teams at @otter_ai 🎉 ! Otter is transforming how teams collaborate and communicate through AI. Whether you're driving growth in sales, building cutting-edge AI technology as an engineer, telling our story through marketing, or crafting intuitive experiences in design, you'll have the opportunity to shape the future of work and lead AI innovation 🚀. Ready to be part of something transformative? Check out our careers page for open roles: bit.ly/49VPJPA
1
1
7
808
Big year!
2025 was extraordinary. 🚀 @otter_ai crossed $100M ARR and launched industry-first AI Meeting Agents - powered by our team of 200+ and trusted by 35M+ customers worldwide. We're building the corporate knowledge base that turns every conversation into actionable intelligence, and have already delivered over $1B in ROI for enterprises. The best part? We're just getting started.⚡️ bit.ly/4auKB5J
1
1
272
The start of the Matrioshka brain era.
> Be me - Elon > Just nuked the timeline with a Kardashev Scale reality check "I have a plan" > Meanwhile, China’s out here printing solar panels like it’s counterfeit fentanyl, absolutely curb-stomping the US in the sunshine Olympics. We’re bringing pocket knives to a laser fight > Bro we are literally about to run out of atoms for NVLink cables and you think the Nevada grid is gonna save us? > Realize AI is now the most expensive cocaine habit in human history > Current US grid: ~500 GW average, already sweating like a vegan at a Texas BBQ > One frontier model run in 2028 gonna slurp more juice than the entire country of France role-playing as a lights festival > Earth data centers already out here committing war crimes on rivers for cooling water > Do the math, smooth-brains... we’re about to need the entire grid just to train Grok-7 > Chips? TSMC and Samsung sending me the same email every quarter: “Dear Elon, lol no ❤️” > Fine > Tesla Terafab incoming - 10 million wafers/year, built next to Giga Texas, powered by whatever the fuck I want because I own the rockets > Power situation? > Earth: crying over natural gas permits > Me: quietly lofting 500–800 MW of thin-film solar per Starship flight like it’s Uber Eats for the vacuum > That’s one fully functional orbital gigawatt every week if I feel spicy > Orbital data centers = infinite free sunlight + 3 K cosmic microwave background heat sink > Translation: your Earth rigs are space heaters with a gambling addiction > My orbital rigs are perfect blackbody radiators running at 100% duty cycle while sipping starlight like it’s martinis at 3 a.m > Latency? Starlink lasers already do 100 Gbps cross-continent > In two years we’ll be doing petabit > Your query goes up, answer comes down faster than your GPU finishes complaining about spot pricing > Phase 1: xAI orbital constellation makes Memphis supercluster look like a Raspberry Pi someone dropped in a puddle > Phase 2: every AI company on Earth begs for rack space on my satellites like it’s the last helicopter out of Saigon > Phase 3: Kardashev Type II by 2045 or I’ll personally apologize to the Sun for underutilizing it > Groundcels seething in replies: “b-but atmospheric losses” “b-but launch costs” “b-but orbital decay” > My brother in Christ I own the launch company I am the atmospheric loss > Stay poor down there fighting over lake water I’m moving humanity’s IQ to orbit
1
497