Arnon

Every e-signature tool proves someone clicked. None prove who clicked. I built SignWho — every signature includes a selfie, biometric verification, and a cryptographic proof certificate. Try it free: signwho.com No account needed for signers. Patent pending.

Collectibles (sports cards, Pokémon, coins, watches), eBay Motors parts (try finding a side mirror for a 2003 Camry on Amazon), refurbished electronics, and the long tail of weird stuff nobody else sells. GMV is ~$75B. Mostly 35+ collectors, hobbyists, and small business owners. Quiet but sticky.

I Haven’t thought about eBay in over 10 years. They’re still doing $10B+ in revenue. Serious question: Who is actually using it?

@TFTC21 DocuSign needs Sam Altman's iris scanner to prove you're human. I built SignWho — selfie + biometric on every signature, runs in your browser, no centralized identity database, no Orb. signwho.com

World, Sam Altman's digital identity project, just unveiled World ID 4.0, what the company calls "full-stack proof of human" infrastructure. The partner list: Tinder, Zoom, DocuSign, Shopify, Okta, AWS, and Vercel. Altman opened by saying we're heading to a world where AI generates more content than humans. Pantera Capital says we've already crossed that threshold. World's answer is an iris-scanning device called the Orb that creates a unique cryptographic ID proving you're a real person. 18 million people across 160 countries have already verified. Tinder is rolling out "verified human" badges in the U.S. after a Japan pilot. Zoom built a feature called "Deep Face" that verifies the person on a video call isn't a deepfake. DocuSign is adding proof-of-human checks to digital signatures. Shopify is enabling verified-human commerce. The most significant announcement is AgentKit, infrastructure that lets AI agents carry cryptographic proof they're acting on behalf of a verified human. Okta built an agent delegation system on top of it. The problem World is solving is real. The question is whether a centralized iris-scanning identity layer controlled by the same person whose company helped create the problem is the right answer. Altman is the CEO of OpenAI. He built the flood. Now he's selling the ark.

@TurnerNovak Built the alternative. One founder. Zero employees. And it actually proves who signed — selfie + biometric on every signature. signwho.com

Fun fact: At 2.1 million employees, Docusign is the largest employer in the United States.

@james406 Built one for real — just one founder, zero Slack channels, and it actually proves who signed. signwho.com

i think i messed up i asked Claude Code to "build a Docusign clone with zero errors" i guess it took it too literally, and immediately spawned 3 million employee agents who don't do anything so far all they've done is create a Slack and ask "any update on this?" to each other all day lesson: startups are hard

Worth pausing on this one. The model is exciting, but if you use chat.deepseek.com or the app, your data goes to servers in China — subject to their National Intelligence Law. Italy already banned it outright, and several other countries restricted it on government devices. The beautiful irony though: open weights are the answer here. Run it locally, or through a Western provider like Together, Fireworks, or Groq — same model, no data flowing to China. That’s what “open source” actually buys you.

🚀 DeepSeek-V4 Preview is officially live & open-sourced! Welcome to the era of cost-effective 1M context length. 🔹 DeepSeek-V4-Pro: 1.6T total / 49B active params. Performance rivaling the world's top closed-source models. 🔹 DeepSeek-V4-Flash: 284B total / 13B active params. Your fast, efficient, and economical choice. Try it now at chat.deepseek.com via Expert Mode / Instant Mode. API is updated & available today! 📄 Tech Report: huggingface.co/deepseek-ai/De… 🤗 Open Weights: huggingface.co/collections/de… 1/n

@MaziyarPanahi DistilBERT + safetensors-first is a smart packaging choice. How's the inference speed on A-series chips vs M-series? Curious if the MLX backend handles the smaller models differently than ONNX would.

What's in 1.0.0: BERT, DistilBERT, RoBERTa, XLM-RoBERTa, ELECTRA, DeBERTa-v2/v3 on MLX OpenMedKit Swift package for Xcode Native iOS demo app shipping with the release safetensors-first packaging with automatic fallback De-identify patient data on a phone. Offline. In Hindi, Telugu, Dutch, French, German, and more.

Medical AI models now run on iPhone. No cloud. No API. OpenMed 1.0.0 just shipped. MLX backend for Apple Silicon. Swift package for macOS and iOS. 200+ PII detection models across 8 languages. pip install openmed Open source. Apache 2.0.

Love this. Same philosophy, different vector. We built something similar for the browser — PII detection and redaction that runs 100% client-side with DistilBERT via WebGPU. No app install, works on any desktop. The "no cloud, no API" approach is the only way forward for medical and legal data. USPTO patent issued on the architecture. zeroegress.ai

@matthewmillerai SWE-bench numbers are real but cherry-picked — Gemma4 actually beats Qwen3.6 on VITA-Bench and Tool Decathlon, and they're basically tied on MMLU-Pro. Qwen wins on agentic coding, not across the board. Still an impressive release though.

Qwen 3.6 35B makes Gemma 4 look like a joke. These results are insane for a 35B parameter model.

What Beijing understands that Silicon Valley is slow to admit: The moat in AI was never the model. It was the combination of model + compute + distribution + trust. Open-source strips out the first variable. Alibaba, DeepSeek, Moonshot, and ByteDance are running a coordinated (and state-supported) campaign to commoditize the exact layer that OpenAI and Anthropic built their valuations on. Meta tried this with Llama and wavered. The Chinese labs are not wavering.

⚡ Meet Qwen3.6-35B-A3B：Now Open-Source！🚀🚀 A sparse MoE model, 35B total params, 3B active. Apache 2.0 license. 🔥 Agentic coding on par with models 10x its active size 📷 Strong multimodal perception and reasoning ability 🧠 Multimodal thinking + non-thinking modes Efficient. Powerful. Versatile. Try it now👇 Blog：qwen.ai/blog?id=qwen3.… Qwen Studio：chat.qwen.ai HuggingFace：huggingface.co/Qwen/Qwen3.6-3… ModelScope：modelscope.cn/models/Qwen/Qw… API（‘Qwen3.6-Flash’ on Model Studio）：Coming soon～ Stay tuned

The takes I'm seeing focus on the engineering. That's the wrong story. "Impressive engineering" isn't why Alibaba gave it away for free. Training frontier models costs billions. Alibaba's Qwen API revenue is a rounding error next to OpenAI's. This isn't a product launch. It's a strategic weapon. Every Qwen download weakens the pricing power of US closed-source labs by a fraction. Multiply by millions. Multiply by every enterprise asking "do we really need GPT-5.2 for this?" That's the release strategy. The moat in AI was never the model. It was model + compute + distribution + trust. Open-source strips out the first variable. Alibaba, DeepSeek, Moonshot — a coordinated, state-supported campaign to commoditize exactly the layer OpenAI and Anthropic built their valuations on. Meta tried this with Llama and wavered. The Chinese labs are not wavering. For 80% of enterprise use cases, Qwen3.6 on your own hardware is fine. It doesn't need to beat Claude Opus 4.7. It just needs to be good enough. The American bet is that the remaining 20% — long-horizon agents, complex reasoning — will sustain $500B valuations. Maybe. But US labs are burning billions to hold a lead that shrinks every six months, against competitors allowed to lose money indefinitely as national policy. That's not a fair fight. It's not supposed to be. Qwen3.6 isn't a revolution. It's a signal: the AI value chain is restructuring, and the assumption that US labs will capture most of that value is starting to look like wishful thinking. Worth paying attention to. Are open models displacing closed API calls in your stack yet?

LM Performance：Qwen3.6-35B-A3B outperforms the dense 27B-param Qwen3.5-27B on several key coding benchmarks and dramatically surpasses its direct predecessor Qwen3.5-35B-A3B, especially on agentic coding and reasoning tasks.

@Docusign Congrats on the recognition. Worth noting though — trust in the platform isn't the same as trust in the signature. A DocuSign signature proves someone clicked a button. It doesn't prove who.

🏆 Docusign has just been named Newsweek’s #1 “Most Trustworthy Company” in the Software and Telecommunications industry - 2 years running! Trust is more than a feature within Docusign IAM - it’s the foundation of how we operate. Learn more: ow.ly/Yx4M50VxSKh

@feijiyeye01 @TJ_Research Copilot is one answer — but it locks you into M365. Another option: strip the PII before it reaches any AI. Runs in the browser, nothing leaves the device. Works with ChatGPT, Claude, Gemini, any tool: zeroegress.ai

@TJ_Research 这个问题反直觉 数据安全。 员工把公司内部数据粘贴到ChatGPT里，这是合规噩梦。 第二个选择指向的就是Copilot——因为它跑在企业已有的M365安全边界内，数据不离开Azure租户，满足GDPR、SOC2、HIPAA等合规要求。 竞争对手越活跃，企业越需要”官方答案”，Copilot就是那个官方答案。

微软的末日？工作场合的AI智能体 举个例子，我是一家汽车配件的SALES，我需要把内部的数据（存在EXCEL里）和新产品的描述（存在WORD里）做成DECK（POWERPOINT）拿去做路演。 正常的操作流程是，我下面的分析员打开EXCEL和WORD，把需要东西放入POWERPOINT，发一个PPT文件给我。 有了AI 智能体后，我把所有需求，数据来源，目的告诉他，他完全取代分析员做的事情，过程里EXCEL和WORD, POWERPOINT需要被调用。这就是老黄说的，软件是工具，被调用，不会被取代。 然后不要忽略了效率问题，原本分析员一天做一个DECK，现在AI一天做10个DECK，工具被调用次数指数上升。所以AI智能体是一个在所有软件更上一层的软件。 那么问题来了，微软会不会出自己的AI智能体？答案是一定会，也已经有了。那这时候就有了一个分发（DISTRIBUTION）的问题，谁会选ANTHROPIC？一定是现在已经用了他们产品的公司，会付更多的钱订阅COWORK（等于雇佣了一个效率非常高的分析员）。如何让更多的企业知道ANTHROPIC？需要销售跑业务！B2B的行业不会有网络效应，都是企业销售跑出来的。初创公司除外（他们站在科技最前沿，但只是冰山一角）。而对于微软来说，只需要利用现有的渠道，推出自己最新的COPILOT。 最后，也是最重要的，对大多数的企业来说，还记不记得疫情是ZM的例子？明明ZM是个更好的产品，为什么大企业里不用？安全。即使后面ZM解决了所谓的安全问题，到现在6年了，ZM仍然没有威胁到微软。给AI智能体权限调用你公司WORD, EXCEL然后在云上面跑？牵涉到公司隐私的数据不敢的。 所以从大多数的企业角度看，我认为终局有两个最有可能的方向： 1. 建私云（成本很大，大公司可以），推理在私云上跑 2. EDGE AI，推理不上云（有点类似于员工/AI去公司上班，下班时啥都不会带走，所有公司信息全部停留在公司电脑上） 中间状态可能是未来3-5年的主流：混合架构。敏感数据（财务、法律、HR）走本地或私有云推理，非敏感的工作（市场研究、公开数据分析、邮件起草）走云端LLM 微软不会是末日，反而是AI agent时代的赢家 — 不是因为他们的AI有多强，而是因为他们掌握了企业的数据、信任和分发关系，这就是我一直拿他和苹果做比喻。

@bradlightcap HIPAA-compliant API is a start — but it doesn't protect PII that's already in the document before upload. Built a tool that strips it in the browser first. Nothing leaves the device. Free: zeroegress.ai

introducing openai for healthcare it includes chatgpt for healthcare, as well as models optimized for care providers and workflows both our APIs and chatgpt support HIPAA compliance requirements we're partnering with HCA, boston children's hospital, MSK, stanford health and many more openai.com/index/openai-f…

From Italy's perspective, its dependence on Israel is relatively greater. The bottom line: Israel can absorb the loss of its defense ties with Italy fairly easily — it represents a tiny share of its arms imports. For Italy, the implications are more complex: it loses a significant customer for its arms industry as well as a supplier of advanced technologies that are harder to replace, particularly in the fields of satellites, cyber, and missiles. That said, even for Italy this is not an existential scenario — it has alternative suppliers and customers.

BREAKING: Italy will suspend the automatic renewal of its defence agreement with Israel.

@realEstateTrent Built an alternative that sends signing links directly — no envelopes landing in spam. Plus every signature includes a photo + biometric proof of who signed. Free during beta → signwho.com

Dear Docusign: If I'm wondering why an investor hasn't signed their subscription docs - and I politely nudge them, and after a bunch of back and forth we come to find out their Docusign envelopes go to spam: This causes hassle for us. Please fix, or I'd rather use fax. Thx.

@The_Cyber_News So the problem is real enough for Microsoft to act on it. But that's one platform, one update. I built a small open source tool that strips EXIF in the browser before you send anything, so you don't have to wonder what each app does or doesn't strip. exifree.com

🛡️ Microsoft to Remove EXIF Data for Images Shared on Teams Source: cybersecuritynews.com/microsoft-remo… Microsoft has announced an important update for its Teams platform. As part of the March 2026 feature rollout, Microsoft Teams will now automatically remove EXIF metadata from all images shared across chats and channels. This privacy-by-default measure aims to protect users from unintentionally leaking sensitive location and device information to internal colleagues, external guest partners, or potential threat actors. EXIF data is hidden text embedded inside digital photos. When capturing a picture, the file quietly records highly specific details. #cybersecuritynews #microsoftteams

@XBToshi Nothing to hand over — even with a court order.

E2EE means nothing if your operating system is a snitch. Signal protects the pipe, but iOS logs the endpoint. "Burn after reading" is a joke when the OS saves the receipt for a month. This is exactly why we built Ghost Chat and Dead Drop on walls.rip. No OS notification hooks. Ephemeral browser sessions only. You close the tab, the keys burn, and it’s gone forever.

@briandstone Built an alternative — every signature includes a selfie + biometric proof of who actually signed. Not just a click. signwho.com

Docusign is so insanely overpriced. eSignature is now built into Google Workspace, completely eliminating the need for DocuSign who charges $300 per user per year to send a max of 100 “envelopes” for signature. That is their most basic plan. I’d say it won’t be long before they fold.

@talkingcockatoo @billyjhowell You don’t have to. Strip the PII before it reaches the AI. We built a tool that does this in the browser — nothing leaves your device: zeroegress.ai

@billyjhowell How do you give it you PII and feel ok about it?

5 things I’m going to immediately build with this: 1. Subscription scanner that visualizes all my subscriptions across cards and when they renew. 2. DCA dashboard. Tells me when market dips and recommends when I should dollar cost average into indexes. 3. Master dashboard that shows cash, stocks, crypto, 401k, etc. 4. AR dashboard that tracks invoices and income across bank accounts and stripe accounts 5. Budgeting dashboard that shows where I’m overspending. What bills I’m getting ripped off on, etc.