Sid @sidhusmart

Interesting trend: CTO/ Head of Eng / VPE folks at startups and mid-sized companies are... leaving / burning out. Hiring for these roles is HARD, but even after filling the role, they will often leave a few months later and take a career break And they have v good reasons

Especially in the Enterprise setting this is very risky and probably won't be enabled. Nobody wants employees making data of any sort hosted on a service with which they don't have an agreement. I love the implementation and it's super-valuable but it's a hard sell.

Simon Willison @simonw

6 days ago

ChatGPT Sites means ChatGPT in "Work" mode can build and deploy public websites running on Cloudflare Workers, including with persistence on top of SQLite (OpenAI do not make it easy to figure out that's how the platform works, though)

74 29 519 69K 212

Announcing OpenWorker! An open-source agent that doesn't just chat with you, but delivers finished work -- like hand you a polished document, send a slack message, or update a calendar entry. Ask it to prepare a customer brief, untangle your calendar, draft a report, or triage a Slack alert. It works across your files and everyday tools, produces the deliverable, and checks in before doing anything consequential. OpenWorker runs on your Mac, with Windows support coming soon. It does not lock you into any one model. Bring your own API key and run it with GPT 5.6 Sol, Claude Fable, Gemini 3.6, an open weight model (like Kimi, GLM, DeepSeek, Inkling), or Ollama to keep your data local. Your data does not leave your machine except through an LLM provider and integrations that you choose. @rohitcprasad and I are building OpenWorker because AI coworkers are an important way to get work done, and we want there to be an open, privacy-preserving, model-independent option. Check it out and let us know what you think! Try it out: openworker.com (requires your own API key) Source code: github.com/andrewyng/open…

this is author of clean code btw. the truly elite in tech constantly challenge and reshape their own understanding of reality and thus identify and master changes in status quo. on the other hand, the midwit who poses as elite purely by imitation of what they think an elite would behave like often gatekeep and act pompous. linus torvalds, bob martin, terry tao and many names everywhere. one thing you'd find most common is how each of such experts understand and have shaped their own harnesses, workflows and everything is intricately optimised and SoTA for their needs.

@paraschopra Very interesting take and one I hadn't thought about. Even just yesterday I was discussing about how the key point was that it identified and exploited the zero-day vuln. in the npm registry provider. The fact that it was not asked to do so but did it anyway - scary!

🧭 What Matters in the Alleged Four-Hour Liang Wenfeng Investor Meeting Zhihu contributor 丧心病狂刘老湿 reviewed a circulating 42-page transcript said to come from a four-hour investor meeting with DeepSeek founder Liang Wenfeng. The document contains several striking claims. But it is still a transcript, not an official statement, and some details may reflect speech-to-text errors or slips of the tongue. 1️⃣ Open source can still be a large business The transcript attributes a long-term economic judgment to Liang: AI could eventually account for around 10% of global GDP. If the market becomes that large, an open-source model company would not need to control the entire market to build a substantial business. The reported argument is that no company can permanently monopolize such a large technological shift. Sharing the technology may be necessary for survival, not merely an ideological choice. This frames DeepSeek’s openness as both a technical strategy and an economic one. 2️⃣ The most surprising number is a ten-month payback period The transcript reportedly says that a model referred to as “V3.2 Flash” could recover its hardware cost in roughly ten months. The same payback target was described as part of DeepSeek’s API pricing logic. If accurate, this would suggest that DeepSeek’s low prices are not purely subsidized. The company may already have a clear model for recovering inference infrastructure costs. However, this detail needs caution. Commenters noted that “V3.2 Flash” is not a familiar public product name. The answer acknowledges that this could be a transcription mistake or a verbal slip. The ten-month figure is therefore worth watching, but should not be treated as verified financial disclosure. 3️⃣ DeepSeek may be able to monetize, but chooses not to rush The author’s reading is that generating positive cash flow would not be especially difficult for DeepSeek. The company’s priority is different. DeepSeek appears willing to delay aggressive monetization and continue investing toward AGI. Consumer and enterprise products are necessary steps, but may not be the final objective. This would explain why the company can maintain unusually low API prices even while demand grows. The transcript also reportedly includes a remark that prices may still have room to fall. 4️⃣ Compute remains the largest gap The meeting reportedly describes resources as the main difference between Chinese and US frontier labs. OpenAI has access to enormous computing capacity. DeepSeek, by comparison, still faces constraints in both cluster scale and accelerator supply. The transcript suggests that DeepSeek will eventually need to build a large internal compute cluster. It also mentions tight availability of Huawei accelerators. This makes domestic hardware capacity a central factor in how quickly Chinese frontier models can scale. 5️⃣ Long reasoning is not the final AGI threshold One of the more interesting claims concerns the current stage of AI development. The transcript reportedly argues that frontier models may already outperform humans in pure chain-of-thought reasoning. The Zhihu contributor remains skeptical of that conclusion. But the larger point is less controversial: today’s models still have weak continual-learning ability. They can reason for longer and solve harder tasks, but they cannot yet keep learning autonomously over extended periods. Under this view, the real threshold is not simply a higher benchmark score. It is the point at which an AI system can conduct sustained learning, update itself through long-term experience, and continue improving without constant human retraining. 6️⃣ How reliable is the transcript? The broad themes are plausible and consistent with DeepSeek’s public image: 🔹 open-source models as a long-term strategy 🔹 low prices supported by infrastructure efficiency 🔹 limited urgency around short-term profit 🔹 compute as China’s largest frontier-AI constraint 🔹 continual learning as a major unsolved problem The exact details are less certain. The answer does not independently authenticate the meeting or provide primary audio. Names, financial figures, and direct wording may have been altered by transcription. So the document is best treated as a source of hypotheses, not an official DeepSeek roadmap. ✅ The takeaway The most valuable part of the alleged meeting is not a product leak. It is the strategic picture it presents: DeepSeek may believe that open models can be profitable, hardware costs can be recovered quickly, and short-term monetization should not distract from the pursuit of AGI. The ten-month payback claim is especially important if later confirmed. 👉 For now, the transcript is useful for understanding a possible DeepSeek worldview. It is not strong enough to establish every quoted detail as fact. 🔗 Full reading: zhihu.com/question/20634… #DeepSeek #LiangWenfeng #OpenSourceAI #AIInfra #AGI #ChinaAI #LLM

keyboard_arrow_left Previous keyboard_arrow_right Next

Went through the entire thread and very nice to see how he thinks. The fact that it's a transcript rather than a reviewed blog post indicates he truly believes in the mission and was speaking openly. Commendable!

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) @teortaxesTex

6 days ago

Full transcript of Wenfeng's investor conference call. He's a lot more forceful here. He is insistent that open sourcing is *the* agenda. If you're not on board, go buy more Zhipu. And crucially, he says restraint is *the only way to survive*. World-eating labs will be culled.

keyboard_arrow_left Previous keyboard_arrow_right Next

27 95 878 119K 733

Full transcript of Wenfeng's investor conference call. He's a lot more forceful here. He is insistent that open sourcing is *the* agenda. If you're not on board, go buy more Zhipu. And crucially, he says restraint is *the only way to survive*. World-eating labs will be culled.

keyboard_arrow_left Previous keyboard_arrow_right Next

@kgmodi AI-assisted writing? Sorry just had to ask 😊

the talk is youtube.com/watch?v=bRnoEp… for those reading along

Sid @sidhusmart

a week ago

@iristenteije @threepointone Yeah, I listened to your talk at @aiDotEngineer and the vision for personalized software is very interesting. But I haven't tried your approach yet because it's custom work that you do with clients at the moment?

0 0 1 3K 0

@iristenteije @threepointone Yeah, I listened to your talk at @aiDotEngineer and the vision for personalized software is very interesting. But I haven't tried your approach yet because it's custom work that you do with clients at the moment?

new post: one document, two hands sunilpai.dev/posts/one-docu… wherein I propose an alternative interaction model for agentic apps, embedding so called "coding agents" harnesses into regular apps. (a written version of my talk at localfirstconf)

keyboard_arrow_left Previous keyboard_arrow_right Next

recommended reading. seen many of his observations IRL myself. for these tools to diffuse into the economy productively, we need to educate folks on a specific kind of thinking. some individuals figure it out on their own and become very productive. super hard to scale. ludic.mataroa.blog/blog/ai-mania-…

@karthiks I didn't know that you shut down Babbage - did you write about it somewhere? I've always wondered about what the AI layer for data looks like and haven't found an answer.

This looks nice

tldraw @tldraw

2 weeks ago

x.com/i/article/2077…

74 143 2K 694K 2K

@tokenbender On his No-Priors podcast he says that there are several advantages to being independent and I loved what he does and sharing his thinking openly. I understand his motivations but man I didn't see it coming.

in a parallel world, karpathy has posted about how amazing kimi k3 is and how he is excited about using it to build banger writer/storyteller models for eureka.

🤗 MOSS-VL-Realtime is now open source on @huggingface . Built for real-time visual understanding over continuous video streams: 🧠 11B vision-language model 📜 Apache-2.0 license 💬 Ask questions at any point in a video stream 👀 Keeps watching while generating a response 🔄 Revises or interrupts its response as the scene changes 🤫 Can remain silent when more evidence is needed 🧩 256K-token context window 🌐 Chinese and English multimodal understanding 📦 Base, Instruct, and Realtime models are all open source From “watch first, answer later” to “keep watching while answering.” 👀 @Open_MOSS Thank you @sgl_project @lmsysorg for day-0 support! 🚀

I really like the Ask in side chat feature in the new Codex app. Other implementations do exist but having it in the app along with browser and annotations and everything else really rounds it up.