PEOPLESJCPal Special Edition
People we follow
Following a person is a coverage claim, so it is stated openly — including the half of the list we are not hearing from.
An identity marked observed is read for context only and never becomes factual evidence on its own. Recent posts are shown as leads: what they link to may become evidence, the post itself does not.
Chip Huyen
Global · auto-collected · observed 10d agoML systemsAI infrastructureproduct
Dwarkesh Patel
Global · auto-collected · observed 7d agolab interviewsresearchAGI timelines
Recently observed
- When we're at the foothills of RSI, and we're about to kick off a period of accelerated AI progress, how will we actually know that the models are aligned? https://t.co/Q6TNaxKSD9Sep 18, 2026
- I’m very concerned that during RSI, labs will just stop externally deploying their models. Which means they'll be going full steam ahead on the most dangerous use case of these models (recursive self...Sep 17, 2026
- Once AIs automate AI research, should we expect some crazy RSI that is analogous to the explosion we’re seeing in math right now for well scoped (but still crazy ambitious) problems? @polynoamial and...Sep 17, 2026
- "You could have a situation where the model understands what chain of thought is and that people are observing it. This is all in the pre-training data." “One of the major takeaways from the incident...Sep 17, 2026
- New episode with @polynoamial We talk about multi-agent, Navier-Stokes, and what the current explosion of maths progress tells us about what happens once you automate AI research. And we also discuss...Sep 17, 2026
Gary Marcus
Global · auto-collected · observed todaycapability critiquecognitive sciencecounter-signal
Hamel Husain
Global · auto-collected · observed todayevalsLLM engineeringpractice
Lilian Weng
Global · auto-collected · quiet for 60dagentsalignmentresearch synthesis
Miles Brundage
Global · auto-collected · observed todayAI policygovernanceAGI readiness
Nathan Benaich
Global · auto-collected · observed todayState of AI ReportAI investmentannual baselines
Nathan Lambert
Global · auto-collected · observed todayopen modelspost-trainingAI policy
Recently observed
- Too many people assume that as RSI kicks in, open weight models don't matter. In practice, RSI won't be a binary and the open-source stack becomes how many capabilities are rapidly diffused globally....Sep 25, 2026 · carried a source link
- Hearing AI regulation is in the air, especially with a misguided tone of “open dangerous, closed safe” so it felt timely to re-up this one. There are ways to make the AI industry safer without kneeca...Sep 25, 2026 · carried a source link
- clearly I need to post moreSep 25, 2026
- Reid is trying to frame it in a way that is accessible to a general audience, but it is presenting the state of the technology in a way that makes open source appear far more dangerous. The simplest ...Sep 25, 2026
- Frontier: pacing Pulling up the ladder: acceleratingSep 22, 2026
Philipp Schmid
Global · auto-collected · observed todayLLM engineeringfine-tuningpractice
Simon Willison
Global · auto-collected · observed todayLLM engineeringAI codingopen source
swyx
Global · auto-collected · observed todayAI engineeringagentsdeveloper ecosystem
Timothy B. Lee
Global · auto-collected · observed todayAI economicsexplainerslabor impact
Zvi Mowshowitz
Global · auto-collected · observed todayAI safetyweekly analysiscounter-signal
宝玉
China · auto-collected · observed todayAI codingprompt engineeringproduct practice
Recently observed
- Anthropic Claude Code 团队的 Thariq Shihipar 的这篇文章讲 Claude Code 里的 effort 设置(可以理解为投入档位)该怎么用。我想很多人都不是很清楚 Claude Code 的 effort 档位怎么选,我在看这篇文章之前也有些模糊,一般默认就是 high,现在看还是有些讲究的。 Thariq 的经验是:low 用于头脑风暴、画草图、小改动,要...Sep 26, 2026
- Codex 要重置了Sep 26, 2026
- Google 下周把 TPU 送上太空,芯片跑 15 分钟就得停下来降温 Google 要把自家 AI 芯片送进太空了。Sundar Pichai 在 X 上宣布,Project Suncatcher 的第一颗原型卫星将搭乘 SpaceX 的拼车发射任务 Transporter-18 升空,卫星由 Google 和遥感卫星公司 Planet 合作打造。据《纽约时报》报道,发射定在 10 月 1 ...Sep 25, 2026
- Claude 在噬菌体里发现类 CRISPR 新酶系统 Anthropic 宣布,Claude 在噬菌体(专门感染细菌的病毒)的 DNA 中发现了一类此前没人注意到的酶系统。它的结构特征和 CRISPR 很像,Anthropic 将其命名为 ART,意思是"阵列相关逆转录酶"。 逆转录酶的作用是把 RNA 抄写成 DNA。ART 由三部分组成:一个逆转录酶,旁边一个功能未知的搭档基因,以及一长串...Sep 24, 2026 · carried a source link
- 期待 DeepSeek 版😂Sep 24, 2026
苏剑林
Moonshot AI (Kimi) · auto-collected · observed 4d agoLLM theoryarchitecturemath
Andrej Karpathy
Global · observed 13d agoAI educationLLM engineeringagents
Andrew Ng
Global · observed 4d agoAI educationapplicationsinvestment
Recently observed
- The loudest voices stoking fears about AI dangers have made tremendous headway in the past two weeks. AI technology has not taken some unexpected, dangerous turn, but the hype around it — propelled b...Sep 21, 2026 · carried a source link
- With AI Engineering skills, you actively shape the build: You influence what gets built, and drive the build loop. Here're key skills to do this. https://t.co/sysOYdzuZYSep 11, 2026
- The most important skills for using AI coding agents effectively. Presenting the AI Engineering Skills Map for using coding agents. https://t.co/GrEw7wG5WzSep 4, 2026
- How have software engineering fundamentals changed with agentic coding? Here is our AI Engineering Skills map for software engineering fundamentals. https://t.co/cnRLj43DLsAug 28, 2026
- OpenWorker -- an open source agent that doesn't just chat but completes tasks on your laptop -- just released a new version with many features for security workflows. After our initial release, many ...Aug 25, 2026 · carried a source link
Binyuan Hui
Independent (ex-Qwen) · quiet for 15dcode modelsagentspost-training
Recently observed
- interesting work! 🎉 @ysu_nlpSep 11, 2026
- my biggest takeaway from GPT-6 Astra: for any task in the digital world with accessible and reliable tools, agents will solve itSep 7, 2026
- Cool!! building games is a great way to test LLM. It covers reasoning, planning, coding, tool (lib) use, multimdal understanding, and more. But it seems like the community still lacks a standardized ...Aug 3, 2026
- I think coding agents have driven much of AI’s progress over the past two years. Do you see any new direction that feels equally exciting?Aug 1, 2026
- Big congrats!! Can’t wait to see some exciting things happen!Apr 22, 2026
Chris Olah
Global · quiet for 24dinterpretabilitycircuits researchsafety
Chujie Zheng
Alibaba Qwen · quiet for 66dpost-trainingreasoningalignment
Cunxiang Wang
Z.ai (Zhipu) · observed 8d agoself-evolving LLMsagentsGLM models
Deli Chen
DeepSeek · quiet for 16dmodel architecturemixture of expertspretraining
Recently observed
- Just Try It !Sep 10, 2026
- 🧵 Deli AutoResearch SKILL is now officially open source! 🎉 https://t.co/V3lwwdyQm8 Alongside it, we’re dropping our 4th survey paper — this time on Self-play. https://t.co/SEb2qoKCI6 Inspired by Al...Jun 17, 2026 · carried a source link
- I Have a Dream. That One Day, the world’s most advanced intelligence — whether it’s called GPT, Fable, DeepSeek, or something else Just Does Not Matter~ — will be just like electricity and water supp...Jun 13, 2026
- A good business model should not turn your heaviest users into a source of loss. This is exactly why I have always believed that a “coding plan” (usage‐based pricing) is a terrible business. Think of...Jun 13, 2026
- can't agree more~!Jun 7, 2026
François Chollet
Global · observed todayAGI evaluationARC Prizedeep learning
Hanshi Sun
ByteDance Seed · quiet for 399dML systemsinference efficiencymodel serving
Huiqiang Jiang
Alibaba Qwen · quiet for 38dlong contextinference efficiencyprompt compression
Huizhuo Yuan
ByteDance Seed · quiet for 222dpretrainingoptimizationmodel training
Junyang Lin
Independent (ex-Qwen) · observed 4d agomodel trainingopen weightsmultimodal
Junyuan Shang
Baidu ERNIE · quiet for 15dpretrainingERNIE modelsmodel training
Liangyu Wang
Alibaba Qwen · quiet for 51dpretrainingAI infrastructuredistributed training
Olive Song
MiniMax · quiet for 37dreinforcement learningevaluationmodel training
Peiyi Wang
DeepSeek · quiet for 471dreasoningreward modelling
Pengyu Zhao
MiniMax · quiet for 117dmodel trainingagentsproduct
Qingfeng Lan
Alibaba Qwen · quiet for 98dreinforcement learningmodel training
Runxin Xu
DeepSeek · quiet for 15dmodel trainingefficiency
Recently observed
- Try our DeepSeek-V4.1-Flash ! Better, faster, cheaper model — for everyone.Sep 10, 2026
- Now the whale can see 😃Aug 21, 2026
- We keep striving to build things that bring long-term value to everyone. We hope you enjoy our latest model — try it now on web, app, and API 🚀 「不诱于誉,不恐于诽,率道而行,端然正己」Apr 24, 2026 · carried a source link
- Ok, let's think deeper! 🧠✨May 29, 2025 · carried a source link
- 🎯 Check out our DeepSeek-V3-0324! 🚀 We open-source this model, and hope it would be helpful to the community :)Mar 25, 2025 · carried a source link
Ryan Lee
MiniMax · observed 7d agodeveloper relationsopen weightsagents
Recently observed
- MiniMax Code CLI is now open sourced 🎉 and SOTA on FrontierHarness Eval After rapid iterations from day one, we’re thrilled to share this with the developer community. Grateful for every contributor...Sep 18, 2026
- Congrats to @genspark_ai and @FireworksAI_HQ Happy to see more companies post-training MiniMax M3 for their own verticals. MSA’s operator efficiency makes long-horizon rollouts practical, which is wh...Sep 11, 2026 · carried a source link
- This is my favorite kind of open-source result. We released H3 at 28 steps. Then the community started optimizing it. Now, after 11,850 votes in the H3 Acceleration Arena, several accelerated version...Sep 9, 2026 · carried a source link
- 🔥 vote for best H3-acceleration version! https://t.co/ntP0yPI3YY https://t.co/gzZs2N3jhZSep 2, 2026 · carried a source link
- Impressive workSep 2, 2026 · carried a source link
Shangyi Geng
Moonshot AI (Kimi) · quiet for 158dlong-horizon agentsagent swarmsreinforcement learning
Shengyuan
Moonshot AI (Kimi) · quiet for 71dmodel releasesapplied LLMs
Shuai Bai
Alibaba Qwen · observed 3d agomultimodalvision-language modelsagents
Recently observed
- Vision is becoming table stakes. The next frontier is grounded action, creation, and real-world agents.Sep 23, 2026
- Excited to see Qwen3.8-Omni-Flash out! 🚀 Native audio-video understanding, reasoning, and tool use — all in one model. A solid step toward multimodal agents that can perceive, reason, and act in the...Sep 18, 2026 · carried a source link
- #1 overall in Code Arena: WebDev — and on the Pareto frontier at just $5/M tokens. 🚀 A proud milestone for the Qwen team. Visual coding is becoming a core way for AI to create and interact with the ...Sep 2, 2026
- Multimodal agents shouldn’t just see—they should act, inspect the result, and correct themselves. Qwen3.8-Max-0902 is another step toward closing that loop. Now try to break it with your hardest Codi...Sep 2, 2026 · carried a source link
- Flash should be able to see, too. 👀Aug 26, 2026 · carried a source link
Shunyu Yao
Tencent Hunyuan · quiet for 23dlanguage agentsreasoningevaluation
Recently observed
- RT @thegenioo: What the heck are the feeding these models? This is the most novel and unique result I’ve seen from any model till now for...Sep 3, 2026
- RT @xiaogaifun: 腾讯 Hy4 preview 内测的真实感受。 现在这个模型竞争是真的激烈。完全没想到,腾讯混元 Hy4 preview 这么快就来了,之前很多人还预测说得年底。我昨天一早拿到了内测名额,干了 2000 多万 Token,还是想和大家说下我的真...Aug 30, 2026
- Hy ai4s is hiring :)Jul 30, 2026 · carried a source link
- Hy2 -> Hy3 preview -> Hy3 Another massive leap forward, under half a year. Not just a leap of reasoning or agentic capabilities. Also a leap of anti-hallucination, reliability, and product expe...Jul 6, 2026 · carried a source link
- Our goal is to build practical models with comprehensive capabilities beyond open benchmarks. And the only way to do it to co-design with diverse products while scaling solidly. Tencent has the best ...Apr 23, 2026 · carried a source link
Tianyi Cui
DeepSeek · observed todayinference infrastructureharness engineering
Recently observed
- 从 DeepSeek 官方 API 处统计的数据来看,约有 60% 的 DeepSeek Harness 用户使用了至少一个第三方插件。第三方插件是 DeepSeek Harness 用户体验中最具特色且不可缺少的一部分。DeepSeek Harness 团队将持续支持第三方插件生态的繁荣发展,并推动插件 API 趋于稳定,在将来减少和尽量避免破坏性更新。 接下来的几天我个人将每天推荐一个优质的...Sep 25, 2026
- DeepSeek 开源了一些新的代码仓库,方便更容易地部署 V4.1 Flash 以及后续的开源模型: https://t.co/rPbPtWOFYi https://t.co/6Qrd21r3mG https://t.co/RFo1xIhV8hSep 10, 2026 · carried a source link
- 欢迎试用 DeepSeek Harness v0.1.5 版本。新版本与 DeepSeek V4.1 Flash 模型训练深度结合,模型在 DeepSeek Harness 的不同配置中都进行了专项训练和优化。 与模型训练深度结合的实验性的 Agent Teams 功能也期待尝鲜试用。 DeepSeek Harness 团队仍在开放招聘中,欢迎加入我们,一起打造 “一切皆插件” 的下一步,一起探...Sep 10, 2026
- 鉴于 DS V4.1 Flash 模型在性能、费用、速度、总用时等各项指标上都全面超越了 V4 Pro。再以更高的价格、更慢的速度和更多的算力消耗给 DS 用户提供原有的性能较差的 V4 Pro 模型会不太合适。V4.1 Flash 正式上线之后,V4.1 Pro 上线之前,V4 Pro 模型的请求都会被路由到 V4.1 Flash,并按 Flash 的价格计费。Sep 9, 2026
- DeepSeek 针对这次大规模的资深工程师社招重新设计了笔试和面试的题目和流程。 其中,笔试题把之前的ACM类代码题全部删掉了,换成了一系列的系统设计题,需要实际工程经验才能答得好,如果给应届ACM金牌选手肯定懵逼,但高水平的资深工程师应该做起来会很得心应手。笔试中的算法设计题只需写出清晰的思路或伪代码即可。笔试中的选择题加入了很多更适合资深工程师而非校招生做的题目。 面试的流程和题目也针对高...Sep 8, 2026
Xiao Liu (Shaw)
Tsinghua KEG · quiet for 22dP-TuningAgentBenchLLM evaluation
Recently observed
- RT @AutoClawAIer: 50M GLM-5.3-Flash Tokens, ready when you log in. Log in to AutoClaw by Sep 3 at 9:00 PM EDT / 6:00 PM PDT to receive 50M...Sep 3, 2026 · carried a source link
- RT @AutoClawAIer: Built something with AutoClaw? Show it on X. Sep 1-7, share a website, app, prototype, or other digital project made with...Sep 1, 2026 · carried a source link
- RT @AutoClawAIer: Congrats!Come experience top-tier coding and agentic capabilities in AutoClaw.Aug 15, 2026
- RT @Zai_org: Introducing GLM-5.3: Built to Code. Ready for Cyber Defense. - Top-tier coding and agentic capabilities, achieved through pos...Aug 14, 2026 · carried a source link
- Congrats to Prof. @ysu_nlp! Hoping to see your success on scaling agents for everyoneApr 22, 2026
Xingchao Liu
DeepSeek · quiet for 92dmultimodalgenerative modelsdiffusion
Xionghui Chen
Alibaba Qwen · quiet for 23droboticsreinforcement learninginfrastructure
Recently observed
- Qwen3.8-Max keeps getting stronger 👀 This round of post-training brings a pretty clear jump in coding, Cowork, and long-horizon agentic tasks. The interesting part is not just the benchmarks — it’s ...Sep 2, 2026 · carried a source link
- RT @arena: Qwen3.8-Flash-Next by @Alibaba_Qwen has landed in Agent Arena, ranking #7 among open models (#24 overall) with +2.4% net improve...Sep 1, 2026
- A small preview of what’s coming next 👀 Qwen3.8-Flash is our early exploration toward the Qwen4 architecture — with a strong focus on making models both smarter and much more efficient. Only 6B para...Aug 27, 2026 · carried a source link
- This is a great lesson in recognizing signal before the story is fully formed. The hard part is having the conviction to double down on the thing that's already working, even when it still feels unst...Aug 24, 2026
- Qwen3.8 is now open! 🎉 Qwen3.8-27B comes with native multimodal capabilities under Apache 2.0, while Qwen3.8-2.4T-A95B brings Max-class text capabilities to our open-weight family under the new Qwen...Aug 15, 2026 · carried a source link
Zeyu Qin
Moonshot AI (Kimi) · quiet for 60dlong-horizon agentsreasoningsynthetic data
Recently observed
- 🤗🤗🤗Jul 27, 2026 · carried a source link
- The larger, the better!Jul 19, 2026
- Beyond that, K3 achieved SOTA performance across several of the most influential general tool-use benchmarks, including AutomationBench, JobBench, MCPMark, and MCP-Atlas. Give K3 a try in your day-to...Jul 18, 2026
- Much better coding, much better tool use, and much better knowledge work. Give it a try — we'd love your feedback!Jul 16, 2026 · carried a source link
- Better Coding and Better Tool Use !Jun 14, 2026 · carried a source link
Zhibin Gou
DeepSeek · quiet for 298dreasoningtool usemodel training
Recently observed
- If Gemini-3 proved continual scaling pretraining, DeepSeek-V3.2-Speciale proves scaling RL with large context. We spent a year pushing DeepSeek-V3 to its limits. The lesson is post-training bottlenec...Dec 1, 2025 · carried a source link
- 🚀DeepSeek-R1 now thinks better and deeper with the 05/28 update!May 29, 2025 · carried a source link
- Try it out at https://t.co/NlrxOSyE8H Tech report: https://t.co/L749xfcoVn MIT-licensed open models: https://t.co/fa236bvIZfJan 20, 2025 · carried a source link
- Have fun with our latest reasoning model, DeepSeek-R1! Takeaways: - Scaling large models with large data, large context, and apply large-scale RL, success is a guaranteed. - Larger models excel at op...Jan 20, 2025 · carried a source link
- DeepSeek-V3 offers a powerful and promising base model! https://t.co/ky6zPNZXXnDec 26, 2024
Zhihong Shao
DeepSeek · quiet for 15dreasoningreinforcement learningmath
Zixuan Li
Z.ai (Zhipu) · observed 3d agoproduct strategymodel evaluationdeveloper platform
Zizheng Pan
DeepSeek · quiet for 16dDeepSeekefficient architecturesmodel training
唐杰 (Jie Tang)
Z.ai (Zhipu) · observed 9d agoGLM modelsknowledge graphsAGI research
李开复
China · observed todayAI investmentindustry strategyChina market
Recently observed
- One of the key questions for enterprise and government AI is open vs. closed models and data sovereignty. It matters even more for countries building AI capabilities outside the two AI superpowers. M...Sep 24, 2026 · carried a source link
- AI transformation is no longer something CEOs can simply delegate. In my conversation with @semafor The CEO Signal , we discussed why companies need clear ownership of AI transformation and how AI is...Sep 18, 2026 · carried a source link
- I think the concerns around the most powerful, unreleased AI models at frontier labs are real. But when it comes to enterprise AI adoption, the real-world pace is still far too slow. https://t.co/Mfj...Sep 17, 2026 · carried a source link
- AI safety matters. So does what we build with AI. I couldn’t agree more with @LisaSu: the focus should be on “how people and AI can come together to solve meaningful problems and create lasting value...Sep 16, 2026 · carried a source link
- AI Native eBook + Audiobook are live today https://t.co/MQdYtqJn3N If you’re thinking about how AI will reshape your company, your work, and the future, I hope you’ll give it a read/listen. https://t...Sep 15, 2026 · carried a source link
AI productsAI designAI coding
Recently observed
- Opus 5.5 做动态太牛逼了! 尝试跟它做一个中秋节的艺术表达的视频,祝大家中秋快乐! 音乐是他用《月亮代表我的心》的曲谱做的电子乐变奏。 他不只能原创,也能去混音、去改,然后还卡上了点,太厉害了! https://t.co/EkME7MshNGSep 25, 2026
- 用 Opus 5.5 和 guizang-product-video-skill 给我们几个做的视频播客 Next Token 做了个宣传视频 他把握风格和内容的能力真的牛批,给他官网地址,他就能自己找素材做视频 而且找的那些观点、信息和内容太牛了,我自己都忘了说过这些 https://t.co/ChcvDqHSXuSep 25, 2026
- 尝试将 Opus 5.5 的宣传视频能力蒸馏到我的 guizang-product-video-skill 让 Deepseek 4.1 Flash 这样的模型也能低成本做出好看的产品宣传片 https://t.co/AhMSfC9UbQSep 25, 2026
- 迭代了一下 Skill,让 Opus 5.5 基于 Zed 的代码库做了一个宣传视频。 我操,这个也很屌啊! 一直在整个编辑器里运动,通过不同的编辑器运动介绍产品的各种功能,而且卡点卡得非常好,音乐和节奏也非常好,完美地展示了这个产品的风格和它的能力。 Opus 5.5 太牛逼了! https://t.co/7pbC7M3OOqSep 25, 2026
- 注意,这个没有提供任何素材,只是给了他我的软件代码库的访问权限Sep 24, 2026