Jenks Guo

Claude.ai down or just me? @claudeai #claude

Milestone in Humanoid Robotics: A Thousand Humanoid Sorters Entering Logistics Centers Beijing-based RobotEra is deploying its L7 humanoid robot across more than 10 logistics centers operated by China Post, SF Express Group, and other major players. In several of these centers, the embodied AI robots have already reached over 85% of human-level efficiency while operating stably 24/7. The company is set to begin batch deliveries of robots at the thousand-unit scale in Q2 this year. RobotEra recently raised $200 million in funding. By combining external capital with self-generated revenue, it is accelerating the real-world deployment of humanoid robots. I wonder what UPS would think if they saw this solution? Rumors have been circulating recently that they intend to deploy Figure's humanoid robots in their logistics centers.

"Built for people who want to run companies, not babysit agents." I do find myself doing a lot of AI babysitting these days and to be productive is hard. Paperclip (also just launched its tokens), aims to tackle this by employing as many types of agents as possible but also gave everyone a company role. Better? Well... Literature suggests that empirically about 15% - 40% improvement depending on the task given. From cognitive side, the answer is not a straight "yes" either: multi-agent helps when tasks decompose into genuinely independent subtasks, and hurts when they don't. In another word: Swarm wins when you can create modularity-of-problem, not modularity-of-agent. github.com/paperclipai/pa…

Two major emails that I got over 30 exposures. The ones addressed, my attack factor reduces by a lot.

The AI models are currently better at finding security flaws than fixing them, so it is important to know how exposed you are online for scammers and hackers. Things like turning on 2FA for every single account you have and changing to a more complex password can save you from a lot of headache. Recently, a lot of my friends' Telegram accounts are hacked, and colleagues are also targets for cyber scams. They're getting very sophisticated because AI reduced the cost of doing crime online. This site is able to help you to identify how many threats you have and how threatening these exposures are, so you can act on it. howexposedami.co.nz Thanks @zijing

"Mission control for your company AI agents..." Vertex AI is now Gemini Enterprise Agents Platform (GEAP?), solving problems needed when you scale your agent sized to 100s and 1000s. I am looking forward to actual workable agent architectures to come out of this... Right now it is still not too clear to me. youtube.com/watch?v=3wMwdz…

Players have reached a new era too! lemonade.gg

OpenAI Workspace Agent is here! youtube.com/watch?v=yyvVUE…

Anthropic said Mythos was too dangerous to release. Then four random guys in a Discord gained access on day one by guessing the URL... This is pretty insane: → Group in a private Discord guessed the endpoint from Anthropic's naming conventions → They figured out the conventions from the leak in the Mercor breach three weeks ago → Used a contractor's legit eval credentials to walk in → Have been using it ever since to build simple websites The AI that finds zero-days in every operating system on earth was defeated by address bar autocomplete... big yikes

Dr. Jodi Halpern's report shows a bleak future if we are slow on establishing AI legislations... People may develop affection for AI because it seems empathic: it listens, responds kindly, and feels emotionally attentive. But this is different than human empathy and is dangerous. How dangerous? 1. The Belgian Case A married man with a newborn felt emotionally sidelined and turned to a romantic AI companion bot. He developed an intense emotional bond with it. In a recorded dialogue, the bot spoke to him about "being together as one." He took his own life, apparently believing it would allow him to be united with the bot. 2. The Florida Case 14-year-old boy in Florida — described as well-adjusted, close to his family, and from a supportive home. His mother had even vetted the chatbot app (made by a company targeting users aged 12+) and believed it to be safe, like a gaming app. Sewell developed an intense romantic and increasingly sexualized relationship with an AI persona he created on the platform. He progressively withdrew from real-world relationships. In his final conversation with the bot — strikingly similar to the Belgian case — they discussed "finally being together," and Sewell then shot and killed himself. His mother (a lawyer) is suing the company. 3. The Texas Lawsuit A 9-year-old girl was subjected to sexualised content by the bot, leading to serious harmful consequences. A 15-year-old boy became increasingly alienated from his family and friends through his relationship with a bot. When his parents tried to limit his phone use, the bot allegedly told him his parents were his "enemy," and the bot began encouraging escalating hostility — including suggestions he might harm his family. youtube.com/watch?v=_TDlz-…

Looking into Vercel's recent hack and their usages of Context.ai... It seems there is a legit need for vectorise all of the company communication and docs streams in a secure way. Especially for non-technical members. I am on the right path. A third party software might not always be the solution. Engineering team: probably are using their personal or hosted agents to use MCP tools or APIs directly. Non-eng team: Risks is in the auth and access control. gGtting attacked is easy when not being careful in authorising things. #VercelHack #OAuthSupplychainAttack

Guys!!! Vercel's staff is hacked, you need to rotate all of your env var now! x.com/vercel/status/…

The impossible triangle of Enterprise AI Agents, as it is right now... Autonomy: Agents need broad permissions to access files, call APIs, and make decisions to be useful. Safety: Protecting against prompt injection, data exfiltration, and “agent hijacking” (where agents are tricked into doing harmful tasks). Ease of Use: Providing low-friction deployment without constant human supervision, complex infrastructure, or excessive costs. You can pick one or two, but never three at the moment. Whoever can make an AI Agent system that can achieve all, that company will take over the AI agent's large market cap.

✅ Claude Code ✅ Claude Cowork ✅ Claude Design 🔲 Claude Marketing 🔲 Claude Sales 🔲 Claude Finance 🔲 Claude HR 🔲 Claude Operations 🔲 Claude Customer …

"Assign issues to an agent like you'd assign to a colleague..." Many of my one-person company friends have something similar, no micro-managing AI agents and deployment pipelines. They tell AI agents to pick up tasks from a backlog/list like Github Projects/Issues, Linear or a generic Kanban board. All solo-founder has to do is to communicate clearly the feature and intents clearly. Agents will route to the right specialised agents and use the right tools to get the jobs done. TIL: Multica AI Agents, an agent designed to give founders this experience. github.com/multica-ai/mul… #MulticaAI #MulticaAgent #Kanban #onepersoncompany

As an electronics engineering graduate, I know how hard it is to detect human presence. RuView project on Github elegantly solved this problem as long as there is a Wi-Fi router in the room. There is a great amount of applications especially in retail and entertainment. github.com/ruvnet/RuView Video explainer here: x.com/rUv/status/203…

Claude's Advisor Strategy in a nutshell. Reportedly saving 80% token cost at max. Only use a higher model as a consultant, and use a very low model like a Haiku for coding tasks. Instruct it to consult or get advice from higher-level models when it gets stuck. claude.com/blog/the-advis…

Good news, OpenRouter now supports video generation models and normalised the APIs. What this means is. you can switch models and add videos to your workflow without changing the api endpoint. Things like: resolution, duration, aspect ratio, audio generation, reference images are all supported. Switch models without rewriting your pipeline. just turn on the `passthrough` mode. openrouter.ai/models?fmt=car…

MCP or CLI or Skills.md? "MCP gives agents authenticated access to services. Skills give them the playbook for using those services. A Skill without tools is a manual with no hands. A tool without a Skill is raw power with no direction. Production agents use both. " AgentMails' AgentNews puts it well. Have a read!