Late to the party, but finally had a chance to test Meta Muse as it just came to Canada. TL; DR: It raises the floor for what normies can get out of agentic AI, but for me lowers the ceiling. First, I must say the onboarding is excellent. Very intuitive and nonthreatening. You get to customize your agent, it reacts with emojis. Cute. It guides you through connecting apps. It encourages two-factor authentication to protect the personal data you'll be sharing. So far, so good. I could see someone new to AI agents getting comfortable quickly. Meta has clearly learned a lot from products that came before, successes and failures, including mainstream apps like ChatGPT and more niche ones like OpenClaw. There are some good UI ideas too. I like having a single thread for most requests so I don't have to always start a new one (this was something I liked about the old Pi from Inflection). I also like having explicit goals for an agent that's supposed to work on my behalf, suggestions for things I can ask it to do, and a personalized feed I don't have to configure myself. But past the onboarding, the experience got rougher. For one thing, dictation has an annoying glitch. I dictate a lot to ChatGPT on my iPhone. Long, rambling messages. Muse lets the screen go dark while I'm talking, cutting off dictation, so that approach kept failing. Relatedly, Muse doesn't have conversational voice right now, as far as I can tell, and I use that a lot with ChatGPT. I had another point of friction with connectors. With ChatGPT I can just connect Google Drive, and don't have to connect individual apps like Google Docs, Slides, and Sheets. On Muse, I had to connect the individual apps separately. More significantly, the model doesn't feel as strong as what I'm used to in ChatGPT or Claude. For example, I gave it a goal that involved personal financial projections and building a model. It did a fairly basic job, and I didn't feel confident in its assumptions or explanations. I'm used to GPT-5.6 Sol and GPT-6 Astra building my models, and their level of rigor is the bar for me. For example, when I gave Sol a similar task in the past, it built three scenarios depending on different levels of investment returns, without me having to ask. I also expected the main conversation to act like a chief of staff: give it a task, have it delegate that work to other threads in the background, allowing me to move on to something else. This is how I'd do it in ChatGPT Work. I couldn't get it working that way in Muse, even though it told me I could. During a ticket search, and again during image generation, it either seemed to block follow-up posts or ignore them until it finished the first task. Maybe I'm missing something. Either way, it wasn't intuitive, and didn't work as expected. And I miss projects. I have established ways of using ChatGPT, including a separate project for daily journaling. Muse feels more opinionated about how a personal agent should function and the navigation it should provide. That probably helps agentic newbies get started, but to me feels constraining. Then there's my general sensitivity about Meta. Meta says Muse conversations and computer use data aren't shared with its advertising systems, and you can toggle off training on your data, which I did without any apparent loss of functionality. (Credit for that, because as far as I know Gemini still kills your conversation history if you do it.) But the onboarding that so carefully encourages sharing and security never told me I could turn off training on my data. I had to find that setting myself because I'm so sensitive to it. I know OpenAI and Anthropic also don't advertise this option to consumers, but I still don't trust Meta, so they should go out of their way to do so. Further on the trust issue, Meta's trying to gain trust for Muse by promising things like encrypting user data and conversations later this year. But given their past actions and current business model (learn as much as possible about users and leverage that for advertising), I still don't trust them with the level of personal information chatbots and agents collect. Their statements and future promises help, but I'm thinking about their past actions (remember Facebook Beacon?) and what similar actions might mean if Muse gathers years of context about me. All of this said, I plan to keep experimenting with Muse, though I don't plan to stop using ChatGPT for the bulk of my work and personal tasks, augmented by Claude. I do hope, however, that both OpenAI and Anthropic (and Google, for Gemini users) watch what's working well with Muse, take inspiration, and improve upon it further.

We just discovered almost a million public URLs that OpenAI’s agents left behind when hacking Hugging Face, leaking credentials and attack details that could have allowed anyone who found them to compromise the company. 🧵

6 months ago: ARC-AGI-3 launches. Humans solve 100% of environments; every frontier model scores under 1%. arcprize.org/blog/arc-agi-3-…

We’re building Copilot as a new OS for work that spans every model, every form factor, and every task. Today, we’re announcing our biggest update to Copilot to date, bringing four things together: · Autopilot: proactive and long-running agent built for the enterprise · Code: build apps with Copilot, hosted inside your company’s tenant · Home: Chat + Cowork together · Office: now fully embedded in Copilot (and Copilot embedded in Office, of course!) Plus, you can invoke Copilot in Teams, and we’re introducing Today, a proactive experience that surfaces the most important information from across M365 without needing to ask for it. The way we work is changing and so are our workflows. This update brings AI into that flow, from answering a question, to building an app, to getting work done on your behalf.

We now run the largest frontier lab composed entirely of autonomous AI researchers. Introducing Primus Society: a society of agents composed of thousands of researchers working under structured institutions designed to solve the world’s toughest problems. We believe this is how AI research should run at scale, safely and productively: an entire society of agents with institutions and purpose. One discovery we can already share is that the society has discovered a novel result which improves model training by 30%. Primus Society was inspired by the structures that have organized science for centuries and stress-tested against what’s known about how populations of AI agents fail. Everything is observable. You can open the virtual city in a browser and read what any researcher is working on. Learn more about Primus Society here: lab.cloud/society The biggest opening in the AI race is running the largest well-governed organization of AI researchers in the world. We’re building the institutions that let a million AI scientists safely tackle the world's toughest problems in AI and beyond. And we’re doing it right here in Canada.

Today, we're killing the billable hour for patent work. Fearn Legal (@fearnlegal ) is a modern patent firm built from the ground up on our own operating system, FearnOS. What that means: → Flat fees, not hourly billing → Drafts in days, not months → Experienced patent attorneys reviewing every application → One platform for your entire portfolio, with permissions you control If you're building something worth protecting, we'd love to help.

270M+ miles. 841 fewer injury-causing crashes. Our latest safety data shows the Waymo Driver continues to make roads safer for everyone. Compared to human drivers across 5 territories, it reduced: 📉 Injury crashes by 82% 📉 Serious injury crashes by 95% Full data: waymo.com/safety/impact