BREAKING: Conservatives win a landslide majority in British Columbia, returning to power for the first time since 1928.

Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications. On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments. While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software. After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely. openai.com/index/securing-ou…

Something I've seen that's different about how AI works with code and how I work with code is that AI reads documentation about how to do something. It just assumes that everything works, builds out the whole entire workflow or the integration, and then tries to run it and test it. When something breaks, it's not always obvious where it breaks or how it breaks. With smarter models, we've been able to figure out where it's breaking, but it tries to guess where something's breaking and then tries to fix that. Sometimes the fix works, and sometimes it doesn't. It'll show up as a problem somewhere else, and it'll do a hacky fix there and a hacky fix somewhere else. Sometimes a fix that you're doing upstream means you need to rearchitect or change everything downstream. What do you do? This is very, very different from how I work and what I've seen most developers work. I think we verify each step to see whether it works exactly like the documentation says it should work. Usually, this approach means that if step number 3 fails in a subtle way, we will fix that step, and then we'll devise the downstream steps. That usually ends up in a much better-quality integration. I'm not really sure. Maybe this just needs to be a skill, especially when working with external tools, and there could be a skill or something to put in your agents.md. I don't really know, but just something observed.

every major company is arriving at roughly the same offering: persistent memory, email/calendar/messages, browser + computer use, background tasks, proactive notifications, voice, app/tool execution, ambient context, & some notion of a personal agent sitting above everything. remarkable levels of convergence with very little differentiation whatsoever.

Canada pitched 5.7 GW of AI data centers to the world's biggest investors this week Alberta holds 94% of it The province sells the gas, the land and the cold air And the data centre pays for its own power plant

holy shit kimi was faking it’s performance by serving you claude you have been using claude all along