It's challenging to expand vertically while also fighting a battle horizontally. Usually expanding vertically is done by companies after they've already taken over a lot of their industry or the industry has matured (a.k.a. less competition).
iMessage's attachment renderer is also sandboxed, actually. You might be talking about this vulnerability: https://projectzero.google/2021/12/a-deep-dive-into-nso-zero..., but the second half of the vulnerability exploit chain was finding a way out of the CoreGraphics (Apple's image rendering library) sandbox to hack the rest of the system.
iOS does sandbox their processes really well, that's why this hack was as difficult as it was. The Safari thing you're talking about is JIT - iOS disables all runtime code generation across the system except in Safari (for JavaScript), but this is a completely separate thing from sandboxing.
Server GPU/memory has also been getting more expensive along with normal customer stuff. Hyperscalars with a lot of hardware do try to keep prices constant for a bit (since prices for hardware they bought is amortized), but they too are bound by economic markets and have to raise prices.
No, the model is simulating the entire TUI, not just terminal commands. It also actually is that slow (because it's generating a diff for the entire screen each time you make a keypress).
Macs can install/run iPad/iPhone apps through the App Store if the app developer enables the feature. A bunch of app developers don't though, and iOS apps have DRM that macOS ones don't
But it's specifically a potting epoxy designed to cause damage to the electronics if removed. And any destruction of the epoxy has to not significantly affect the outward appearance of the device. At worst it's defense-in-depth.
I think this is a classic [Commoditize Your Complement](https://news.ycombinator.com/item?id=17047348) - Nvidia wants open models because their business is hardware and it's complement is AI models, so they want AI models to be commoditized so that hardware is the industry with leverage. OpenAI/Anthropic/etc want closed models so that the AI model development/data has leverage over the hardware providers.
This is exactly what the YouTuber was trying to illustrate - the whole legal justification of Flock cameras is that it's equated to someone sitting down in the location and writing down license plate numbers.
Because the scripts run outside the harness, they would have to create their own MCP clients and manually connect to the same servers to run MCP tools. While Pi codemode runs inside the harness itself, so the script can call MCP tools directly and have the call go through the harness.
They can't natively call MCP tools. Claude importing a library, reading the global MCP config, creating a new MCP client, etc. is not the same as Pi codemode being able to do that inline
To be fair to GitHub, outages are pretty short (~40 minutes), and they rarely impact critical parts of the service, nor do they affect one’s ability to work (git offline)
Claude’s scripts can’t call MCP tools, meaning everything has to be CLIs or libraries. At which point you lose the “everything has self-documenting input-output schema” that Codemode is going for.