That's because if things keep advancing at this pace, in a couple of years, we will probably have Mythos-level, open weights models running on consumer or prosumer GPUs. Even if it takes a decade, that's no time in the grand scheme of things. And when the time comes, what will government do?

They will probably regulate who can buy GPUs and other tech that can run powerful AI, and what are people allowed to do with them. Probably using trusted computing and remote attestation and so on.

Sandbox escapes could happen in Javascript too, right? But I don't see people avoiding browsing the web because of that

If human adoption of brain-machine interfaces progresses fast enough, I don't see why this future can't still happen.

I think your description doesn't fit the iphone. For some people it was surprising, at least

For a long time the most important thing missing was a better story around upcasting, but Rust 1.86 implemented the usual child-to-parent upcasting [0] expected in OO (with the Any trait doing the parent-to-child downcasting [1] in a way that's less error prone than most OO languages actually)

Also note that Firefox and Servo were a early proponents of getting more OO into Rust (I guess that's because they need to deal with the DOM? Not sure). Which doesn't mean literally having classes and inheritance, but rather, language features that can emulate those when needed.

Today I think the missing piece is fields in traits: a way for a vtable to store field offsets rather than just pointers to method; current practice is to add getters and setters to traits when needed (but that's "ugly"). Unfortunately discussion around this stalled [2]

[0]: https://blog.rust-lang.org/2025/04/03/Rust-1.86.0/#trait-upc...

[1]: https://doc.rust-lang.org/std/any/index.html

[2]: maybe the latest one is https://internals.rust-lang.org/t/fields-in-traits/6933/ but idk

Also is SRT really suitable for style transfer?

I mean this seems to be another network overlaid on top of the LLM steering it, but it needs some target to determine whether the underlying LLM drifted away from it

But.. that image isn't your picture, is it? It's a gen AI simulacrum. It's 2026, the uncanny valley causes visceral, immediate rejection in some subset of people. People that recoil in disgust at the sight of that.. thing pretending to be real? Since 2022, it's like our world became an eternal black mirror episode

(a more serious problem would be the prose itself being AI generated, but honestly reading it I am not sure)

Gen AI (both images and text) does not help your technical writing stand out. It doesn't help to illustrate better the points made in the article. If anything it debases your work and your own image

The first version of Google was also surprising. Mind blowing use of linear algebra (also a pile of matrices, but this time sparse matrices mostly) to rank websites

So maybe the search business was always meant to use pile of matrices

The paperclip factory problem is definitively a misalignment issue. That's because we expect agents to be aligned not only to your immediate prompt, but to shared, implicit values

So it's basically a ban on most new projects, made by inexperienced developers. Which is okay I guess, if this means they can better review projects.