lemagedurage

Poor quality is the problem and since the bandwidth of editors doesn't magically grow with the amount of input, rejecting AI content is a very effective way to keep average quality up.

It's already typical to call into a native library with Python bindings for performance critical work in Python. Even Pydantic's validation runs in Rust code.

Given that the speedup is modest (~10%) and the Python 3 migration caused enough pain for users, I get it though. If it was a 2x speedup it would be a different story IMO

You can click the big image or the url at the right and use it in browser.

https://blueblazin.github.io/blockviz/

I tried it on some of my code and it's pretty neat, though not fully intuitive to follow more complex code structures for me.

Their UI for viewing diffs is pretty good.

I like what they did with the website UI. I hope they can replace the text with something less AI heavy and more sincere/professional.

Looks like VSCode's editor UI on top of pi.

That's interesting. A go version to iterate on the code and a rust version to run the code might be an optimal split here.

For VSCode: 4x speedup vs TS 7, 3x vs TS 7.1

Thanks for the input, glad to hear both parties benefit.

Could you elaborate on what type of optimizations were added to TS 7.1? As in, were they more algorithmic or related to memory management?

I'm expecting IDEs to adapt to new workflows, albeit very slowly.

There is potential to build a better UI than what's constrained within a CLI's grid of characters.

You could be implicitly paid to understand what you're delivering and reading code could be a way to achieve that.

Agree.

I do feel like there's merit to having an open source implementation of anything, no matter who/what wrote it. I'm just hoping the results are validated well.

Given that you bring the weights from NVIDIA's DLSS. So basically, the repo contains reverse engineered machinery that produces the exact same output given the same model.

We're not simply changing the seed. The LLM provides some probability to tokens, and the SynthID watermarking makes it slightly more likely to choose some and slightly less likely to choose others within that distribution.

Theoretically, that can be true.

In reality, Gemini and Anthropic use SynthID watermarking which affects token probability distribution, i.e. their tournament sampling can pick lower-probability tokens which the LLM's distribution would otherwise not have. They likely use this over unbiased watermarking because SynthID is resistant against text edits.

That's not true. Watermarks are messing with the next token generation probabilities based on some random seed. The quality is neccesarily lower, the difference is simply too small to notice, typically.

Yep, I'd hypothetically economically gladly build an alternative app store charging "only" 5% on app revenue. The blocker would be that Google has a hold on what store is installed by default on Android. Apple even more so. This is hardly a fair market.

Being pedantic about the captcha is the less-human thing to do here.

Agree, and for me I feel like I often have more implicit intentions than I write in a prompt. A plan helps me verify whether an agent gets these right or not. Plus, it highlights tradeoffs I might've not thought about. Removing both feels like lowering a quality bar.

On the other hand, for a low effort hobby project: just do the thing.

Yeah, Tailwind is great as long as you have some way to reuse markup, like React. You don't have to care about class names and whether CSS defined somewhere else affects styles of a current component at all. That last point makes sense for LLMs as well: all the styles that affect the current file are automatically in context since styling is local. Admittedly, there are other ways to achieve this but the ergonomics of Tailwind are just good. Having nice defaults and good training data is an added bonus.

A challenge of releasing tech on a platform of people that are picky about tech.

Sometimes it feels like we're living in an episode of Silicon Valley, the show.

Well I don't think the position of planets when you're born has a large correlation to how your life will go.

On the other hand, how an AI writes will have a big correlation to whether the written text would likely be written by an AI.

The latter is more of a direct relationship.

Maybe A and B are not correlated, and Y and Z are? What pattern are people falling for here?

I just feel like, from the POV of AI companies, that reducing the amount of em dashes they use to "blend in" more and talk more human-like, for the sake of being less detectable, wouldn't be a big priority.

That just proves that perfect classification is impossible. Classification doesn't need to be 100% accurate to be useful.

The 80% accuracy from the article would be one reason to believe there's significant correlation, no?

Why would sounding human be a goal rather than a byproduct of trying to communicate efficiently?

The problem seems to be deeper rooted. Cursor doesn't see cloning a repo with Cursor and code execution as separate security boundaries.

Cursor ships with Workspace Trust disabled by default [0]. A repo that includes .vscode/tasks.json with "runOn": "folderOpen" will already run arbitrary code [1].

[0] https://cursor.com/docs/agent/security#workspace-trust

[1] https://www.oasis.security/blog/cursor-security-flaw

You're right, looks like they changed it, though Task should still work.

> In version 2.1.63, the Task tool was renamed to Agent. Existing Task(...) references in settings and agent definitions still work as aliases.

https://code.claude.com/docs/en/sub-agents