ghxst

Video by Matt Brown.

I think they released this in 2023 somewhere, it's been hidden behind an experiment in the devtools settings.

The absolute safest way to grab all traffic is capturing a netlog (https://www.chromium.org/developers/design-documents/network...) unfortunately there's been a lot of quirks with network captures through devtools and even with some of the recent bugfixes (https://issues.chromium.org/issues/40254754) they can still be lossy.

Tried a similar approach but found that putting the browser in a VM has a tendency to expose a few data points that stand out as less trust worthy which means you end up getting a lot of captchas on some websites (like using swiftshader for renderer, not having some fonts installed, among other things), lying about these can typically be detected as well (like injecting noise into a canvas, modifying the advertised renderer). If you've found any solutions to these please share.

This is awesome

The hardware in the pictures of the NYT article don't resemble what I am familiar with when it comes to mobile data farming, they look like traditional sim equipment for texting.

I'd be surprised if you had no one around you that didn't have an old android laying around collecting dust. I'm still on an old A52s and it works fine.

I usually hit robots.txt when I want to make fetch requests to a domain from the console without running into CORS or CSP issues. Since it's just a static file, there's no client-side code interfering, which makes it nice for testing. If you're hunting for vulnerabilities it's also worth probing (especially with crawler UAs), since it can leak hidden endpoints or framework-specific paths that devs didn't expect anyone to notice.

In my experience a good aimbot is impossible to tell from a normal player when you play ranked at a high competitive level, at least not with any degree of certainty that I think is worth banning people for. The cheaters you lose to at that level are the ones calling out your positions to their team because they can see you on the minimap while they aren't supposed to, things like that.

The DMA cheaters are caught when riot gets access to the vendors firmware and ban the people that are using it, not by the cheats themselves. Colorbots run on the same PC so these can be caught in various ways.

This article reads like it was written by an LLM and it doesn't mention how these "undetected" DMA cheats are actually caught. The anti-cheat teams join discords of vendors to get access to the cheats and flag users based on the heuristics they observe from the vendors firmware (that DMA card / hardware has to show up as _something_). So yeah, your setup can work (as long as you’re sticking to the drivers and input methods they tolerate), and the same goes for private DMA cheats.

I think the idea of Computational Irreducibility fits here, prediction becomes either as hard or as costly as the process itself.

What editor do you use, and how did you set it up? I've been thinking about trying this with some local models and also with super low-latency ones like Gemini 2.5 Flash Lite. Would love to read more about this.

I’ve tried creating similar solutions and feel LLMs still lack accurate control over (or understanding of) length, stress / accents, and phonetics for consistent name generation. For usernames for example I’ve yet to create a generator that uses LLMs that beats simple Markov chains. Maybe because results are subjective it makes rating / training a lot harder? I like the site and your approach though and great job on lookup speed! If anyone has any tricks or suggestions I'd love to hear them.

"What is an interface?" reminds me of one of my favorite interview questions: "What happens when you open a browser and visit a webpage?". There’s no single right answer and when asked in the right way I find it helps surface knowledge and skills candidates might not otherwise bring up. Some go into OS internals, others talk about networking, some focus on UI and page load performance. Even when candidates expect the question it still reveals how they think and what they spent time reading up on.

I had no idea that Microsoft set up a discord server for copilot.

Love the binary and wave clocks, instantly got me thinking about how it could work as a subtle graphical element in a landing page footer or something like that.

Multiple people within our company reporting issues. Mostly from US, us in the EU still seem fine as of right now.

edit: Never mind, it's down for me now as well.

Thanks for sharing. Sounds like they have (hopefully _had_) a really messy system in place.

And just to be clear, I didn’t mean to downplay what happened to you, I completely understand how serious it is.

Thank you, that's what I intended to say.

I feel like this argument is a byproduct of being relatively well-off in a Western country (apologies if I'm wrong), where access to therapists and mental healthcare is a given rather than a luxury (and even that is arguable).

> programmers that think that real world problems can be solved by an algorithm that's been useful to them.

Are you suggesting programmers aren't solving real-world problems? That's a strange take, considering nearly every service, tool, or system you rely on today is built and maintained by software engineers to some extent. I'm not sure what point you're making or how it challenges what I actually said.

> Haven't you thought about that it might be useful just to you and nothing more? It's the same pattern again and again, first with blockchain and crypto, then nfts, today ai, tomorrow whatever will come.

Haven't you considered how crypto, despite the hype, has played a real and practical role in countries where fiat currencies have collapsed to the point people resort to in-game currencies as a substitute? (https://archive.ph/MCoOP) Just because a technology gets co-opted by hype or bad actors doesn't mean it has no valid use cases.

> Think about it: how nn LLM that by default create a react app for a simple form can be the right thing to use for a therapist?

LLMs are far more capable than you're giving them credit for in that statement, and that example isn't even close to what I was suggesting.

If your takeaway from my original comment was that I want to replace therapists with a code-generating chatbot, then you either didn't read it carefully or willfully misinterpreted it. The point was about accessibility in parts of the world where human therapists are inaccessible, costly, or simply don't exist in meaningful numbers, AI-assisted tools (with a human in the loop wherever possible) may help close the gap. That doesn't require perfection or replacement, just being better than nothing, which is what many people currently have.

I can see how recommending the right books to someone who's struggling might actually help, so in that sense it's not entirely useless or could even help the person get better. But more importantly I don't think most people are suggesting LLMs replace therapists; rather, they're acknowledging that a lot of people simply don't have access to mental healthcare, and LLMs are sometimes the only thing available.

Personally, I'd love to see LLMs become as useful to therapists as they've been for me as a software engineer, boosting productivity, not replacing the human. Therapist-in-the-loop AI might be a practical way to expand access to care while potentially increasing the quality as well (not all therapists are good).

> I considered that an appropriate punishment would have been a pay cut for a few months

This can absolutely cripple a family, I'd be really cautious wishing that upon someone if they wronged you without malice, though I completely understand where you are coming from.

In this case at the very least, I'd want to know what went wrong and what they’re doing to make sure it doesn’t happen again. From a software-engineer’s standpoint, there’s probably a bunch of low-hanging fruit that could have prevented this in the first place.

If all they sent was a (generic) apology letter, I'd have switched banks too.

How did you pursue the matter?

If the training data was "censored" by leaving out certain information, is there any practical way to inject that missing data after the model has already been trained?

AMD just signed a 2 million dollar contract with them to get the new hardware on mlperf, so it looks like the AMD chapter continues. https://x.com/__tinygrad__/status/1935732517933613216

https://abrahamjuliot.github.io/creepjs/ https://github.com/abrahamjuliot/creepjs

Usually my go to. The readme, source code and GitHub issues are great source of information, and the website itself is useful to test against.

edit:

For anything network fingerprinting related, especially censorship related I usually browse https://github.com/net4people/bbs/issues

IMO calling it "easy mode" really misses the mark. If you ever get the chance to hear directly from the engineers working on Starship, I think you’d come away with a deeper appreciation for the scale and complexity of what they’re building. The solutions they work on go far beyond "just" launching a rocket.

I don’t think it’s _that_ silly, there are plenty of cases where comparisons are useful. SpaceX’s development approach is radically different from traditional aerospace, but it’s clearly working for them. A bunch of Falcon 9s blew up too, and now it’s one of the most reliable and frequently used launch platforms in history. Why would you expect Starship to follow a completely different trajectory?

Starship hasn't had a mission yet that I'm aware of. I love the Saturn V but I don't think this is a fair comparison. Just because your software didn't compile first try doesn't mean it's bad. Those two vehicles fundamentally have different approaches to development and that's fine.

Big congratulations to the engineers! Also had no idea that .honda is a TLD that's really cool.