dbmnt

"throughput is limited by my approval"

This is exactly where I'm at with AI. (Mostly via Claude Code, but I'm not sure the harness, or my workflow in particular, is the important part.)

I am increasingly wondering though, is there really a valid reason to keep blocking on my approval? Most of the time, I wind up saying yes anyway, because the model has a valid, efficient solution.

What if it's faster at this point to just fix the mistakes?

Scary thought, but it seems like we're close, or already there.

As misguided and misinformed as some of it may be, the "anti AI" and "no more datacenters" pushback in the US is actually working to some extent. Multiple states and locales have issued construction/permit moratoriums already. If you think it's a threat, help slow it down.

Another reason is hardware fragmentation. An Android developer has to account for over 20,000 unique devices, compared to an Apple developer targeting around 40 supported models at most.

Apple has a very different business model than Google (Alphabet). It's not even close to the same in terms of how they make most of their money.

No, because the GU in GUID stands for globally unique. Adding it to a second image would cause a detectable collision.

Also, C2PA, another technology mentioned in the article, means tampering is easily detected.

I'm so jaded that my first thought was "oh no, the AI is going to learn terrible customer service."

I hate what corporate IT has become. Hate.

OMG I didn't know about this. It works! Hover over the speaker icon and adjust with the mouse wheel.

Sounds interesting. Tailscale without handing off key management?

I'm working on https://sndlnk.io, a website that allows you to paste in a sharing link from any streaming service and get back a universal link that works with all of them.

It's for those moments where you need to share something and you don't know which steaming service your friend uses. Or maybe for a group chat or social media post where people use different services.

It works with video games too!

I'm also curious if they used Mythos. Google definitely had access to it via Project Glasswing. However, your logic doesn't hold. That sentence only tells us they were using multiple models, but offers no real clues as to which proprietary model(s) they were using.

> I wonder if there could be a large security situation playing out behind the scenes right now.

There absolutely is in the sense that Mythos via Project Glasswing has uncovered over 10,000 critical vulns and counting. I don't know that this incident is directly related, but there's a lot going on at the moment in this area.

Yes but parent was saying use passphrases, which is the same, just more like "correct horse battery staple". Parent then correctly pointed out there are a large number of sites that enforce special characters, numerical digits, etc., also being part of the password. So that idea falls apart very quickly in practice.

But more specifically why build a stack tied to a single vendor?

You talk about competent SRE being hard to find and manage but then you describe needing several AWS backend specialists.

I think I'd rather have a generalized SRE team with portable infrastructure.

Maybe that's just me. I watched an org get burned by Google App Engine. I find these proprietary stacks to be a giant trap.

I built a news aggregator that pulls in hundreds of RSS feeds and uses multiple large language models to synthesize daily briefs. There's also Tech and Finance editions, plus local editions for some US locales. https://feedpunk.com

I'm also building a modern HTTPS-only transport utility called curb. It's an alternative to curl and wget. It's written in Go using only the standard library. curb can stream output or download files and picks the right behavior based on what the server returns and whether the output is going to a human or a pipe. It also has a '--vet' mode that runs the body through security sieves; this is meant to add some protection and friction for the 'curl | sh' use-case. https://gocurb.dev

I hear you on IP law, but how it applies to AI training is far from settled.

I don't believe we should have software patents, and I am highly skeptical of the US copyright system in general.

As for why I didn't use a piracy analogy: humans don't need to pirate books to access them for free. They can just go to the library. That is exactly my point. Reading books isn't a crime. Why would we stop an AI from reading publicly available material just because it's automated and upsets the commercial status quo?

"Amazon's CEO knew what he was doing" is not a fact. That's speculation.

When it comes to highly technical, fast moving developments like frontier AI and blue team / red team perspectives, I could see any CEO getting out over their skis. Now mix in some incompetent Trump admin officials, including apparently Howard Lutnick. I am guessing many of these people don't understand the subject matter very well at all.

Imagine a super intelligent speed reading human in a library. If they read all the books and are able to understand, conceptualize and summarize that knowledge for others, is it theft? The books weren't stolen, after all, just read. The knowledge in the books wasn't taken away; it's still there for others to read.

I personally do not believe knowledge can be stolen.

That's not how CVEs work.

> Open 3 terminal windows. Try to switch back & forth between just two of them with a keyboard shortcut

cmd+` gets me there, no problem at all

> Open a browser and two terminal windows. Try to switch terminal and the browser window, without also bringing the other terminal above the browser window

you got a point there. alt+tab is gonna surface both terminal windows above the browser.

Knoll's Law of Media Accuracy: "Everything you read in the newspapers is absolutely true, except for the rare story of which you happen to have firsthand knowledge."

See also, Gell-Mann Amnesia effect.

Most reporting is garbage once you get into the details.

Gboard hasn't been updated in 4 years and as a result the UI doesn't always display properly. It's especially jarring on iOS 26. It doesn't fit into the OS keyboard target area properly (on my iPhone 17 Pro, at least).

I've tried pretty much every reputable third-party keyboard app in the App Store. Unfortunately, there's really nothing better than the stock one.

I’m struggling with the utility of this logic. The argument seems to be "because malware can intercept /proc output, any tool relying on it is inherently unreliable."

While that’s theoretically true in a security context, it feels like a 'perfect is the enemy of the good' situation. Unless the author is discussing high-stakes incident response on a compromised system, discarding /proc-based tools for debugging and troubleshooting seems like throwing the baby out with the bathwater. If your environment is so compromised that /proc is lying to you, you've likely moved past standard tooling anyway.

I just installed it and it seems really promising. Glad you shared it here.

I’ve spent more time than I care to admit searching for a good keyboard app in the App Store, and I’ve tried a lot of them. This one never surfaced for me in any of my usual searches, which is a shame (likely more on Apple’s search than on you).

I really like the T9-style approach, and I appreciate the clean App Privacy section and straightforward privacy policy.

It's right there in the FAQ.

"Your Affinity V2 license (via Serif) remains valid and Serif will continue to keep activation servers online. But please note that these apps won’t receive future updates."

It's an annual fee. It would raise the cost to $300k/yr.

https://apnews.com/article/h1b-visa-trump-immigration-8d3969...

Ah, ok... a transparent proxy just to hide the origin IP. Thanks for clarifying. A lot of people are assuming full proxying, but I understand you were describing a hypothetical.

This is wrong. It shows a fundamental misunderstanding of how certificate authorities (CAs) work.

A certificate has to be signed by a trusted CA (one your browser already trusts).

A DNS provider could mint a self-signed cert for pornhub.com, but your browser would reject it immediately.

Even if they tried to trick a real CA, Certificate Transparency (CT) would expose the bogus certificate:

https://en.wikipedia.org/wiki/Certificate_Transparency

Instead, NextDNS is very likely abusing the EDNS Client Subnet feature to provide website operators with a spoofed client location. Much more simple and less nefarious.

They're (ab)using the EDNS Client Subnet feature:

https://en.wikipedia.org/wiki/EDNS_Client_Subnet