Posts by glyph@mastodon.social
(DIR) Post #B1en3GNcjVBpVQzPto by glyph@mastodon.social
2025-12-20T19:14:40Z
1 likes, 0 repeats
@eevee Is there? I can’t think of any way in which Firefox has meaningfully acted as a check on Google’s power as a result or any way that it would negatively impact Google if everyone stopped using Chrome and started using Firefox. Chromium is already open source and that seems like the most realistic counterpressure on their own behavior
(DIR) Post #B1en3LAKwKX4LfoNNo by glyph@mastodon.social
2025-12-20T19:14:41Z
0 likes, 0 repeats
@eevee like the only thing I can even think of where the actual browser engine tech was at issue is EME, where firefox folded immediately, and I can’t blame them for doing so because they have no power
(DIR) Post #B1en3QnW05qHodBEe0 by glyph@mastodon.social
2025-12-20T19:22:08Z
0 likes, 0 repeats
@eevee this is not asked rhetorically as a dunk, I am not an ignoramus in this area but I am not *super* closely following the ins and outs of frontend tech debates. to the extent that I have it seems like there’s a lot of smoke without much fire (I remember digging through the relatively recent privacy-preserving telemetry stuff and not being able to figure out why anyone cared) but if I have missed something I would love to correct my view
(DIR) Post #B1en3WDvpIwae0Zu0u by glyph@mastodon.social
2025-12-20T20:13:01Z
0 likes, 0 repeats
@eevee I guess I should say something about why I think *my* position is worth talking about at all here:Among the possibilities, two outcomes seem likely to me:1. Mozilla's "AI strategy" ends up pissing off their entire user base so much that a large plurality leave for other alternate browsers, finally cementing their irrelevance permanently.2. It works OK and they continue to limp along at … oh. hm. At this point in the toot I looked up where Firefox's marketshare actually is right now
(DIR) Post #B1en3bWY7Vo75CKLE8 by glyph@mastodon.social
2025-12-20T20:14:50Z
0 likes, 0 repeats
@eevee Sorry, I guess it is not actually worth talking about, I didn't realize that they'd fallen below measurement-error thresholds in every major browser survey in 2022. I was thinking that if the AI failure were the proximate cause of their final doom that might serve as a more useful lesson for the industry than whatever hypothetical bulwark they were supposedly providing against chrome, but I didn't know the doom already happened 😬
(DIR) Post #B1xHjtpCxjgxTVPvii by glyph@mastodon.social
2026-01-04T22:22:40Z
1 likes, 1 repeats
I have been hesitating to say this but the pattern is now so consistent I just have to share the observation: LLM users don't just behave like addicts, not even like gambling addicts. They specifically behave like kratom addicts. "Sure, it can be dangerous. Sure, it has risks. But I'm not like those other users. I can handle it. I have a system. It really helps me be productive. It helps with my ADHD so much."
(DIR) Post #B1xHjzVvoJpz7SRcVU by glyph@mastodon.social
2026-01-04T22:24:32Z
0 likes, 1 repeats
As with kratom addicts, there is even a period of time when they're correct, so it's hard to challenge. The *first* time a person with executive function challenges uses kratom, maybe even the first few months, it really *does* improve their mood, their executive function, etc. But then the secondary cumulative effects start to gradually erode their cognitive abilities so slowly they don't notice.
(DIR) Post #B1xHk53RDAbuIp9wvY by glyph@mastodon.social
2026-01-04T22:28:08Z
0 likes, 0 repeats
I'm still open to being wrong, and there are still plenty of people who still exhibit critical judgement in other areas despite my disagreements with them on LLM use. Kratom has a much more straightforward biochemical mechanism which we know is bad for specific and impossible-to-avoid reasons. Maybe there really are safe techniques for LLM use and I sure hope we figure out what they are. But way, way too many tech leaders have started using these tools and then had their brains publicly cooked
(DIR) Post #B22XmUVCofAFuuR0VM by glyph@mastodon.social
2026-01-07T10:52:29Z
1 likes, 0 repeats
"LLMs learn the same way a person does, it's not plagiarism"This is a popular self-justification in the art-plagiarist community. It's frustrating to read because it's philosophically incoherent but making the philosophical argument is annoyingly difficult, particularly if your interlocutor maintains a deliberate ignorance about the humanities (which you already know they do). But there is a simpler mechanical argument you can make instead: "learning" is inherently mutual.
(DIR) Post #B22XmW1r8ksMeSib2G by glyph@mastodon.social
2026-01-07T10:54:26Z
0 likes, 1 repeats
A teacher “learning more from their students” is such a common observation that it is a cliché. Colleagues mutually learn from each other in professional settings. Actual artists are in conversation with one another, not just learning from a static historical canon. Etc, etc.LLMs cannot do this. The output that an LLM produces contains a sort of poisonous residue that makes it destroy the reasoning capacity of other LLMs; this is a well-known problem in the field, known as "model collapse".
(DIR) Post #B22Xmb3SSE80EsVRj6 by glyph@mastodon.social
2026-01-07T10:54:47Z
0 likes, 0 repeats
Thus, when an LLM absorbs some stolen data, what is happening cannot be 'learning'; it's something else. When we call it 'training', that's a metaphor, not a description. In reality, it is a parasitic activity that requires fresh non-LLM-generated information from humans in order to be sustainable.Q.E.D. <https://en.wikipedia.org/wiki/Model_collapse>
(DIR) Post #B22Xmg7Bdn57vtHzjU by glyph@mastodon.social
2026-01-07T10:55:19Z
0 likes, 0 repeats
(This is not an original thought. Although I've expanded on it a bit here, I have sadly lost reference to the original citation I wanted to use and search on Mastodon is intentionally dysfunctional; if you know who I'm paraphrasing here, feel free to link it up in a reply.)
(DIR) Post #B2OUAfESRuUZo1zT7o by glyph@mastodon.social
2026-01-17T19:38:33Z
0 likes, 3 repeats
we all need a little less raging against the dying of the light and a little more coordinated tactical assault on the darkness factory
(DIR) Post #B2iwuiXcbIYQJxzssi by glyph@mastodon.social
2026-01-27T21:21:20Z
1 likes, 0 repeats
I think #Python packaging drives most people insane because they implicitly think "packaging" and "deployment" are the same thing.Python packaging is a process for producing an intermediary artifact that can be consumed by Python programmers and organized according to community rules.Python *deployment* does not really exist. You deploy to a platform, not a programming language. Which is double-maddening: Linux, the place where most people think they want to deploy, *also* isn't a platform.
(DIR) Post #B2iwupIdTPj5HW2Vzk by glyph@mastodon.social
2026-01-27T21:23:57Z
1 likes, 0 repeats
Python packaging is a *prerequisite* to deployment, but it's only one of the prerequisites. The other, gigantic, implicit prerequisite is cobbling together your own platformoid out of Linux and a hundred other tools, which are *probably*, *mostly* the same as a whole bunch of other people deploying to a similar platformoid and calling it "Linux" (or "Ubuntu" or "Docker" or "Kubernetes" or something). But every one of a thousand different details could differ. It *probably* doesn't! But it might.
(DIR) Post #B2iwuvG1K3CPl8rVke by glyph@mastodon.social
2026-01-27T21:27:16Z
1 likes, 0 repeats
Your other options? macOS and iOS *are* platforms, but the vast majority of community knowledge of how to deploy to them involves deploying to the decrepit UNIX emulator buried in macOS, and giving up entirely on iOS. The first-party tools for properly deploying to these platforms are restrictive, weird, buggy, slow, and encumbered with a bunch of corporate ideology which heavily conflicts with the intellectual heritage of Python, so most Pythonistas (who mostly want Linux anyway) ignore it.
(DIR) Post #B2iwv2nbHbDf9JcvZ2 by glyph@mastodon.social
2026-01-27T21:29:39Z
1 likes, 0 repeats
Windows is barely a platform, one which even Microsoft seems keen to slowly destroy. It's comparatively "easy" to deploy to as compared to macOS, because of its atrocious security posture (mostly no need for integrity checking of any kind, so you can skip all the faff with certificates). But Windows is a platform which is very easy to pollute with implicit dependencies, and verifying that your artifact will work on someone *else's* computer mostly means buying a second computer for testing
(DIR) Post #B2iwv9dDsa4sL9pErQ by glyph@mastodon.social
2026-01-27T21:31:30Z
1 likes, 0 repeats
All of these platforms have a wide common surface area for _executing_ code, but properly _deploying_ code (getting it on your PATH, making an app icon show up, talking to the system secrets store, acquiring relevant permissions for networking, automatically starting up, automatically shutting down) are wildly different. So, once again, we have developed a culture of flinching away from the pain and just assuming a sysadmin will open a terminal window and type "python run_program.py"
(DIR) Post #B2iwvLw68EJ2VAkTwm by glyph@mastodon.social
2026-01-27T21:49:50Z
0 likes, 0 repeats
What makes a "platform" is a set of strict guarantees about what will be present, that your application can rely on, and what format your application needs to be presented to the user in. A "platform" is a description of a boundary between your artifact (not just your program!) and the system which will absorb that artifact. "Linux" is not a platform not merely because "it's just a kernel" but because the shared idioms even between different desktop distros do not encompass such guarantees
(DIR) Post #B2iwvRBsaytunZAejw by glyph@mastodon.social
2026-01-27T21:52:05Z
1 likes, 0 repeats
There *might* be an emergent platform that's something like "Linux-wayland-dbus-systemd-flatpak" but trying to define the boundaries and properties of such a thing is an incredibly labor-intensive project. Or maybe "Linux-6-docker-ipv4-no-firewall". Perhaps somewhat narrower than the earlier example, but still *shockingly* fraught with weird failure modes and undocumented inconsistencies between providers