there’s this idea in the AI development zeitgeist that “rare data” is the only advantage left. this stuff is almost certainly motivated by that type of thinking. if they can have data in their datasets that no one else does, they view that as the only way left to be ahead of the competition.
chrash0
this latest AI stuff was just the last straw for me. Spotify has routinely shown disrespect for the artists they rely on
what i mean is that Claude Code wouldn’t report as a Linux desktop browser without some nefarious behavior. what i mean by “surely” is that it would be shitty of Anthropic or OpenAI to do that for no benefit on their part. curl, wget, httpie, xh, whatever client use a specific user-agent that would be a reasonable default to use for any agent without fucking around pretending just to skew metrics.
those header report that they’re bots.
sure, but is that implying that 5% of this traffic is bad faith or rogue agents? the tools built into Claude Code and Codex surely have their own user-agent settings.
and why would they spoof Linux instead of Windows?
the AI explanation and reasoning is weak. what does “largely disappears” mean without numbers? Cloudflare doesn’t do vibes; give me numbers. in any case “desktop” surely means that the user-agent reflects a desktop browser and not a CLI tool, right? otherwise this number would be closer to 90% than 10%. unless the agents are intentionally spoofing browser user-agents i don’t buy it.
no mention of Steam’s and Framework’s hardware offering that come with Linux? no mention of the clear mass migrations?
here’s an AI explanation: LLMs are helping people who otherwise would struggle move to Linux
why work on a feature that your users categorically reject? this isn’t me projecting, i literally have this conversation with a buddy all the time. he’s a programmer and talks about all the problems with Windows like there aren’t options. use the OS the way the devs designed it, pick a different one, or stfu
you say that like that makes it not worth doing
not precisely. Rust has let mut and Kotlin has val; ie the member layout and the mutability are treated differently, since those languages have mutability rules.
this is a circular argument anyway. the language designers could have figured out a way to do type inference like, eg, Java, but, again, explicitly chose not to.
i mean, i’m not arguing for this position necessarily. i agree with you.
that said, in practice, the “bad data” comes out in the wash. for example, all of Reddit is likely in their datasets. who knows what their data cleaning process is like, but diversity in the data is generally a good thing from a training perspective.