Yeah. I think it’s funny/entertaining because it screws over Adobe, and fuck Adobe. But it doesn’t look like anything I’d like to actually use. This same “developer” ripped Microsoft Office too… but we’ve already got LibreOffice (not as booty garbage as it used to be, apparently) and OnlyOffice doing that, and they’re actually maintained and tested and patched by a wide userbase and often dedicated developers.
Adding a disclaimer here that I’m gonna get a little long-winded:
Contributing is hard and requires waiting. Some open source projects also discourage/ban AI and AI-assisted PRs, often because the developers just can’t keep up with the sheer amount of slop code someone can generate and shove out.
I’ve been helping diagnose some issues with RamSleuth, @MadGoat’s Linux utility for RAM speed/timings and some other bits. A model does most of the work, but I still have to read a bunch, run various commands, install some things, double-check information, and then write up a human-readable report with what the issue is. And then I wait for MadGoat to have the time to look at the issues, decide what they want to do about them, and then issue a fix (or not).
All this means no instant gratification, no “claude, make aida64 on linux, make no mistakes”, and I have to actually talk to someone else about things. I can’t sit in my little AI bunker and be master of the world.
Even my own utilities take ages just working with myself. I have my own AI-built git server utility and I already ran into an issue where it didn’t negotiate shallow checkouts, so my Gentoo Portage overlay would just stop syncing once there were 15 commits. I still need to tweak the UI/UX and un-slop the instructions in case anyone else were to want to use it. At current it’s “good nuff” because it’s just me using it for my own projects.
My own voltage/clock monitor only reads my specific motherboard’s sensors, and it’s a hand-rolled entry for which “in#” entry is actually the vCore input. I would need to do a lot of thinking and reading about things to figure out a good way to pull numbers from other platforms. It’s part of what I’ve been doing with RamSleuth, because Alder Lake (Intel 12th Gen) reports things differently from both Ryzen (what MadGoat has) and older Intel platforms. And it’s a mixed-RAM arch, it can have DDR4 or DDR5 DIMMs, so something meaning to read RAM information needs to consider that as well. The PDF Intel has with the DDR4/DDR5 registers (and every other register 12th gen has) is so long that you’ll explode your context and lobotomize your agent by trying to feed it the whole thing. So you have to - gasp! - look through the table of contents, find the correct section, and clip those pages so a model can pull the register table and know what each entry actually means.
That’s the longer version of the reason above, why I’m not really interested in these AI slop reverse-engineerings with completely vibe coded websites that look like they had 0 input. It comes across to me like someone who wants to push the “make the thing I want right now” button and not engage any further. So I’m just not confident in their ability to maintain the project long-term.
