hisham hm

🔗 Let’s chronicle the AI hellscape

Researchers Jailbreak Chatbots with ASCII Art

“ArtPrompt, developed by researchers in Washington and Chicago, can bypass large language models’ (LLMs) built-in security features. The tool generates ASCII art prompts to get AI chatbots to respond to queries they are supposed to reject, like those referencing hateful, violent, illegal, or harmful content. ArtPrompt replaces the “safety word” (the reason for rejecting the submission) with an ASCII art representation of the word, which does not trigger the ethical or security measures that would prevent a response from the LLM.” - full article - Tom’s Hardware; Mark Tyson (March 7, 2024)

In that same ACM TechNews, they published a post with the cheerful and absurd title “Tiny Swimming Robot Thinks for Itself”.

🔗 That time I almost added Tetris to htop

Confession time: once I *almost* added a terminal version of Tetris as an Easter egg in htop.

I managed to implement a real crude but working version of it code golfing to make it as short as possible and got it pretty tiny, then added it to the help screen so it would activate by typing h, t, o, p (since h would take you to the help screen and the other keys would be nops in that screen).

Then there’s the question of how to hide an Easter egg in a FOSS codebase… The best I could think of was to make it into a long one-liner starting at column 200 so that most people looking at the code without word-wrapping editors would miss it. But after everything was coded, I decided that trying to “sneak code in”, even in my own codebase, was a bad practice and the good intention of innocent fun wasn’t worth it.

My fascination with Tetris goes way back. I first implemented it when I was in high school, and it getting it done really gave me pause: that was a real program, something that people paid real money for in Nintendo cartridges. It was the first time I thought I could really call myself a programmer for real. At the same time, it was my first contact with the ethics of software. I had never heard of FOSS then, and yet I asked myself: “what if my friends ask for the source code? what should I do?”

Years later, when we did the first CD version for our GoboLinux distro, I took an existing ncurses version of Tetris and hacked it into our installer, adding a progress bar that showed the status of files copying from CD to disk, while the user played the game (distro installers took forever back then!). Everyone loved it–except for the fact that it was supposed to auto-quit when the installation was finished but we changed the list of packages last minute so it got the count wrong.

A lot of people just kept playing for a long time without realizing the installation was done! (But it wasn’t too bad, they could just press Esc or something to quit and finish the install.)

Our early Gobo releases were full of little fun tweaks like that. In one release we included an emulator and legend has it that some hidden folder contains a ROM (not Tetris!), but not even I remember where that is, and that ISO probably isn’t even online anymore. (We really should have preserved our old stuff better!)

The memory of the Tetris installer in Gobo having a last-minute bug was another thing that dispelled me from the idea of the Tetris Easter egg in htop: while having bugs is just normal, I couldn’t bear the thought of htop having some serious bug caused by code added for silly reasons…

htop has its fair share of “unnecessary code”, such as the “big-digit LCD” meter and the themes, which are more artsy than utilitarian and I stand by them. If anything, I think software in general should be more artsy.

But “hidden Tetris in htop causes buffer overflow” would be terrible PR for the project (and my reputation by extension, I guess). That along with the bad taste in the mouth of the idea of hiding code in FOSS left made me drop the Easter egg idea.

I wish I still had that code, though! If only to keep it to myself as an autobiographical side-note.

Come to think of it, after writing all of this I realize I probably _should_ have included that code… as a comment!! Maybe that’s the way to do Easter eggs in FOSS? Add a fun/silly feature but leave it commented out, so that someone tinkering with the code finds it, enables it and has fun with it for a bit. I know that *I* would have enjoyed finding something like that in a codebase.

Oh well, maybe someday I’ll pull this off in some project.

🔗 Finally upgraded FlatPress

DreamHost finally bugged me to upgrade FlatPress from my super-old version running PHP 7 to the latest one running PHP 8.1. Making a tiny blog post just to test it — hope this works!

🔗 Finally got rid of a/ and b/ in git diff outputs!

You know how there are these little annoyances that are just mild enough so that you do nothing about it?

In the world of open source there’s always this notion of “if you want something to be different, the code is there, you can change it”, but most often this is not practical: I would never go about carrying a patched version of Git with me to every machine I work on just because of the annoying `a/` and `b/` prefixes that show up on Git diffs.

But those tiny prefixes always made me unable to select and paste a filename with a double-click and a middle-click on the terminal.

Today, after who knows how many years, I decided to make a search about it — “I can’t be the only one annoyed by this, right?” — and lo and behold: someone did ask about this on StackOverflow, and there is a global configuration to disable those prefixes:

git config --global diff.noprefix true

And just like that, this annoyance is gone!

🔗 Turns out gcc has imperative argument handling

The Linux program with most contrived argument handling logic ever has got to be gcc.

Everything in it has a reason, of course, but the end result is that you get a weird mix where the order matters for some args and not for others PLUS there are imperative arguments:

Say you want to link a static library into your program (I’m going to use […] to skip other flags)

gcc -o myprogram [...] myprogram.c libmylibrary.a [...]

This works, but now you want to add plugins to your program. So you add some runtime dynamic linking logic and add -ldl.

Oops, you realize your plugins can’t find some symbols from the static library, only those already used by the main program. The compiler threw away everything from libmylibrary.a that was “unused”.

-Wl,–whole-archive to the rescue!

Wait, what’s that? Two flags joined by a comma?

Turns out gcc is a main driver command which launches other programs, and passes arguments along to them. -Wl,–something means that it will pass the flag –something to the linker. You can add after -Wl, anything that is understood by ld, the GNU Linker.)

But you have other libraries you’re linking as well, and now you start getting duplicated symbol errors when compiling, because it is linking too much stuff! The solution? Wait for it…

gcc [...stuff...] -Wl,--whole-archive libfoo.a -Wl,--no-whole-archive [...other libs...]

The arguments in gcc when dealing with linker options are not only positional, they are imperative!

And I mean that in a quite literal sense. They interpreted like a sequence with side-effects: you set a flag, the next libraries is affected by it, you unset the flag, the following libraries aren’t affected anymore.

I thought find was a strong contender for Unix command with the weirdest argument handling, but I guess gcc takes the cake. 🍰


Follow

🐘 MastodonRSS (English), RSS (português), RSS (todos / all)


Last 10 entries


Search


Admin