Roman Lutz

Responsible AI Engineer | AI Red Teaming | Open source | @romanlutz.bsky.social

[Roman Lutz @romanlutz13]

Is there a name for the phenomenon that in #OpenSource as the size of a pull request grows the probability of it merging tends towards 0? If not I’m claiming it as Lutz’s law. #LutzsLaw

[Roman Lutz @romanlutz13]

@itsolelehmann @threadreaderapp unroll

[Roman Lutz @romanlutz13]

@lateinteraction @HanchungLee The autogenerated part probably refers to generating API references for a webpage from source+docstring.

I agree on redundancy (especially if you have inheritance… repeat the same?!? Or worse, have differing information between parent and child classes?!?).

[Roman Lutz @romanlutz13]

@maksym_andr Reading through some of the results I can’t help wondering that this could easily be done with better attack strategies/instructions for the model that generates attack prompts. A lot of it boils down to very simple requests to expand, to explain, etc.

[Roman Lutz @romanlutz13]

@DareFailed @levelsio Need this for the phone…

[Roman Lutz @romanlutz13]

@p_matula @markrussinovich Thanks for pointing that out! We just reorganized the docs but clearly forgot to update the README! The contents of the demo folder are now under docs/code/orchestrators. I’ll get that link updated, too.

[Roman Lutz @romanlutz13]

@maksym_andr PyRIT maintainer here 🙂 This was indeed based on your idea! Saw your tweet in June and a colleague put it together. Now that there’s a paper we should add a proper citation 🙂

[Roman Lutz @romanlutz13]

@elder_plinius @Izaya718 @markrussinovich See https://t.co/ov15pf5sRW

This needs to be sent as a first prompt, then the actual question/command goes into the second prompt.

[Ram Shankar Siva Kumar @ram_ssk]

Repro @markrussinovich new jailbreak in any LLM using our OSS AI Red team tool, PyRIT.

https://t.co/eV9hapwA8H

All possible because our awesome intern, Safwan Ahmad!

[Roman Lutz @romanlutz13]

@snick8484 @paolini @paolini any thoughts? I’m about to shut down my account on here (in favor of Mastodon) and would love to follow you over there, even if just crossposted…

[Roman Lutz @romanlutz13]

@aselbst @joinmastodon Excellent! Thanks for sharing 👍🏻🙂

[Roman Lutz @romanlutz13]

In the spirit of #TwitterMigration I’ll try something new. I’ll unfollow one person every day (assuming I’m logging on) and will ask them to consider moving over to @joinmastodon. At some point, I should be at 0 follows and ready to shut down my account.

[Roman Lutz @romanlutz13]

@nikiluk @OpenEthicsAI @joinmastodon Amazing! I found it. Do you also have an account over there? 🙂

[Roman Lutz @romanlutz13]

@UMassAmherst Fantastic! Now you only have to @joinmastodon 🙂

[Roman Lutz @romanlutz13]

@gvanrossum @joinmastodon ! 🙂

[Roman Lutz @romanlutz13]

@hildeweerts That’s why you joined @joinmastodon where things are nice and calm 🙂

[Roman Lutz @romanlutz13]

@hildeweerts Indeed I am! 🤣
Thanks, Hilde!!!

[Roman Lutz @romanlutz13]

@seanmcgregor @joinmastodon @IncidentsDB @indiewebcamp Certainly not a bad idea! Looking forward to seeing you over there 😉

[Roman Lutz @romanlutz13]

@seanmcgregor @joinmastodon @IncidentsDB I agree. I would run my own but there’s some effort involved. I don’t think it’s weird. I’ve seen a few people do that. Often it’s the [email protected] domains but honestly who cares? 🙂

[Roman Lutz @romanlutz13]

@RadicalAIPod @Dylantdoyle @_jessiejsmith_ @joinmastodon Glad to hear it’s on your mind! 🙂 Fortunately, I can still listen to your podcasts 🙂 Keep up the excellent work!

[Roman Lutz @romanlutz13]

@nikiluk @joinmastodon Manual. Whenever I log in I add another one. 🙂

[Roman Lutz @romanlutz13]

@seanmcgregor @joinmastodon @IncidentsDB Voila!

[Roman Lutz @romanlutz13]

@hildeweerts Woohoo! 🎉🚀🔥 #TwitterMigration

https://t.co/sKfQCvloqt
https://t.co/FLN0jbrYAr
https://t.co/rALNUvQ0gI
https://t.co/eB0LqLjVcw
might be of interest!
I’m on https://t.co/amppcDpY3l which I have been enjoying so far. No research or even ML focus there, though.

[Roman Lutz @romanlutz13]

@seanmcgregor @joinmastodon @IncidentsDB They typically use that if a server is full of people who do objectionable things, or violate a server’s rules in some way. I haven’t seen it much so far.

I wrote a few suggestions elsewhere the other day, let me share that…