How and when should LLM guardrails be deployed to balance safety and user experience?
Our #EMNLP2025 paper reveals that crafting thoughtful refusals rather than detecting intent is the key to human-centered AI safety.
📄 arxiv.org/abs/2506.00195
🧵[1/9]
[9/9] Big THANKS to my amazing collaborators @jiajiah.bsky.social @pigeonzow.bsky.social Motahhare Eslami, Jena Hwang @faebrahman.bsky.social, @carolynrose.bsky.social @maartensap.bsky.social from @ltiatcmu.bsky.social
Pareto.ai @sfu.ca @ai2.bsky.social ♥️
📂 github.com/EEElisa/LLM-Guardrails
Me with my cat on the plane
Some life updates here: got a car, managed to escape grad school, working on expert data curation and future of work these days, had a new cat, moved back to SF, finally feels alive now.
I’m back!
Look at my kitty ❤️❤️
Ok, guys I cleared out the fridge and I… OH NO
Thank u!
Got Covid again 😢
Skeet!
Moss shower mat?
Vibe > existing follower count in getting new followers
People tweeting vs people skeeting
People skeeting
OMG Casey Neistat on bsky! Welcome Casey!
From what I’ve see, most proposed LLM based tool is really just as a proof of concept. If all the “moderation tool” do is to prompt the GPT to produce a formatted JSON file, there’s no way to tune and no barriers of entry.
✨
Repost as reminder 👀
The memes are too good
/honk Duck is getting out of control
following the protocols summer like 👀
The eternal September is pretty awesome after all.
/squeeee? 🤔
Bsky post giving out jazzing vibe ❤️
Can confirm.
Must! Share! Capybaras!
Arram ✨
This too, shall pass 🧘
Twitter right now: Apollo gets crazy after getting shot in the heart. 🙈
Oh this is the Commons!
