Freda Shi

Interesting piece to read on how VLMs encode properties of objects!

🐻

That said, I also like the less intellectually exciting ones as they are solid analyses or methodological enhancements that work for realistic scenarios. It's even hard for me to "rank" them. I'm already prepared to see a reversed ranking from reviewers. (2/2)

I frankly don't think the ICML policy of having authors rank their papers will work. I have 3-5 submissions (depending on ICLR results) this time to quite different subcommunities. I know my favorite submissions could be quite controversial---people either like or dislike them a lot. (1/2)

Thrilled to announce the 1st Workshop on Computational Developmental Linguistics (CDL) at ACL 2026 🎉 A new venue at the intersection of development linguistics × modern NLP, spearheaded by @fredashi.bsky.social @marstin.bsky.social, and and outstanding team of colleagues!

A thread 🧵

(1) is "what AI is good at," and (2) is "what typical humans are bad at." But AI is not so good at *creatively* generating ideas that "make sense." Proposing such ideas is a task that does not match either criterion above, and is where human intelligence will keep shining.

2/2

AI coding is now the best at implementing things that are (1) either already standardized or not very ambiguous in their natural-language description, and (2) with many details.

Yes, I just vibe-coded a helper for some organizational matters, and will do more.
1/

Yeah I feel it’s indeed easier to have an okay statement with LLM draft, but its emphases are almost always off…

You mean narrative CV or something else? I tried using LLM to generate one but… garbage

NSF GRFP Application Tips for NLP, AI, CS Reflections and advice from my successful NSF GRFP proposal in NLP. Why I think my applications worked well, what I wish I did differently, and links to my actual statements and feedback from the GRFP...

It's #NSF #GRFP application season again so it's time to re-up my GRFP application advice post!

Also, check out the cool bsky comment integration I've added to the blog! Engagement with this post will go under the blogpost on my site as comments!

saxon.me/blog/2024/gr...

Reviewer timeliness by area (based on a small sample of 14 papers * 4 reviewers each): Multilingual LMs >> VLMs > LLM Reasoning.

PoeTree. Poetry corpora in 11 languages PoeTree is a standardized collection of poetry corpora comprising nearly 335,000 poems in ten languages (Czech, English, French, German, Hungarian, Italian, Norwegian, Portuguese, Russian, Slovenian, ...

We have updated our collection of multilingual poetry corpora PoeTree. With addition of Norwegian, and few metadata fixes it now has a loud label of 1.0.0 release! 🌳

versologie.cz/poetree/vers...

Should be fun to read!

I can only go through <40 slides in a one-hour talk... verified multiple times. My job talk has 39 slides (including acknowledgement), and so are my recent talks.
Always impressed when folks present 100 slides in the same amount of time.

AI isn't replacing radiologists Radiology combines digital images, clear benchmarks, and repeatable tasks. But demand for human radiologists is at an all-time high.

In 2016 Hinton predicted that AI would replace all radiologists in five years. Ten years later, why hasn't it happened? This post is a great explainer.

www.understandingai.org/p/ai-isnt-re...

Thanks, Chenhao! It's my pleasure helping with streamlining things in the backend - quite some pain but also fun :-)))

🚀 ACL ARR is looking for a Co-CTO to join me lead our amazing tech team and drive the future of our workflow. If you’re interested or know someone who might be, let’s connect!

RTs & recommendations appreciated.

Same! I‘ve literally unbidden 0 out of my recommended batch.

Also, is it now common to assume all reviewers are irresponsible and start to threaten them before deadline?

Is there a specific reason that NeurIPS does not show AC identity to reviewers? I'm very curious about who sent the polite reminders, and of course, even more curious about the rude ones.

#CoreCognition #LLM #multimodal #GrowAI We spent 3 years to curate 1503 classic experiments spanning 12 core concepts in human cognitive development and evaluated on 230 MLLMs with 11 different prompts for 5 times to get over 3.8 millions inference data points.

A thread (1/n) - #ICML2025 ✅

Just in case this personal latex environmental setup is helpful to a broader crowd ⬇️

I use local complication w/ VSCode latex workshop plugin + git for version control.
Pros: better hierarchical organization & doesn’t require internet.
Cons: sometimes need to resolve conflict (usually fine if commit frequently) & require a powerful laptop to match overleaf compilation speed.

I'm late to the NAACL party, but I've just arrived in Albuquerque! I'll talk about measuring faithfulness of verbalized reasoning on *Sunday* at Repl4NLP at *9:45*.

Screenshot of a slide that says "what does it take to convince ourselves that a system is exhibiting compositionality?" with a side comment "mostly AI, but humans too!!" for the word system

hello NAACL friends I'm giving a keynote today at RepL4NLP at 1:30PM local time, come say hi! I'll mostly be musing about things with light research discussions

If you are also at NAACL, let's chat!

Tutorial website here: language-grounding.github.io. More material forthcoming.

On my way to NAACL✈️! If you're also there and interested in grounding, don't miss our tutorial on "Learning Language through Grounding"!
Mark your calendar: May 3rd, 14:00-17:30, Ballroom A.

Another exciting collaboration with @marstin.bsky.social @kordjamshidi.bsky.social, Jiayuan, and Joyce!

📢Curious why your LLM behaves strangely after long SFT or DPO?
We offer a fresh perspective—consider doing a "force analysis" on your model’s behavior.
Check out our #ICLR2025 Oral paper:

Learning Dynamics of LLM Finetuning!

(0/12)

Congratulations Danica and Joshua!