Assaf Arbelle

“There is nothing quite so useless, as doing with great efficiency, something that should not be done at all.”

🎺 Here comes the official 2024 NeurIPS paper browser: - browse all NeurIPS papers in a visual way - select clusters of interest and get cluster summary - ZOOOOM in - filter by human assigned keywords - filter by substring (authors, titles) neurips2024.vizhub.ai #neurips by IBM Research Cambridge

Overleaf now incorporates an AI assistant, as of this morning. It is opt-out, not opt-in. I have very mixed feelings about this: and by "mixed", I mean ranging from annoyance to anger. www.overleaf.com/learn/how-to...

Gumbel-softmax (magic) trick: Gradients for high dimensional discrete variables are hard. Gradients for high dimensional continuous variables are easy. You swap a discrete variable by a continuous one and🤞no one realizes. Algos that need sequences of hard choices are not amused.

Super cool work! If you're interested in context-aware models- We've done it for autoregressive models through structured dialogue data

We just dropped CAT4D, text to dynamic 3D models that you can render in real time. Not posting a video because Bluesky is garbage in this respect; go straight to the real time viewer on a desktop browser and look around. The cat kneading dough is my favorite. cat-4d.github.io

Time sink 😡

Let’s make a generation of amazing open source image generation models from high quality data. The best image generation models train on human preferences. Unfortunately, many of these datasets are closed. Let’s change that! 🧵 we're building a community dataset and we need help reviewing!

Hello Bluesky!

You can access keynotes, papers, and spotlights on Robotic Learning for free! So cool!🤖 youtube.com/watch?v=0joZ... #Robotics #DeepLearning #CoRL2024

interesting paper on test-time training for solving ARC. a cool fact is that dreambooth can be considered as test-time training for subject driven generation. test-time training is very powerful t.co/OzbUEthhSo

go.bsky.app/VDFDYsm

I like this new analogy for working with LLMs by @emollick.bsky.social "treat AI like an infinitely patient new coworker who forgets everything you tell them each new conversation, one that comes highly recommended but whose actual abilities are not that clear" www.oneusefulthing.org/p/getting-st...

Welcome to the new Bluesky 🦋weekend arrivals! A few tips: 📸 set up a real profile w a bio 📨 post (a personal intro, a first skeet) 🎯 interact (like, and repost more, there is llittle algorithmic help here) 🔎 search for starter packs 😊 enjoy the friendly non-tech billionaire-politics space🦋🌈🚀

Creating a 🦋 starter pack for people working in IR/RAG: go.bsky.app/88ULgwY I can’t seem to find everyone though, help definitely appreciated to fill this out (DM or comment)!

Here we grow again. Introducing my 2nd AWESOME #ComputerVision Starter Pack. Missed you? Let me know in the thread. go.bsky.app/Gjcs339

My list is now FULL, 150/150. So many AWESOME people still joining this place!

After a phenomenal couple of weeks of following great accounts, I am yet again overwhelmed by all the available content and hence in a dire need of bookmarking features/ optimized frameworks for ‘later’ reading.

Feature request: Filter feed by LLM instead of keywords! I want to be able to say things like "No politics". Or "Research papers on multi-modal models". I find it crazy that most platforms don't let you control your feed besides through behavior (likes/mutes), let alone use AI to do so.

so! i now have about 1,000 more followers than i had two weeks ago, and it's all because i was featured in an #nlp starter pack, so i guess i better start pulling my weight around here? i'm yuval (preferred english pronounciation "you've all"), early-career faculty at bun-gurion university cs dept >

#KostasThoughts: Be generous with your retweets and likes as they can signal boost the work of early researchers.

I'm making a list of AI for Science researchers on bluesky — let me know if I missed you / if you'd like to join! go.bsky.app/AcP9Lix

Created a #robotics and #vision starter pack. Feel free to suggest yourself or others missing from the list! go.bsky.app/HcQYMj

NeurIPS Conference is now Live on Bluesky! -NeurIPS2024 Communication Chairs

some little bluesky tips 🦋 your blocks, likes, lists, and just about everything except chats are PUBLIC you can pin custom feeds; i like quiet posters, best of follows, mutuals, mentions if your chronological feed is overwhelming, you can make and pin make a personal list of "unmissable" people

How do LLMs learn to reason from data? Are they ~retrieving the answers from parametric knowledge🦜? In our new preprint, we look at the pretraining data and find evidence against this: Procedural knowledge in pretraining drives LLM reasoning ⚙️🔢 🧵⬇️

LLMs generate novel word sequences not contained in their pretraining data. However, compared to humans, models generate significantly fewer novel n-grams. RLHF = 30% *more* copying than base! Awesome work from the awesome Ximing Lu (gloriaximinglu.github.io) et al. 🤩 arxiv.org/pdf/2410.04265

A new paper, "Let Me Speak Freely" has been spreading rumors that structured generation hurts LLM evaluation performance. Well, we've taken a look and found serious issue in this paper, and shown, once again, that structured generation *improves* evaluation performance!

The last ten years in computer vision and multimodal AI have been a ride, yet I still feel there's so much more to explore.

I like how the Bluesky app lets you customize the pinned feeds. If you want more focus on your CV/ML conference buddy content, you can demote the Discover feed, or just delete it.

Hugging Face transformers pipelines now support vision language models for easy local inference 🫰🏻 you can also use inference API to infer hosted vision LMs (via Python, JS and cURL) huggingface.co/docs/api-inf...

Hi @bsky.app ! Love it here, but I can’t live without bookmarks… how can I keep track of all these wonderful posts without🦉? Don’t want to lose them 😭.

Here is a list of ML OSS & Open Source / Science enthusiasts I found on Bluesky 🦋 go.bsky.app/8MFcfXd Let me know if you find such people here! I'm still new here and probably the list misses many must-add people, so let's built it together💪

Whenever AI "generates" something impressive, the first question we should always ask is: "What does the closest sample in the training data look like?" LLMs are amazing interfaces for accessing the world's information but they need to be treated as the "search and synthesis" tools they are.

After my general computer vision starter pack is now full (150/150 entries reached), here is one specific to 3D Vision: go.bsky.app/Cfm9XFe

Productivity tip 💡 When you procrastinate on tasks you don't want to do until the last minute, you only need ONE MINUTE to complete them. Absolute time saver!

Every time I see someone post this image it goes viral

In a gratuitous attempt to acquire more followers myself 😁, I've made a start on a "starter pack". Hopefully as more people from 🐦 make it over to 🦋, we can extend this a bit. Suggestions welcome! I've noticed not all accounts seem to be eligible to be added, anyone know what's up with that? 🤔

Start gathering people working on conversational AI. DM/reply if you want to be added to this starter pack. go.bsky.app/Nz6zoJD

I recently started one on vision and language: go.bsky.app/TENRRBb

#ComputerVision community reaction 😒 (with audio)

OK, here it is: the starter pack to get to know the #multimodalists on Bluesky. Who is missing? go.bsky.app/QPLSxuX

Starter packs allow a maximum of 150 entries and so my computer vision starter pack is now full! 😳 It’s fantastic to see the CV community embrace @bsky.app! go.bsky.app/PkAKJu5

Hey :) I'm looking for 3 emergency reviewers for ARR submissions🚨📷 they are all in LM interpretability! Should be submitted within the next 36 hours 🙃 If you are interested, please DM or email me! #NLProc #NLP

Thanks @adhirajghosh.bsky.social for organizing this collection of Vision-Language researchers! These topic-specific starter-packs are very useful to get good visibility on all the interesting work people are doing in the field.

Looking for potential emergency reviewers for submissions in Interpretability and Model Analysis for #NAACL2024! Topics include: LLM Hallucination, Evaluation, Privacy Policy Analysis, and Unlearning. Please reach out if you have the bandwidth to help!🫡 #NLProc

Can anyone point me to a good starter pack for Computer Vision or Multimodal (Vision & Language) Researchers?

An easy way to follow a lot of great researchers from IBM Research!