Letiția Pârcălăbescu
@letiepi
Followers
1K
Following
1K
Media
13
Statuses
409
👩🎓 PhD, 🧠 AI Researcher, 📺 ML videos creator on YouTube: https://t.co/yd92HLx342 ▶ @AICoffeeBreak
Joined August 2020
@letiepi has a new Twitter handle: @AICoffeeBreak. Keeping the professional profile separated from the private one. Once again, I would like to express my gratitude for being on this amazing list!.
My favorite ML educators: [1/n]. - @JayAlammar: visual guides on NLP.- @fishnets88: intuition & real-world anecdotes.- @CShorten30, @ecsquendor, @ykilcher : netflix of research papers.- @letiepi: bite-size animated videos on ML.- @nlpguy_: experience on NLP in industry.
0
2
14
RT @AICoffeeBreak: 🧠🤖 Can we trust AI in science?.I'm excited to be speaking at the final event of the Young Marsilius Fellows 2025, themed….
0
2
0
RT @AICoffeeBreak: 💡 AlphaEvolve is a new AI system that doesn’t just write code, it evolves it. It uses LLMs and evolutionary search to ma….
0
4
0
RT @AICoffeeBreak: Excited to share that I’ll be joining the Summer School “AI and Human Values” this September at the Marsilius-Kolleg of….
0
2
0
RT @AICoffeeBreak: Long videos are a nightmare for language models—too many tokens, slow inference. ☠️.We explain STORM ⛈️, a new architect….
0
4
0
Quick thread with paper summary: .
The last paper of my PhD @HD_NLP is accepted at ICLR 2025!🙌.We investigate the reliance of modern Vision & Language Models (VLMs) on image🖼️ vs. text📄 inputs when generating answers vs. explanations, revealing fascinating insights into their modality use and self-consistency.👇.
1
0
3
The last paper of my PhD @HD_NLP is accepted at ICLR 2025!🙌.We investigate the reliance of modern Vision & Language Models (VLMs) on image🖼️ vs. text📄 inputs when generating answers vs. explanations, revealing fascinating insights into their modality use and self-consistency.👇.
0
4
89
RT @AICoffeeBreak: We all know quantization works at inference time, but researchers successfully trained a 13B LLaMA 2 model using FP4 pre….
0
16
0
RT @AICoffeeBreak: Just say “Wait…” – and your LLM gets smarter?!.We explain how just 1,000 training examples + a tiny trick at inference t….
0
7
0
RT @AICoffeeBreak: 🎙️ Yesterday, I gave a keynote on large language models outfitted with visual understanding, and the faithfulness of the….
0
5
0
RT @ecekt2: It was very nice seeing @letiepi aka @AICoffeeBreak again after many years😊 She gave a wonderful keynote at the Railway Museum….
0
1
0
RT @AICoffeeBreak: We explain 🥥COCONUT (Chain of Continuous Thought), a new paper using vectors for CoT instead of words. We break down:. -….
0
12
0
(3)🔎We provide an update of the accuracies reached by state-of-the-art VLMs on the VALSE 💃benchmark .🎯 Even modern VLMs still struggle with most phenomena tested by VALSE💃, although there are strong improvements from models such as mplug-owl3.
0
0
6
(2)🔎We evaluate VLMs' self-consistency when generating post-hoc and CoT explanations. 🎯 Most VLMs are less self-consistent than LLMs. For all models, the contributions of the image are significantly stronger when generating explanations compared to answers.
1
0
5
(1)🔎We measure how much VLMs use text and images when generating predictions or explanations. 🎯 We find that VLMs are heavily text-centric when producing answers and natural language explanations.
1
0
6
The last paper of my PhD @HD_NLP is accepted at ICLR 2025!🙌.We investigate the reliance of modern Vision & Language Models (VLMs) on image🖼️ vs. text📄 inputs when generating answers vs. explanations, revealing fascinating insights into their modality use and self-consistency.👇.
1
9
55
RT @AICoffeeBreak: An educational and a bit historical deep dive into LLM research. 💡Learn what breakthroughs since 2017 paved the way for….
0
15
0