Arxiv paper – MINERVA: Evaluating Complex Video Reasoning In this episode, we discuss MINERVA: Evaluating Complex Video Reasoning by Arsha Nagrani, Sachit Menon, Ahmet Iscen, Shyamal Buch, Ramin Mehran, Nilpa Jha, Anja Hauth, Yukun Zhu, Carl Vondrick, Mikhai...

Listen to the AGI Breakdown podcast on Minerva here: aibreakdown.org/arxiv-paper-...

Advancing the frontier of video understanding with Gemini 2.5- Google Developers Blog Explore Gemini 2.5, enhancing video understanding and combining audio-visual data and code for new interactive applications

The newly released Gemini 2.5 Pro (Preview 05/06) sets the state-of-the art on Minerva with 63.5% accuracy. Human accuracy is 92.5%.

developers.googleblog.com/en/gemini-2-...

MINERVA: Evaluating Complex Video Reasoning Multimodal LLMs are turning their focus to video benchmarks, however most video benchmarks only provide outcome supervision, with no intermediate or interpretable reasoning steps. This makes it challe...

📜 Paper: arxiv.org/abs/2505.006...
📊 Dataset: github.com/google-deepm...

This is work with my amazing colleagues and collaborators Arsha Nagrani, Sachit Menon, Ahmet Iscen, Shyamal Buch, Ramin Mehran, Nilpa Jha, Anja Hauth, Yukun Zhu, Carl Vondrick, Mikhail Sirotenko, and Cordelia Schmid

We're excited to release Minerva 🕵️‍♀️, a benchmark to evaluate if AI can truly reason about videos, from spotting game-changing moments in sports 🏀 to understanding character motivations in short films 🍿. We provide the "why" behind the answers! Pointers below 👇

And the ICLR decisions

6yo daughter: Papa, are you the boss of Google?
Me: No
6yo daughter: Why?

Excited to share Long-Video Masked Autoencoder (LVMAE) our team just published at NeurIPS'24! We boost the context length of video models using an adaptive decoder and a dual-masking strategy and achieve SotA on several video benchmarks.

Paper: arxiv.org/abs/2411.13683

Whoa, massive news! Excited for you and looking forward to seeing what you'll build there!

Another nice way to get an ETA is

import tqm
for i in tqdm(range(len(dataset)):
...

Professor knocks - "Hey, I have a 'research' project for you"

Thanks, looks promising!

Oh nice, seems to work for the first few papers I tried. Thank you!

Is there a better way to find the publication venue of an ArXiv paper than searching for the title on Google / Google Scholar / OpenReview and checking authors' websites?

Tap, tap. Is this thing on?