Igor Melnyk
Arxiv Papers
Running out of time to catch up with new arXiv papers? We take the most impactful papers and present them as convenient podcasts. If you're a visual learner, we offer these papers in an engaging video format. Our service fills the gap between overly brief paper summaries and time-consuming full paper reads. You gain academic insights in a time-efficient, digestible format. Code behind this work: https://github.com/imelnyk/ArxivPapers
Where to listen?
Podcasts in the app Replaio Radio Coming soonPodcasts are coming to the app soon. Install now and be the first to see a whole new take on podcasts
Install for free
Android 5M+ downloads · 4.8 rating
iOS soon
Episodes
Foundational Autoraters: Taming Large Language Models for Better Automatic Evaluation 20.07.2024 38:59
https://arxiv.org/abs//2407.10817 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify: https://podcasters.spotify.com/pod/show/arxiv-papers
[QA] Scaling Retrieval-Based Language Models with a Trillion-Token Datastore 19.07.2024 15:23
This paper explores how increasing datastore size enhances retrieval-based language models' performance, demonstrating that smaller models with large datastores outperform larger models in knowledge-intensive tasks. https://arxiv.org/abs//2407.12854 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/a...
Scaling Retrieval-Based Language Models with a Trillion-Token Datastore 19.07.2024 30:26
This paper explores how increasing datastore size enhances retrieval-based language models' performance, demonstrating that smaller models with large datastores outperform larger models in knowledge-intensive tasks. https://arxiv.org/abs//2407.12854 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/a...
[QA] Beyond KV Caching: Shared Attention for Efficient LLMs 19.07.2024 16:21
This paper presents a Shared Attention mechanism that improves the efficiency of large language models by sharing attention weights across layers, reducing computational resources while maintaining performance. https://arxiv.org/abs//2407.12866 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-pape...
Beyond KV Caching: Shared Attention for Efficient LLMs 19.07.2024 23:14
This paper presents a Shared Attention mechanism that improves the efficiency of large language models by sharing attention weights across layers, reducing computational resources while maintaining performance. https://arxiv.org/abs//2407.12866 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-pape...
[QA] Private prediction for large-scale synthetic text generation 18.07.2024 8:27
Approach for generating differentially private synthetic text using large language models through private prediction, enabling creation of thousands of high-quality data points for various applications. https://arxiv.org/abs//2407.12108 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id169...
Private prediction for large-scale synthetic text generation1 18.07.2024 14:16
Approach for generating differentially private synthetic text using large language models through private prediction, enabling creation of thousands of high-quality data points for various applications. https://arxiv.org/abs//2407.12108 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id169...
[QA] Chip Placement with Diffusion 18.07.2024 7:51
Novel diffusion model for macro placement in digital circuit design outperforms existing reinforcement learning methods by placing all components simultaneously. https://arxiv.org/abs//2407.12282 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify: https://podcasters.spoti...
Chip Placement with Diffusion 18.07.2024 12:16
Novel diffusion model for macro placement in digital circuit design outperforms existing reinforcement learning methods by placing all components simultaneously. https://arxiv.org/abs//2407.12282 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify: https://podcasters.spoti...
[QA] Unraveling the Truth: Do LLMs really Understand Charts? A Deep Dive into Consistency and Robustness 17.07.2024 9:28
This paper evaluates Visual Language Models for Chart Question Answering, revealing performance variations and proposing improvements for more robust systems in diverse question and chart scenarios. https://arxiv.org/abs//2407.11229 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476...
Unraveling the Truth: Do LLMs really Understand Charts? A Deep Dive into Consistency and Robustness 17.07.2024 16:29
This paper evaluates Visual Language Models for Chart Question Answering, revealing performance variations and proposing improvements for more robust systems in diverse question and chart scenarios. https://arxiv.org/abs//2407.11229 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476...
[QA] Does Refusal Training in LLMs Generalize to the Past Tense? 17.07.2024 7:47
Refusal training gaps: Past tense reformulations can jailbreak LLMs. Future tense less effective. Alignment techniques may not generalize. Code and artifacts available. https://arxiv.org/abs//2407.11969 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify: https://podcaster...
Does Refusal Training in LLMs Generalize to the Past Tense? 17.07.2024 12:50
Refusal training gaps: Past tense reformulations can jailbreak LLMs. Future tense less effective. Alignment techniques may not generalize. Code and artifacts available. https://arxiv.org/abs//2407.11969 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify: https://podcaster...
[QA] No Train, all Gain: Self-Supervised Gradients Improve Deep Frozen Representations 16.07.2024 7:10
FUNGI enhances vision encoder features using self-supervised gradients, improving performance across datasets and tasks without additional training. Code available at https://github.com/WalterSimoncini/fungivision. https://arxiv.org/abs//2407.10964 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-...
No Train, all Gain: Self-Supervised Gradients Improve Deep Frozen Representations 16.07.2024 7:28
FUNGI enhances vision encoder features using self-supervised gradients, improving performance across datasets and tasks without additional training. Code available at https://github.com/WalterSimoncini/fungivision. https://arxiv.org/abs//2407.10964 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-...
[QA] LLM Circuit Analyses Are Consistent Across Training and Scale 16.07.2024 6:58
Study tracks how mechanisms evolve in large language models during training, finding consistent emergence of task abilities and functional components across different model scales. https://arxiv.org/abs//2407.10827 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify: https...
LLM Circuit Analyses Are Consistent Across Training and Scale 16.07.2024 13:09
Study tracks how mechanisms evolve in large language models during training, finding consistent emergence of task abilities and functional components across different model scales. https://arxiv.org/abs//2407.10827 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify: https...
[QA] SPREADSHEETLLM: Encoding Spreadsheets for Large Language Models 15.07.2024 10:11
SPREADSHEETLLM introduces efficient encoding for large language models to enhance understanding and reasoning on spreadsheets, achieving superior performance and compression ratios in various tasks. https://arxiv.org/abs//2407.09025 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476...
SPREADSHEETLLM: Encoding Spreadsheets for Large Language Models 15.07.2024 17:48
SPREADSHEETLLM introduces efficient encoding for large language models to enhance understanding and reasoning on spreadsheets, achieving superior performance and compression ratios in various tasks. https://arxiv.org/abs//2407.09025 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476...
[QA] Transformer Layers as Painters 15.07.2024 7:28
The paper explores the impact of removing or reorganizing information in pretrained transformers, finding differences in layers and suggesting potential improvements for model usage and architecture. https://arxiv.org/abs//2407.09298 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id169247...
Transformer Layers as Painters 15.07.2024 11:17
The paper explores the impact of removing or reorganizing information in pretrained transformers, finding differences in layers and suggesting potential improvements for model usage and architecture. https://arxiv.org/abs//2407.09298 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id169247...
[QA] Lynx: An Open Source Hallucination Evaluation Model 14.07.2024 9:22
LYNX is a state-of-the-art hallucination detection model that outperforms other language models on a new benchmark, HaluBench, for identifying unsupported information in text generation. https://arxiv.org/abs//2407.08488 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify:...
Lynx: An Open Source Hallucination Evaluation Model 14.07.2024 9:28
LYNX is a state-of-the-art hallucination detection model that outperforms other language models on a new benchmark, HaluBench, for identifying unsupported information in text generation. https://arxiv.org/abs//2407.08488 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify:...
[QA] Deconstructing What Makes a Good Optimizer for Language Models 14.07.2024 11:27
Comparing optimization algorithms for language models, finding no clear winner. Introducing simplified versions of Adam for improved performance and hyperparameter stability. https://arxiv.org/abs//2407.07972 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify: https://pod...
Deconstructing What Makes a Good Optimizer for Language Models 14.07.2024 14:24
Comparing optimization algorithms for language models, finding no clear winner. Introducing simplified versions of Adam for improved performance and hyperparameter stability. https://arxiv.org/abs//2407.07972 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify: https://pod...
Similar podcasts
Listen to the Arxiv Papers podcast in Replaio
Radio and podcasts in one app - free, with no sign-up. Install today and do not miss the launch
Replaio is not a podcast publisher; show names, artwork and audio belong to their authors and are distributed through public RSS feeds.
