AI Papers Podcast for 04/24/2024

1. Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
2. The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions
3. FlowMind: Automatic Workflow Generation with LLMs
4. Music Consistency Models
5. How Good Are Low-bit Quantized LLaMA3 Models? An Empirical Study

AI Papers Podcast for 04/23/2024 PhysDreamer: Physics-Based Interaction with 3D Objects via Video Generation TextSquare: Scaling up Text-Centric Visual Instruction Tuning Does Gaussian Splatting need SFM Initialization? How Far Can We Go with Practical Function-Level Program Repair? AutoCrawler: A Progressive Understanding Web Agent for Web Crawler Generation 

AI Papers Podcast for 04/20/2024 Dynamic Typography: Bringing Words to Life Toward Self-Improvement of LLMs via Imagination, Searching, and Criticizing MeshLRM: Large Reconstruction Model for High-Quality Mesh MoA: Mixture-of-Attention for Subject-Context Disentanglement in Personalized Image Generation EdgeFusion: On-Device Text-to-Image Generation 

AI Papers Podcast for 04/19/2024 BLINK: Multimodal Large Language Models Can See but Not Perceive Reka Core, Flash, and Edge: A Series of Powerful Multimodal Language Models AniClipart: Clipart Animation with Text-to-Video Priors TriForce: Lossless Acceleration of Long Sequence Generation with Hierarchical Speculative Decoding OpenBezoar: Small, Cost-Effective and Open Models Trained on Mixes of Instruction Data 

AI Papers Podcast for 04/19/2024 Meta releases Llama 3, claims it's among the best open models available