Apr 19, 2025

• Packing Input Frame Context in Next-Frame Prediction Models for Video Generation

  • FramePack enables the diffusion of thousands of frames at full fps-30 using 13B models on a 6GB laptop GPU, showcasing its efficiency in resource management.

• OpenAI's new reasoning AI models hallucinate more

  • OpenAI's o3 and o4-mini models exhibit a higher hallucination rate than previous models, with o3 hallucinating 33% of the time on PersonQA, nearly double that of its predecessors.

• AI-Designed Antivenoms: New Proteins to Block Deadly Snake Toxins

  • AI-designed proteins effectively neutralize snake venom toxins, addressing a critical public health issue where 2 million people suffer from snakebites annually, leading to over 100,000 deaths.

• Principles for Building One-Shot AI Agents

  • One-shot AI agents automate complex code maintenance tasks without human intervention, leveraging static analysis and dependency updates to enhance efficiency and accuracy in software engineering workflows.

• Inside ArXiv

  • arXiv, created by Paul Ginsparg, revolutionized scientific communication by allowing researchers to share preprints instantly, bypassing traditional peer review, which can take months, thus accelerating the dissemination of knowledge, especially during crises like the Covid pandemic.

• Ink and Switch Constraint System (2023)

  • The constraint system developed in phase 2 aims to unify key aspects of dynamic mediums, enabling mechanical constructions and bidirectional computation while addressing issues like “floaty-ness” and “blow-ups” that hinder performance.

• How to Write a Fast Matrix Multiplication from Scratch with Tensor Cores

  • Optimized matrix multiplication using Tensor Cores on NVIDIA Tesla T4 GPUs can achieve up to 96% of cuBLAS throughput for large matrices, demonstrating significant performance improvements through techniques like hierarchical tiling, swizzling, and double buffering.

• Inferring the Phylogeny of Large Language Models

  • PhyloLM adapts phylogenetic algorithms to analyze Large Language Models (LLMs), revealing their interrelations and predicting performance metrics based on output similarity.

• [R] Biologically-inspired architecture with simple mechanisms shows strong long-range memory (O(n) complexity)

  • The new biologically-inspired architecture demonstrates strong long-range memory capabilities with O(n) time and memory complexity, achieved through simple mechanisms akin to signal accumulation.

• [P] How to handle highly imbalanced biological dataset

  • Handling highly imbalanced datasets in biological research, such as a peptide epitope dataset with 1 million non-epitope versus 300 epitope peptides, requires advanced techniques beyond simple oversampling and undersampling.