- AI Dev and Research News
- Posts
- ⏰ Featured AI: Open AI Releases PaperBench and Salesforce AI Introduce BingoGuard......
⏰ Featured AI: Open AI Releases PaperBench and Salesforce AI Introduce BingoGuard......
Hi There,
Dive into the hottest AI breakthroughs of the week—handpicked just for you!
Nomic Open Sources State-of-the-Art Multimodal Embedding Model
Nomic has announced the release of “Nomic Embed Multimodal,” a groundbreaking embedding model that achieves state-of-the-art performance on visual document retrieval tasks. The new model seamlessly processes interleaved text, images, and screenshots, establishing a new high score on the Vidore-v2 benchmark for visual document retrieval. This advancement is particularly significant for retrieval augmented generation (RAG) applications working with PDF documents, where capturing both visual and textual context is crucial.……..
Open AI Releases PaperBench: A Challenging Benchmark for Assessing AI Agents’ Abilities to Replicate Cutting-Edge Machine Learning Research
OpenAI has introduced PaperBench, a benchmark designed to evaluate the competence of AI agents in autonomously replicating state-of-the-art machine learning research. PaperBench specifically measures whether AI systems can accurately interpret research papers, independently develop the necessary codebases, and execute experiments to replicate empirical outcomes. The benchmark comprises 20 papers selected from ICML 2024, covering areas including reinforcement learning, robustness, and probabilistic methods. Detailed rubrics, co-developed with original paper authors, specify 8,316 individually gradable tasks to facilitate precise evaluation of AI capabilities.……..
Meta AI Proposes Multi-Token Attention (MTA): A New Attention Method which Allows LLMs to Condition their Attention Weights on Multiple Query and Key Vectors
Meta AI addresses this limitation by introducing Multi-Token Attention (MTA), an advanced attention mechanism that conditions attention weights simultaneously on multiple query and key vectors. MTA integrates convolution operations over queries, keys, and attention heads, thus enhancing the precision and efficiency of contextual information retrieval. Specifically, the MTA framework consists of two convolutional components: key-query convolution, which aggregates multiple token signals within individual attention heads, and head mixing convolution, which facilitates information sharing among different attention heads. Additionally, the implementation employs group normalization with depth-dependent scaling to stabilize gradient flow, further improving model training stability and efficacy……..
Salesforce AI Introduce BingoGuard: An LLM-based Moderation System Designed to Predict both Binary Safety Labels and Severity Levels
Salesforce AI introduces BingoGuard, an LLM-based moderation system designed to address the inadequacies of binary classification by predicting both binary safety labels and detailed severity levels. BingoGuard utilizes a structured taxonomy, categorizing potentially harmful content into eleven specific areas, including violent crime, sexual content, profanity, privacy invasion, and weapon-related content. Each category incorporates five clearly defined severity levels ranging from benign (level 0) to extreme risk (level 4). This structure enables platforms to calibrate their moderation settings precisely according to their specific safety guidelines, ensuring appropriate content management across varying severity contexts.……..
Snowflake Proposes ExCoT: A Novel AI Framework that Iteratively Optimizes Open-Source LLMs by Combining CoT Reasoning with off-Policy and on-Policy DPO, Relying Solely on Execution Accuracy as Feedback
Snowflake introduces ExCoT, a structured framework designed to optimize open-source LLMs through the combination of CoT reasoning and iterative preference optimization, specifically utilizing off-policy and on-policy DPO guided exclusively by execution accuracy feedback. ExCoT dispenses with external reward models and human annotations, relying instead on internally generated reasoning steps and execution results. The method operates in two principal phases: initially, it generates candidate CoT data validated through off-policy DPO, forming the basis for supervised fine-tuning. Subsequently, the model iteratively generates and refines CoT data via on-policy DPO, incrementally improving accuracy through feedback derived from execution correctness..……..
