Classifying the entirety of SafeDocs, Santiago Pedroza utilized a combination of LLMs, Embeddings Models, XGBoost, and LinearRegressors, achieving a significant improvement in accuracy with XGBoost embeddings model reaching 85.26% after hyperparameter tuning. SafeDocsDARPA SafeDocs program
Automated Design of Agentic Systems
Researchers introduce Automated Design of Agentic Systems (ADAS), aiming to automate the creation of powerful agentic systems, potentially inventing novel components and architectures.
Attackers can exfil data with Slack AI
Attackers can exfiltrate data from private Slack channels they are not a member of by using indirect prompt injection, exploiting the Slack AI's inability to distinguish between legitimate user queries and malicious instructions.
Music recommendation system using transformer models
Google Research introduces a music recommendation system employing Transformer models to analyze the sequence of user actions, enhancing personalization based on current context.
LLMs know more than what they say
Latent Space Readout (LSR) techniques applied to Large Language Models (LLMs) enhance evaluation accuracy for hallucination detection and numeric scoring, outperforming traditional fine-tuning methods with significantly fewer examples of human feedback.
JPEG-LM: LLMs as Image Generators with Canonical Codec Representations
JPEG-LM leverages canonical codec representations like JPEG for image generation, outperforming traditional pixel-based and vector quantization methods in efficiency and effectiveness.
Searching a Codebase in English
Semantic search in codebases is more effective when code is translated to natural language before vector embedding, highlighting the distinct semantic nature of code versus prose.
R2GenCSR: Retrieving Context Samples for Large Language Model based X-ray Medical Report Generation
The R2GenCSR framework enhances X-ray medical report generation by integrating context retrieval with Large Language Models (LLMs), improving the extraction of effective information for text generation.
Join Our Global Paper Reading Group for a Deep Dive into "Plan Like a Graph (PLaG)"
The "Plan Like a Graph (PLaG)" method significantly enhances LLM performance by decomposing tasks into sub-tasks that form an execution graph, enabling parallel and sequential task execution without fine-tuning.
PEDAL: Enhancing Greedy Decoding with Large Language Models using Diverse Exemplars
PEDAL introduces a method to enhance greedy decoding in large language models by incorporating diverse exemplars, aiming to improve output quality and relevance.
The Ungrounded Alignment Problem
The Ungrounded Alignment Problem explores how to embed predefined knowledge in machine learning systems that process abstract stimuli without prior labeling, focusing on unsupervised learners recognizing patterns in image sequences.
Do you think the best LLM is memorizing all possible downstream tasks and handle their variations well?
Overfitting the universe eliminates the gap between training and production data, suggesting that large models could theoretically memorize all variations of downstream tasks without overfitting.
Molecular Graph Representation Learning Integrating Large Language Models with Domain-specific Small Models
MolGraph-LarDo integrates Large Language Models (LLMs) with Domain-specific Small Models (DSMs) to enhance molecular property prediction, crucial for drug discovery, by balancing general knowledge with domain-specific accuracy.
TeamLoRA: Boosting Low-Rank Adaptation with Expert Collaboration and Competition
TeamLoRA introduces a novel PEFT method that incorporates expert collaboration and competition modules, aiming to enhance multi-task learning without compromising efficiency.
AI Chases the Storm: New NVIDIA Research Boosts Weather Prediction, Climate Simulation
NVIDIA's new generative AI model, StormCast, enhances weather prediction by emulating high-fidelity atmospheric dynamics, crucial for disaster planning amid increasing extreme weather events.