Symbolic Learning Enables Self-Evolving Agents

Agent symbolic learning represents a systematic framework that propels language agents towards autonomy by enabling them to optimize themselves using symbolic optimizers, moving beyond the need for extensive manual engineering.

Claude 3.5 Sonnet

Claude 3.5 Sonnet emerges as the new leading large language model (LLM), outperforming its predecessors in speed, cost-efficiency, and intelligence, available on Claude.ai and its iOS app with an API cost of $3 per million input tokens and $15 per million output tokens.

Show HN: R2R V2 – An open source RAG engine with prod features

R2R is an open-source Retrieval-Augmented Generation (RAG) system, designed to facilitate the transition from local LLM experimentation to scalable, production-ready applications, featuring a RESTful API for ease of integration.

How to think about creating a dataset for LLM fine-tuning evaluation

Alex Strick van Linschoten emphasizes the importance of evaluating fine-tuned LLMs for accuracy by comparing predictions against structured data derived from press releases, aiming to move beyond gut feelings to quantifiable metrics.

AI Revolutionized Protein Science, but Didn't End It

Google DeepMind's AlphaFold2 significantly advanced protein science by predicting protein structures with over 90% accuracy, yet it did not fully solve the protein folding problem, highlighting the distinction between predicting protein structures and understanding the folding process.

[R] Are Language Models Actually Useful for Time Series Forecasting?

Removing or replacing the LLM component in time series forecasting methods does not degrade results, often improving them instead, challenging the utility of LLMs in this domain.

Compressing graphs and indexes with recursive graph bisection (2016)

Recursive graph bisection significantly improves compression of graphs and indexes by enhancing data locality, extending the theoretical model of Chierichetti et al. from KDD 2009.

[R] Interpretability research in LLMs

Most interpretability research in LLMs has pivoted towards mechanistic interpretability, diverging from traditional methods like counterfactuals and saliency maps.

Optimized Nonlinear Regression Using Data Clustering [P]

Clustering data into smaller groups before forming a regression significantly reduces computation time, offering a novel approach to nonlinear regression optimization.

YannicKilcher - Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools (Paper Explained)

Researchers from Stanford and Yale have critically evaluated the accuracy of AI legal research tools, focusing on their propensity for "hallucinations"—the tendency to generate incorrect or misleading information.

Into the Omniverse: SyncTwin Helps Democratize Industrial Digital Twins With Generative AI, OpenUSD

SyncTwin GmbH leverages OpenUSD and NVIDIA's technologies to create digital twins that optimize industrial efficiency and sustainability.

🤗Welcome Gemma 2 - Google's new open LLM

Google's Gemma 2 introduces two model sizes with 9 billion and 27 billion parameters, both available in base and instruction-tuned versions, showcasing advancements in data training volume and model performance. Models on the Hub

Research Focus: Week of June 24, 2024

RENC, a system designed by Microsoft researchers, significantly reduces CPU power consumption in 5G vRAN servers by up to 45% by dynamically adjusting CPU frequency based on cellular workload variations, demonstrating a blend of innovative techniques for energy efficiency. Read the paper

Selective Prompting Tuning for Personalized Conversations with LLMs

Selective Prompt Tuning (SPT) addresses the challenge of integrating persona profiles into conversations by softly prompting LLMs, overcoming the limitations of textual prompting and direct fine-tuning.

AdaZeta: Adaptive Zeroth-Order Tensor-Train Adaption for Memory-Efficient Large Language Models Fine-Tuning

AdaZeta framework enhances memory efficiency in fine-tuning large language models (LLMs) by introducing a low-parameter tensorized adapter and an adaptive query number schedule, addressing the limitations of previous Memory-efficient Zeroth-order (MeZO) methods.