Tag: LLMs

Unlock faster material discovery! Learn how HalluMat tackles the concerning issue of **LLM hallucinations**-fabricated or inaccurate data-that can derail scientific ...

Discover DiRL, a groundbreaking reinforcement learning method that significantly enhances the performance of diffusion language models. Learn how this innovative ...

Unlock peak AI performance! Discover how prioritized bandits intelligently manage scarce resources like compute & bandwidth, adapting to fluctuating workloads ...

Discover how researchers are pushing AI boundaries with groundbreaking 1-bit LLM quantization! Learn about this extreme precision reduction and its ...

Unlock a new approach to tackling complex challenges! Learn how researchers are combining powerful language models with dedicated solvers for ...

Explore the emerging field of LLM privacy forgetting! Learn how "selective forgetting" aims to remove specific training data from powerful ...

Unlock faster, more efficient large language models! CodeGEMM tackles a key bottleneck in quantized LLMs - dequantization - streamlining matrix ...

Explore a groundbreaking LLM architecture! "Compression is Routing" tackles context limitations & computational costs by using reconstruction error to guide ...

Discover how Google's new Agent Development Kit (ADK) for Go is revolutionizing AI agent development. Build intelligent systems directly within ...

AI agents are generating excitement, but many projects fail to deliver real value. Learn how to navigate the hype cycle ...

As autonomous AI systems become more powerful, ensuring their safe operation is critical. This research proposes a new approach to ...

Discover NeuroAda, a new framework for **efficient fine-tuning** of AI models! It reduces training costs and improves performance by adapting ...

New research reveals how LLMs’ emotional reasoning is skewed by user memory, reinforcing societal biases. Personalization may inadvertently embed social ...

Learn how to configure and verify a distributed training cluster with AWS Deep Learning Containers on Amazon EKS for large ...

GTAlign uses game theory to align LLMs for mutual benefit, improving efficiency, answer quality & user satisfaction. Code available on ...

)) Enhanced Transparency & Neuro-Symbolic Potential The benefits of this approach extend beyond improved accuracy. The stepwise process provides enhanced ...

ProSEA is a new agent framework that uses exploration & plan evolution for iterative problem-solving. It outperforms baselines on FinanceBench ...

New research introduces BaRP, a bandit-feedback approach that optimizes LLM routing by adapting to partial feedback & tuning performance/cost trade-offs.

APE boosts LLM judge reliability by adaptively incorporating auxiliary evaluation dimensions & a 'Collective Confidence' approach, improving human agreement rates.

PuzzlePlex benchmark assesses reasoning & planning of LLMs via diverse puzzles. It reveals insights into instruction vs code approaches for ...

TS-Agent is a new agent improving time series reasoning by leveraging LLMs for synthesis & dedicated tools for analysis, mitigating ...

New framework Dynamic Adjudication Template (DAT) enhances LLMs' complex rule processing, outperforming CoT & enabling smaller models to achieve high ...

New research bridges LLMs & formal methods, using GPT-5 to convert plans into verifiable specifications. Achieves 96.3% accuracy in a ...

New research introduces ConCuR & KernelCoder, a dataset & model generating high-quality GPU kernels using concise reasoning. Outperforms leading models!

MCCE framework unites closed-source & open-source LLMs, enabling collaborative evolution and state-of-the-art results in complex optimization like drug design.

New research explores how LLMs like ChatGPT can understand & optimize business processes (BPMN). Findings show promise for assisting designers ...

New research introduces Encode-Think-Decode (ETD), a technique boosting LLM reasoning by iterating over key layers & employing recursive latent thoughts. ...

Unlock the future of materials! Learn how innovative AI techniques are revolutionizing alloy discovery, accelerating development & paving the way ...

New research explores how training data impacts test-time scaling for LLMs, revealing insights into compute efficiency, performance risks & the ...

New research reveals that advanced LLMs struggle to recognize their own generated text, exhibiting biases & highlighting crucial implications for ...