Researcher Jindong Wang and Associate Professor Steven Euijong Whang explore the NeurIPS 2024 work ERBench. ERBench leverages relational databases to create LLM benchmarks that can verify model rationale via keywords in addition to checking answer correctness.
Next-token prediction trains a language model on all tokens in a sequence. VP Weizhu Chen discusses his team’s 2024 NeurIPS paper on how distinguishing between useful and “noisy” tokens in pretraining can improve token efficiency and model performance.
Can existing algorithms designed for simple reinforcement learning problems be used to solve more complex RL problems? Researcher Dylan Foster discusses the modular approach he and his coauthors explored in their 2024 NeurIPS paper on RL under latent dynamics.
Pranjal Chitale discusses the 2024 NeurIPS work CVQA. Spanning 31 languages and the cultures of 30 countries, this VQA benchmark was created with native speakers and cultural experts to evaluate model performance across diverse linguistic and cultural contexts.
The efficient simulation of molecules has the potential to change how the world understands biological systems and designs new drugs and biomaterials. Tong Wang discusses AI2BMD, an AI-based system designed to simulate large biomolecules with speed and accuracy.
Researchers Chris Hawblitzel and Jay Lorch share how progress in programming languages and verification approaches are bringing bug-free software within reach. Their work on the Rust verification tool Verus won the Distinguished Artifact Award at SOSP ’24.
In their 2024 SOSP paper, researchers explore a common—though often undertested—software system issue: retry bugs. Research manager Shan Lu and PhD candidate Bogdan Stoica share how they’re combining traditional program analysis and LLMs to address the challenge.
The personalizable object recognizer Find My Things was recently recognized for accessible design. Researcher Daniela Massiceti and software development engineer Martin Grayson talk about the research project’s origins and the tech advances making it possible.
Advanced AI may make it easier for bad actors to deceive others online. A multidisciplinary research team is exploring one solution: a credential that allows people to show they’re not bots without sharing identifying information. Shrey Jain and Zoë Hitzig explain.
A lack of appropriate data, decreased model performance, and other obstacles have made it difficult to expand the input language models can receive. Li Lyna Zhang introduces LongRoPE, a method capable of extending content windows to more than 2 million tokens.
Senior Researcher Arindam Mitra introduces AgentInstruct. Using raw data sources, the automated multi-agent framework can create diverse, high-quality synthetic data at scale for the post-training of small and large language models.
Andrey Kolobov discusses WindSeer, a small CNN capable of estimating the wind field around an sUAV in flight more finely and with less compute and data than traditional models. The advancement can help support longer and safer autonomous flights.
