Stay updated with the latest developments in Artificial Intelligence
Does a language model answer better because its compressed memory contains useful document information? Replacing memory vectors with zeros can lower a trained model's score without showing that it outperforms a model trained without memory.
Science fiction is widely credited with unsettling gender. This article asks whether that holds at the linguistic level, in how readily gendered reference gives way to its counterpart, a neutral term, or other social category, and whether it separates the genre from general fiction.
LLM workflows and agents rely on a set of binary decisions like allow/block an input, or discard/retain a retrieved document. When human deferral is unavailable, uncertain cases must still receive an action.
W.W. Norton this month published Eric So’s “The Collision,” which argues broad AI reliance can cause cognitive atrophy by weakening learning, memory and problem solving habits.
The Practical Molecular Optimization (PMO) benchmark standardized evaluation in molecular optimization, but it is built on simple property-based oracles that do not capture structure-based drug design.
Rare but severe failures of large language models can become practically significant at deployment scale. Estimating their probabilities with direct Monte Carlo sampling is highly inefficient, while proposal-based methods typically require auxiliary learning of an additional proposal distribution.
Automatic prompt optimization (APO) uses feedback from model responses to improve prompts automatically. However, its effectiveness is usually measured through overall scores, which do not reveal which questions become correct, which previously correct answers become wrong, or whether improvements…
This work concerns our submission for the Multilingual track of the 2026 BabyLM Challenge. We improve on our baseline using data that has been summarized and translated by an LLM, showing that higher-quality data can come from synthetic sources.
Harness optimization improves LLMs by refining reusable prompts and scaffolding, yet requires further adaptation for program generation tasks such as text-to-SQL and text-to-Python.
Sparse attention reduces the KV-cache traffic that dominates long-context LLM decoding, but practical methods largely operate at coarse granularity. Token-level selection offers finer control, yet presents two challenges: low-dimensional scoring may lose query--key structure needed for accurate…
Pluralistic alignment needs to accommodate diverse preferences under limited human feedback. Using AI simulators to label preference comparisons reduces annotation costs, but variation in simulation accuracy across individuals can leave some people's preferences poorly represented in training.
One-step diffusion distillation achieves strong performance on small datasets but remains insufficient for high-resolution or complex text-to-image settings.
Multimodal in-context learning (ICL) is an important capability of multimodal large language models (MLLMs). Its performance is closely tied to the demonstrations it uses, making demonstration retrieval a core problem for ICL.
Chatbot Arena ranks language models with the Bradley-Terry model, which gives each model one score and so assumes Luce's choice axiom. Choice theory predicts where this fails: near-copies are told apart more decisively than their scores imply.
Compartment models of infectious diseases are often linear in coefficients that are physical parameters or products or ratios of them. We exploit this with closed-form least squares and equip the estimates with distribution-free uncertainty.
Fairness audits with several sensitive attributes are often performed one coordinate at a time. This can miss purely intersectional disparities: a score may be independent of each sensitive coordinate while depending on their joint profile.
Understanding the dynamics of language acquisition in Large Language Models (LLMs) is crucial for uncovering how statistical representations evolve into structured linguistic capabilities.
Post-training reasoning LLMs often requires making a tradeoff between costly supervision and compute inefficiency: supervised finetuning typically requires reasoning traces, which is data inefficient since in many applications one only has prompt-answer tuples and lacks the groundtruth…
Showing 144 of 31,467 stories
Load more storiesGet the top AI news and tools straight to your inbox every week.
Weekly · AI news
Join 15,000+ AI enthusiasts
