Safety and trustworthiness are indispensable requirements for real-world applications of AI systems using large language models (LLMs).

Ouyang et al., 2022, Bai et al., 2022], achieves this by learning a reward model on human preference

Why is this the case? One possibility is that some aspect of the transformer architecture makes these behaviors easy to learn. Under this hypothesis, transformer models do not implement any useful functionality when initialized; however, their loss landscape is structured such that they can be (computation-and sample-) efficiently optimized for behaviors of interest.

Can incorporating context into self-supervised vision algorithms eliminate augmentation-based inductive priors and enable dynamic adaptation to varying task symmetries? This work suggests a positive answer to this question by proposing to enhance the current joint embedding architecture with a finite context -- an abstract representation of a task, containing a few demonstrations that inform about task-specific symmetries, as shown in Figure 2(c).

Given a piece of subject text, many existing detection methods work by measuring the difficulty of LLM predicting the next token in the text from their prefix.

We consider the learning of first-layer embeddings / pre-activations in vision models, and in particular learning the weights with which the input images are transformed in order to obtain these embeddings.

Most widely adopted MoE models are discontinuous with respect to their parameters - often referred to as sparse .

Beginning with the simplification of flattening all actions, we theoretically explore the discrepancies between action-level optimization and this naive token-level optimization.

In light of this, we introduces a novel research task, M ulti-modal S entiment A nalysis for C omment R esponse of V ideo I nduced( MSA-CRVI), aims to infer opinions and emotions according to comments response to micro video.

In mechanical structures like airplanes, cars and houses, noise is generated and transmitted through vibrations. To take measures to reduce this noise, vibrations need to be simulated with expensive numerical computations. Deep learning surrogate models present a promising alternative to classical numerical simulations as they can be evaluated magnitudes faster, while trading-off accuracy. To quantify such trade-offs systematically and foster the development of methods, we present a benchmark on the task of predicting the vibration of harmonically excited plates. The benchmark features a total of 12,000 plate geometries with varying forms of beadings, material, boundary conditions, load position and sizes with associated numerical solutions. To address the benchmark task, we propose a new network architecture, named Frequency-Query Operator, which predicts vibration patterns of plate geometries given a specific excitation frequency. Applying principles from operator learning and implicit models for shape encoding, our approach effectively addresses the prediction of highly variable frequency response functions occurring in dynamic systems. To quantify the prediction quality, we introduce a set of evaluation metrics and evaluate the method on our vibrating-plates benchmark. Our method outperforms Deep-ONets, Fourier Neural Operators and more traditional neural network architectures and can be used for design optimization.