Numerous algorithms have been proposed to $\textit{align}$ language models to remove undesirable behaviors. However, the challenges associated with a very large state space and creating a proper reward function often result in various jailbreaks. Our paper aims to examine this effect of reward in the controlled setting of positive sentiment language generation. Instead of online training of a reward model based on human feedback, we employ a statically learned sentiment classifier. We also consider a setting where our model's weights and activations are exposed to an end-user after training. We examine a pretrained GPT-2 through the lens of mechanistic interpretability before and after proximal policy optimization (PPO) has been applied to promote positive sentiment responses. Using these insights, we (1) attempt to "hack" the PPO-ed model to generate negative sentiment responses and (2) add a term to the reward function to try and alter `negative' weights.

Overparametrized transformer networks are the state-of-the-art architecture for Large Language Models (LLMs). However, such models contain billions of parameters making large compute a necessity, while raising environmental concerns. To address these issues, we propose FinerCut, a new form of fine-grained layer pruning, which in contrast to prior work at the transformer block level, considers all self-attention and feed-forward network (FFN) layers within blocks as individual pruning candidates. FinerCut prunes layers whose removal causes minimal alternation to the model's output -- contributing to a new, lean, interpretable, and task-agnostic pruning method. Tested across 9 benchmarks, our approach retains 90% performance of Llama3-8B with 25% layers removed, and 95% performance of Llama3-70B with 30% layers removed, all without fine-tuning or post-pruning reconstruction. Strikingly, we observe intriguing results with FinerCut: 42% (34 out of 80) of the self-attention layers in Llama3-70B can be removed while preserving 99% of its performance -- without additional fine-tuning after removal. Moreover, FinerCut provides a tool to inspect the types and locations of pruned layers, allowing to observe interesting pruning behaviors. For instance, we observe a preference for pruning self-attention layers, often at deeper consecutive decoder layers. We hope our insights inspire future efficient LLM architecture designs.

Although humans inherently have diverse values, current large language model (LLM) alignment methods often assume that aligning LLMs with the general public's preferences is optimal. A major challenge in adopting a more individualized approach to LLM alignment is its lack of scalability, as it involves repeatedly acquiring preference data and training new reward models and LLMs for each individual's preferences. To address these challenges, we propose a new paradigm where users specify what they value most within the system message, steering the LLM's generation behavior to better align with the user's intentions. However, a naive application of such an approach is non-trivial since LLMs are typically trained on a uniform system message (e.g., "You are a helpful assistant") which limits their ability to generalize to diverse, unseen system messages. To improve this generalization, we create the Multifaceted Collection, a preference dataset with 192k combinations of values beyond generic helpfulness and harmlessness, spanning 65k user instructions. Using this dataset, we train a 7B LLM called Janus and test it on 921 prompts from 5 benchmarks (AlpacaEval 2.0, FLASK, Koala, MT-Bench, and Self-Instruct) by adding various unseen system messages that reflect user preferences. Janus achieves tie+win rate of 75.2%, 72.4%, and 66.4% against Mistral 7B Instruct v0.2, GPT-3.5 Turbo, and GPT-4, respectively. Unexpectedly, on three benchmarks focused on response helpfulness (AlpacaEval 2.0, MT-Bench, Arena Hard Auto v0.1), Janus also outperforms LLaMA 3 8B Instruct by a +4.0%, +0.1%, +3.0% margin, underscoring that training with a vast array of system messages could also enhance alignment to the general public's preference as well. Our code, dataset, benchmark, and models are available at https://github.com/kaistAI/Janus.

Recent advancements in generative artificial intelligence (AI) have transformed collaborative work processes, yet the impact on team performance remains underexplored. Here we examine the role of generative AI in enhancing or replacing traditional team dynamics using a randomized controlled experiment with 435 participants across 122 teams. We show that teams augmented with generative AI significantly outperformed those relying solely on human collaboration across various performance measures. Interestingly, teams with multiple AIs did not exhibit further gains, indicating diminishing returns with increased AI integration. Our analysis suggests that centralized AI usage by a few team members is more effective than distributed engagement. Additionally, individual-AI pairs matched the performance of conventional teams, suggesting a reduced need for traditional team structures in some contexts. However, despite this capability, individual-AI pairs still fell short of the performance levels achieved by AI-assisted teams. These findings underscore that while generative AI can replace some traditional team functions, more comprehensively integrating AI within team structures provides superior benefits, enhancing overall effectiveness beyond individual efforts.

In class-incremental learning, an agent with limited resources needs to learn a sequence of classification tasks, forming an ever growing classification problem, with the constraint of not being able to access data from previous tasks. The main difference with task-incremental learning, where a task-ID is available at inference time, is that the learner also needs to perform cross-task discrimination, i.e. distinguish between classes that have not been seen together. Approaches to tackle this problem are numerous and mostly make use of an external memory (buffer) of non-negligible size. In this paper, we ablate the learning of cross-task features and study its influence on the performance of basic replay strategies used for class-IL. We also define a new forgetting measure for class-incremental learning, and see that forgetting is not the principal cause of low performance. Our experimental results show that future algorithms for class-incremental learning should not only prevent forgetting, but also aim to improve the quality of the cross-task features, and the knowledge transfer between tasks. This is especially important when tasks contain limited amount of data.

Online deep learning solves the problem of learning from streams of data, reconciling two opposing objectives: learn fast and learn deep. Existing work focuses almost exclusively on exploring pure deep learning solutions, which are much better suited to handle the "deep" than the "fast" part of the online learning equation. In our work, we propose a different paradigm, based on a hybrid multilearner approach. First, we develop a fast online logistic regression learner. This learner does not rely on backpropagation. Instead, it uses closed form recursive updates of model parameters, handling the fast learning part of the online learning problem. We then analyze the existing online deep learning theory and show that the widespread ODL approach, currently operating at complexity $O(L^2)$ in terms of the number of layers $L$, can be equivalently implemented in $O(L)$ complexity. This further leads us to the cascaded multilearner design, in which multiple shallow and deep learners are co-trained to solve the online learning problem in a cooperative, synergistic fashion. We show that this approach achieves state-of-the-art results on common online learning datasets, while also being able to handle missing features gracefully. Our code is publicly available at https://github.com/AntonValk/MODL.

Recommender systems are essential tools in the digital era, providing personalized content to users in areas like e-commerce, entertainment, and social media. Among the many approaches developed to create these systems, latent factor models have proven particularly effective. This survey systematically reviews latent factor models in recommender systems, focusing on their core principles, methodologies, and recent advancements. The literature is examined through a structured framework covering learning data, model architecture, learning strategies, and optimization techniques. The analysis includes a taxonomy of contributions and detailed discussions on the types of learning data used, such as implicit feedback, trust, and content data, various models such as probabilistic, nonlinear, and neural models, and an exploration of diverse learning strategies like online learning, transfer learning, and active learning. Furthermore, the survey addresses the optimization strategies used to train latent factor models, improving their performance and scalability. By identifying trends, gaps, and potential research directions, this survey aims to provide valuable insights for researchers and practitioners looking to advance the field of recommender systems.

Broad, open source availability of large pretrained foundation models on the internet through platforms such as HuggingFace has taken the world of practical deep learning by storm. A classical pipeline for neural network training now typically consists of finetuning these pretrained network on a small target dataset instead of training from scratch. In the case of large models this can be done even on modest hardware using a low rank training technique known as Low-Rank Adaptation (LoRA). While Low Rank training has already been studied in the continual learning setting, existing works often consider storing the learned adapter along with the existing model but rarely attempt to modify the weights of the pretrained model by merging the LoRA with the existing weights after finishing the training of each task. In this article we investigate this setting and study the impact of LoRA rank on the forgetting of the pretraining foundation task and on the plasticity and forgetting of subsequent ones. We observe that this rank has an important impact on forgetting of both the pretraining and downstream tasks. We also observe that vision transformers finetuned in that way exhibit a sort of "contextual" forgetting, a behaviour that we do not observe for residual networks and that we believe has not been observed yet in previous continual learning works. The capability of machine learning to automatically learn from data and extract discriminative patters patterns has led to groundbreaking advances in fields such as computer vision, natural language processing, and speech recognition Le-Cun et al. (2015). Over the years, the increase in scale of deep models has been instrumental in driving progress.

Wiki articles are created and maintained by a crowd of editors, producing a continuous stream of reviews. Reviews can take the form of additions, reverts, or both. This crowdsourcing model is exposed to manipulation since neither reviews nor editors are automatically screened and purged. To protect articles against vandalism or damage, the stream of reviews can be mined to classify reviews and profile editors in real-time. The goal of this work is to anticipate and explain which reviews to revert. This way, editors are informed why their edits will be reverted. The proposed method employs stream-based processing, updating the profiling and classification models on each incoming event. The profiling uses side and content-based features employing Natural Language Processing, and editor profiles are incrementally updated based on their reviews. Since the proposed method relies on self-explainable classification algorithms, it is possible to understand why a review has been classified as a revert or a non-revert. In addition, this work contributes an algorithm for generating synthetic data for class balancing, making the final classification fairer. The proposed online method was tested with a real data set from Wikivoyage, which was balanced through the aforementioned synthetic data generation. The results attained near-90 % values for all evaluation metrics (accuracy, precision, recall, and F-measure).

Recently, foundation language models (LMs) have marked significant achievements in the domains of natural language processing (NLP) and computer vision (CV). Unlike traditional neural network models, foundation LMs obtain a great ability for transfer learning by acquiring rich commonsense knowledge through pre-training on extensive unsupervised datasets with a vast number of parameters. However, they still can not emulate human-like continuous learning due to catastrophic forgetting. Consequently, various continual learning (CL)-based methodologies have been developed to refine LMs, enabling them to adapt to new tasks without forgetting previous knowledge. However, a systematic taxonomy of existing approaches and a comparison of their performance are still lacking, which is the gap that our survey aims to fill. We delve into a comprehensive review, summarization, and classification of the existing literature on CL-based approaches applied to foundation language models, such as pre-trained language models (PLMs), large language models (LLMs) and vision-language models (VLMs). We divide these studies into offline CL and online CL, which consist of traditional methods, parameter-efficient-based methods, instruction tuning-based methods and continual pre-training methods. Offline CL encompasses domain-incremental learning, task-incremental learning, and class-incremental learning, while online CL is subdivided into hard task boundary and blurry task boundary settings. Additionally, we outline the typical datasets and metrics employed in CL research and provide a detailed analysis of the challenges and future work for LMs-based continual learning.