Evaluation of large language models (LLMs) has raised great concerns in the community due to the issue of data contamination. Existing work designed evaluation protocols using well-defined algorithms for specific tasks, which cannot be easily extended to diverse scenarios. Moreover, current evaluation benchmarks can only provide the overall benchmark results and cannot support a fine-grained and multifaceted analysis of LLMs' abilities. In this paper, we propose meta probing agents (MPA), a general dynamic evaluation protocol inspired by psychometrics to evaluate LLMs. MPA is the key component of DyVal 2, which naturally extends the previous DyVal~\citep{zhu2023dyval}. MPA designs the probing and judging agents to automatically transform an original evaluation problem into a new one following psychometric theory on three basic cognitive abilities: language understanding, problem solving, and domain knowledge. These basic abilities are also dynamically configurable, allowing multifaceted analysis. We conducted extensive evaluations using MPA and found that most LLMs achieve poorer performance, indicating room for improvement. Our multifaceted analysis demonstrated the strong correlation between the basic abilities and an implicit Matthew effect on model size, i.e., larger models possess stronger correlations of the abilities. MPA can also be used as a data augmentation approach to enhance LLMs. Code is available at: https://github.com/microsoft/promptbench.

Relying on multi-modal observations, embodied robots could perform multiple robotic manipulation tasks in unstructured real-world environments. However, most language-conditioned behavior-cloning agents still face existing long-standing challenges, i.e., 3D scene representation and human-level task learning, when adapting into new sequential tasks in practical scenarios. We here investigate these above challenges with NBAgent in embodied robots, a pioneering language-conditioned Never-ending Behavior-cloning Agent. It can continually learn observation knowledge of novel 3D scene semantics and robot manipulation skills from skill-shared and skill-specific attributes, respectively. Specifically, we propose a skill-sharedsemantic rendering module and a skill-shared representation distillation module to effectively learn 3D scene semantics from skill-shared attribute, further tackling 3D scene representation overlooking. Meanwhile, we establish a skill-specific evolving planner to perform manipulation knowledge decoupling, which can continually embed novel skill-specific knowledge like human from latent and low-rank space. Finally, we design a never-ending embodied robot manipulation benchmark, and expensive experiments demonstrate the significant performance of our method. Visual results, code, and dataset are provided at: https://neragent.github.io.

Multiple-choice question answering (MCQA) is often used to evaluate large language models (LLMs). To see if MCQA assesses LLMs as intended, we probe if LLMs can perform MCQA with choices-only prompts, where models must select the correct answer only from the choices. In three MCQA datasets and four LLMs, this prompt bests a majority baseline in 11/12 cases, with up to 0.33 accuracy gain. To help explain this behavior, we conduct an in-depth, black-box analysis on memorization, choice dynamics, and question inference. Our key findings are threefold. First, we find no evidence that the choices-only accuracy stems from memorization alone. Second, priors over individual choices do not fully explain choices-only accuracy, hinting that LLMs use the group dynamics of choices. Third, LLMs have some ability to infer a relevant question from choices, and surprisingly can sometimes even match the original question. Inferring the original question is an impressive reasoning strategy, but it cannot fully explain the high choices-only accuracy of LLMs in MCQA. Thus, while LLMs are not fully incapable of reasoning in MCQA, we still advocate for the use of stronger baselines in MCQA benchmarks, the design of robust MCQA datasets for fair evaluations, and further efforts to explain LLM decision-making.

Recent works have shown the benefits to LLMs from fine-tuning golden-standard Chain-of-Thought (CoT) rationales or using them as correct examples in few-shot prompting. While humans can indeed imitate correct examples, learning from our mistakes is another vital aspect of human cognition. Hence, a question naturally arises: \textit{can LLMs learn and benefit from their mistakes, especially for their reasoning? } This study investigates this problem from both the prompting and model-tuning perspectives. We begin by introducing \textsc{CoTErrorSet}, a new benchmark with 609,432 questions, each designed with both correct and error references, and demonstrating the types and reasons for making such mistakes. To explore the effectiveness of those mistakes, we design two methods: (1) \textbf{Self-rethinking} prompting guides LLMs to rethink whether they have made similar previous mistakes; and (2) \textbf{Mistake tuning} involves finetuning models in both correct and incorrect reasoning domains, rather than only tuning models to learn ground truth in traditional methodology. We conduct a series of experiments to prove LLMs can obtain benefits from mistakes in both directions. Our two methods offer potentially cost-effective strategies by leveraging errors to enhance reasoning capabilities, which costs significantly less than creating meticulously hand-crafted golden references. We ultimately make a thorough analysis of the reasons behind LLMs' errors, which provides directions that future research needs to overcome. \textsc{CoTErrorSet} will be published soon on \texttt{\url{https://github.com/YookiTong/Learn-from-Mistakes-CotErrorSet}}.

The evolution of video generation from text, starting with animating MNIST numbers to simulating the physical world with Sora, has progressed at a breakneck speed over the past seven years. While often seen as a superficial expansion of the predecessor text-to-image generation model, text-to-video generation models are developed upon carefully engineered constituents. Here, we systematically discuss these elements consisting of but not limited to core building blocks (vision, language, and temporal) and supporting features from the perspective of their contributions to achieving a world model. We employ the PRISMA framework to curate 97 impactful research articles from renowned scientific databases primarily studying video synthesis using text conditions. Upon minute exploration of these manuscripts, we observe that text-to-video generation involves more intricate technologies beyond the plain extension of text-to-image generation. Our additional review into the shortcomings of Sora-generated videos pinpoints the call for more in-depth studies in various enabling aspects of video generation such as dataset, evaluation metric, efficient architecture, and human-controlled generation. Finally, we conclude that the study of the text-to-video generation may still be in its infancy, requiring contribution from the cross-discipline research community towards its advancement as the first step to realize artificial general intelligence (AGI).

Large language models (LLMs) have showcased impressive abilities across various natural language tasks. Efficiently training these massive models remains a major challenge, demanding extensive computational resources and time. To address this issue, a growing body of research has focused on developing techniques to minimize the memory footprint and accelerate LLM training without sacrificing performance. We present Spectrum, a method for selectively training the layers of an LLM based on their signal-tonoise ratio (SNR). Rooted in Random Matrix Theory, Spectrum utilizes the Marchenko-Pastur distribution to pinpoint informative layers according to their SNR. Unlike previous approaches, such as QLoRA [2], which quantizes the entire model, Spectrum strategically targets specific layers and modules for training while keeping others frozen. By concentrating computational resources on the most informative parameters, Spectrum achieves superior performance while significantly reducing training time and memory requirements compared to state-of-the-art methods. Our main contributions are as follows: We propose Spectrum, an approach for efficient LLM training that selectively trains layers based on their SNR.

Reinforcement Learning from Human Feedback (RLHF) (Christiano et al., 2017) has rapidly become a standard method to align Large Language Models (LLMs). One of the main practical issues that all the prominent existing RLHF methods (offline or online) (Ouyang et al., 2022; Rafailov et al., 2023; Azar et al., 2023; Zhao et al., 2023b; Ahmadian et al., 2024) encounter is that their optimal solution heavily depends on the training task in terms of the distribution used to generate the preference data (behavior policy) (Munos et al., 2023; Azar et al., 2023). This makes the existing RLHF methods prone to out-of-distribution (OOD) tasks (Li et al., 2024; Kirk et al., 2024) where the evaluation distribution is significantly different from that of the behavior policy. Also, whenever the base/SFT models significantly differ from the behavior policy, the dependency of the RLHF solutions on the behavior policy makes the preference dataset and reward model less useful (Gao et al., 2022) as RLHF may undo the SFT/pretraining. To address this challenge, we introduce an alternative approach for aligning LLMs from human preferences based on more principled and robust foundations. Our goal is to find a solution that is robust to the changes in the preference dataset, meaning that changes in the distribution from which the completions are sampled do not affect the final outcome of learning significantly. To achieve this goal, we exploit the concept of self-improving (Huang et al., 2022; Bai et al., 2022) language models. By self-improving LLM we refer to a model capable of enhancing its outputs recursively with each inference iteration.

Most popular benchmarks for comparing LLMs rely on a limited set of prompt templates, which may not fully capture the LLMs' abilities and can affect the reproducibility of results on leaderboards. Many recent works empirically verify prompt sensitivity and advocate for changes in LLM evaluation. In this paper, we consider the problem of estimating the performance distribution across many prompt variants instead of finding a single prompt to evaluate with. We introduce PromptEval, a method for estimating performance across a large set of prompts borrowing strength across prompts and examples to produce accurate estimates under practical evaluation budgets. The resulting distribution can be used to obtain performance quantiles to construct various robust performance metrics (e.g., top 95% quantile or median). We prove that PromptEval consistently estimates the performance distribution and demonstrate its efficacy empirically on three prominent LLM benchmarks: MMLU, BIG-bench Hard, and LMentry. For example, PromptEval can accurately estimate performance quantiles across 100 prompt templates on MMLU with a budget equivalent to two single-prompt evaluations. Our code and data can be found at https://github.com/felipemaiapolo/prompt-eval.

We study the problem of online multi-group learning, a learning model in which an online learner must simultaneously achieve small prediction regret on a large collection of (possibly overlapping) subsequences corresponding to a family of groups. Groups are subsets of the context space, and in fairness applications, they may correspond to subpopulations defined by expressive functions of demographic attributes. In contrast to previous work on this learning model, we consider scenarios in which the family of groups is too large to explicitly enumerate, and hence we seek algorithms that only access groups via an optimization oracle. In this paper, we design such oracle-efficient algorithms with sublinear regret under a variety of settings, including: (i) the i.i.d. setting, (ii) the adversarial setting with smoothed context distributions, and (iii) the adversarial transductive setting.

For years, Chromebooks have served as the loyal opposition to PCs. Google's laptops offer many of the same Google services as you can find via the Web, but integrated into an inexpensive package for consumers and students. I sat down with John Solomon, vice president of ChromeOS and education at Google, to ask about the new wave of AI PCs and how Google responds. We talk about how "generic" Chromebooks survive as Google pushes Chromebook Plus, how kids can be encouraged to game on Chromebooks as well as learn, and what Google is cooking up in response to Microsoft's Recall for Copilot PCs. This interview has been lightly edited for length and clarity. Mark Hachman, PCWorld: I saw your presentation at Computex as a way to remind people that there are more than just AI PCs. So, in light of those products, what is the value proposition of a Chromebook these days? John Solomon, Google: As you know, we have Chromebook and Chromebook Plus. In Chromebook, it has always been about and continues to be about delivering really great value, the best place to experience access to Google services. Whether it's Google Workspace, or more broadly, Chrome, we work very hard to make sure that first-party products as well as the Play Store work well on Chromebook.