We introduce Modelizer - a novel framework that, given a black-box program, learns a _model from its input/output behavior_ using _neural machine translation_. The resulting model _mocks_ the original program: Given an input, the model predicts the output that would have been produced by the program. However, the model is also _reversible_ - that is, the model can predict the input that would have produced a given output. Finally, the model is _differentiable_ and can be efficiently restricted to predict only a certain aspect of the program behavior. Modelizer uses _grammars_ to synthesize inputs and to parse the resulting outputs, allowing it to learn sequence-to-sequence associations between token streams. Other than input and output grammars, Modelizer only requires the ability to execute the program. The resulting models are _small_, requiring fewer than 6.3 million parameters for languages such as Markdown or HTML; and they are _accurate_, achieving up to 95.4% accuracy and a BLEU score of 0.98 with standard error 0.04 in mocking real-world applications. We foresee several _applications_ of these models, especially as the output of the program can be any aspect of program behavior. Besides mocking and predicting program behavior, the model can also synthesize inputs that are likely to produce a particular behavior, such as failures or coverage.

While extensive research has focused on ChatGPT in recent years, very few studies have systematically quantified and compared linguistic features between human-written and Artificial Intelligence (AI)-generated language. This study aims to investigate how various linguistic components are represented in both types of texts, assessing the ability of AI to emulate human writing. Using human-authored essays as a benchmark, we prompted ChatGPT to generate essays of equivalent length. These texts were analyzed using Open Brain AI, an online computational tool, to extract measures of phonological, morphological, syntactic, and lexical constituents. Despite AI-generated texts appearing to mimic human speech, the results revealed significant differences across multiple linguistic features such as consonants, word stress, nouns, verbs, pronouns, direct objects, prepositional modifiers, and use of difficult words among others. These findings underscore the importance of integrating automated tools for efficient language assessment, reducing time and effort in data analysis. Moreover, they emphasize the necessity for enhanced training methodologies to improve the capacity of AI for producing more human-like text.

A recent line of MLLMs work[1, 2] has started to explore introducing video modality. On the one hand, video contains more information than images because of the time-dimensional modelling capability, and video is much better in capturing events. On the other hand video cameras are everywhere around us as a common and low-cost electronic surveillance tool. However, how to transform advanced academic results or models into beneficial and practical products is a subject of much greater concern to industry. According to business requirement of the company, our team will focus on a challenging task, video understanding of billiards indoor scenes. We propose HiLight, a video chat model. HiLight is constructed by the two stages, each of which is required to accomplish respective target performance: 1. Video and Text Modality Alignment Stage: This stage requires aligning the video modality with the text modality as well as a strong ability to detecting small objects.

The recent development of large language models (LLMs) has spurred discussions about whether LLM-generated "synthetic samples" could complement or replace traditional surveys, considering their training data potentially reflects attitudes and behaviors prevalent in the population. A number of mostly US-based studies have prompted LLMs to mimic survey respondents, with some of them finding that the responses closely match the survey data. However, several contextual factors related to the relationship between the respective target population and LLM training data might affect the generalizability of such findings. In this study, we investigate the extent to which LLMs can estimate public opinion in Germany, using the example of vote choice. We generate a synthetic sample of personas matching the individual characteristics of the 2017 German Longitudinal Election Study respondents. We ask the LLM GPT-3.5 to predict each respondent's vote choice and compare these predictions to the survey-based estimates on the aggregate and subgroup levels. We find that GPT-3.5 does not predict citizens' vote choice accurately, exhibiting a bias towards the Green and Left parties. While the LLM captures the tendencies of "typical" voter subgroups, such as partisans, it misses the multifaceted factors swaying individual voter choices. By examining the LLM-based prediction of voting behavior in a new context, our study contributes to the growing body of research about the conditions under which LLMs can be leveraged for studying public opinion. The findings point to disparities in opinion representation in LLMs and underscore the limitations in applying them for public opinion estimation.

Machine learning models often make predictions based on biased features such as gender, race, and other social attributes, posing significant fairness risks, especially in societal applications, such as hiring, banking, and criminal justice. Traditional approaches to addressing this issue involve retraining or fine-tuning neural networks with fairness-aware optimization objectives. However, these methods can be impractical due to significant computational resources, complex industrial tests, and the associated CO2 footprint. Additionally, regular users often fail to fine-tune models because they lack access to model parameters In this paper, we introduce the Inference-Time Rule Eraser (Eraser), a novel method designed to address fairness concerns by removing biased decision-making rules from deployed models during inference without altering model weights. We begin by establishing a theoretical foundation for modifying model outputs to eliminate biased rules through Bayesian analysis. Next, we present a specific implementation of Eraser that involves two stages: (1) distilling the biased rules from the deployed model into an additional patch model, and (2) removing these biased rules from the output of the deployed model during inference. Extensive experiments validate the effectiveness of our approach, showcasing its superior performance in addressing fairness concerns in AI systems.

We present TMMLU+, a new benchmark designed for Traditional Chinese language understanding. TMMLU+ is a multi-choice question-answering dataset with 66 subjects from elementary to professional level. It is six times larger and boasts a more balanced subject distribution than its predecessor, Taiwan Massive Multitask Language Understanding (TMMLU). We also benchmark closed-source models and 26 open-weight Chinese large language models (LLMs) of parameters ranging from 1.8B to 72B on the proposed TMMLU+. Our findings reveal that (1.) Traditional Chinese models still trail behind their Simplified Chinese counterparts, highlighting a need for more focused advancements in LLMs catering to Traditional Chinese. (2.) Current LLMs still fall short of human performance in average scores, indicating a potential need for future research to delve deeper into social science and humanities subjects. (3.) Among all the tokenization compression metrics examined, we identify that only the fertility score uniquely demonstrates strong correlations with our benchmark results. We foresee that TMMLU+ will pinpoint areas for future model improvement, thereby narrowing the gap between machine and human linguistic capabilities and supporting researchers in developing Traditional Chinese LLMs. Our dataset, along with the benchmark source code, is accessible at huggingface.co/datasets/ikala/tmmluplus.

Evaluation is critical for assessing capabilities, tracking scientific progress, and informing model selection. In this paper, we present three desiderata for a good benchmark for language models: (i) salience (e.g., knowledge about World War II is more salient than a random day in history), (ii) novelty (i.e., the benchmark reveals new trends in model rankings not shown by previous benchmarks), and (iii) difficulty (i.e., the benchmark should be difficult for existing models, leaving headroom for future improvement). We operationalize these three desiderata and cast benchmark creation as a search problem, that of finding benchmarks that that satisfy all three desiderata. To tackle this search problem, we present AutoBencher, which uses a language model to automatically search for datasets that meet the three desiderata. AutoBencher uses privileged information (e.g. relevant documents) to construct reliable datasets, and adaptivity with reranking to optimize for the search objective. We use AutoBencher to create datasets for math, multilingual, and knowledge-intensive question answering. The scalability of AutoBencher allows it to test fine-grained categories and tail knowledge, creating datasets that are on average 27% more novel and 22% more difficult than existing benchmarks. A closer investigation of our constructed datasets shows that we can identify specific gaps in LM knowledge in language models that are not captured by existing benchmarks, such as Gemini Pro performing much worse on question answering about the Permian Extinction and Fordism, while OpenAGI-7B performing surprisingly well on QA about COVID-19.

Instruction following is one of the fundamental capabilities of large language models (LLMs). As the ability of LLMs is constantly improving, they have been increasingly applied to deal with complex human instructions in real-world scenarios. Therefore, how to evaluate the ability of complex instruction-following of LLMs has become a critical research problem. Existing benchmarks mainly focus on modeling different types of constraints in human instructions while neglecting the composition of different constraints, which is an indispensable constituent in complex instructions. To this end, we propose ComplexBench, a benchmark for comprehensively evaluating the ability of LLMs to follow complex instructions composed of multiple constraints. We propose a hierarchical taxonomy for complex instructions, including 4 constraint types, 19 constraint dimensions, and 4 composition types, and manually collect a high-quality dataset accordingly. To make the evaluation reliable, we augment LLM-based evaluators with rules to effectively verify whether generated texts can satisfy each constraint and composition. Furthermore, we obtain the final evaluation score based on the dependency structure determined by different composition types. ComplexBench identifies significant deficiencies in existing LLMs when dealing with complex instructions with multiple constraints composition.

Pre-trained large language models (LLMs) can now be easily adapted for specific business purposes using custom prompts or fine tuning. These customizations are often iteratively re-engineered to improve some aspect of performance, but after each change businesses want to ensure that there has been no negative impact on the system's behavior around such critical issues as bias. Prior methods of benchmarking bias use techniques such as word masking and multiple choice questions to assess bias at scale, but these do not capture all of the nuanced types of bias that can occur in free response answers, the types of answers typically generated by LLM systems. In this paper, we identify several kinds of nuanced bias in free text that cannot be similarly identified by multiple choice tests. We describe these as: confidence bias, implied bias, inclusion bias and erasure bias. We present a semi-automated pipeline for detecting these types of bias by first eliminating answers that can be automatically classified as unbiased and then co-evaluating name reversed pairs using crowd workers. We believe that the nuanced classifications our method generates can be used to give better feedback to LLMs, especially as LLM reasoning capabilities become more advanced.

Language models achieve impressive results in tasks involving complex multistep reasoning, but scaling these capabilities further traditionally requires expensive collection of more annotated data. In this work, we explore the potential of improving the capabilities of language models without new data, merely using automated feedback to the validity of their predictions in arithmetic reasoning (self-training). We find that models can substantially improve in both single-round (offline) and online self-training. In the offline setting, supervised methods are able to deliver gains comparable to preference optimization, but in online self-training, preference optimization shows to largely outperform supervised training thanks to superior stability and robustness on unseen types of problems.