PyTorchKR Papers

최신 논문

총 857편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2406.06560공식 코드 · Papers with Code

Inverse Constitutional AI: Compressing Preferences into Principles

Feedback data is widely used for fine-tuning and evaluating state-of-the-art AI models. Pairwise text preferences, where human or AI annotators select the "better" of two options, are particularly common. Such preferences are used to train (reward) models or to rank models with aggregate statistics. For many applications it is desirable to understand annotator preferences in addition to modelling them - not least because extensive prior work has shown various unintended biases in preference datasets. Yet, preference datasets remain challenging to interpret. Neither black-box reward models nor statistics can answer why one text is preferred over another. Manual interpretation of the numerous (long) response pairs is usually equally infeasible. In this paper, we introduce the Inverse Constitutional AI (ICAI) problem, formulating the interpretation of pairwise text preference data as a compression task. In constitutional AI, a set of principles (a constitution) is used to provide feedback and fine-tune AI models. ICAI inverts this process: given a feedback dataset, we aim to extract a constitution that best enables a large language model (LLM) to reconstruct the original annotations. We propose a corresponding ICAI algorithm and validate its generated constitutions quantitatively based on annotation reconstruction accuracy on several datasets: (a) synthetic feedback data with known principles; (b) AlpacaEval cross-annotated human feedback data; (c) crowdsourced Chatbot Arena data; and (d) PRISM data from diverse demographic groups. As a short and interpretable representation of the original dataset, generated constitutions have many potential use cases: help identify undesirable annotator biases, understand model performance better, scale feedback to unseen data, or adapt models to individual user or group preferences. We release the source code at https://github.com/rdnfn/icai.

arXiv:2406.07259

Scientific Computing with Large Language Models

We provide an overview of the emergence of large language models for scientific computing applications. We highlight use cases that involve natural language processing of scientific documents and specialized languages designed to describe physical systems. For the former, chatbot style applications appear in medicine, mathematics and physics and can be used iteratively with domain experts for problem solving. We also review specialized languages within molecular biology, the languages of molecules, proteins, and DNA where language models are being used to predict properties and even create novel physical systems at much faster rates than traditional computing methods.

arXiv:2406.06369

Annotation alignment: Comparing LLM and human annotations of conversational safety

Do LLMs align with human perceptions of safety? We study this question via annotation alignment, the extent to which LLMs and humans agree when annotating the safety of user-chatbot conversations. We leverage the recent DICES dataset (Aroyo et al., 2023), in which 350 conversations are each rated for safety by 112 annotators spanning 10 race-gender groups. GPT-4 achieves a Pearson correlation of $r = 0.59$ with the average annotator rating, \textit{higher} than the median annotator's correlation with the average ($r=0.51$). We show that larger datasets are needed to resolve whether LLMs exhibit disparities in how well they correlate with different demographic groups. Also, there is substantial idiosyncratic variation in correlation within groups, suggesting that race & gender do not fully capture differences in alignment. Finally, we find that GPT-4 cannot predict when one demographic group finds a conversation more unsafe than another.

arXiv:2406.04770공식 코드 · Papers with Code

WildBench: Benchmarking LLMs with Challenging Tasks from Real Users in the Wild

We introduce WildBench, an automated evaluation framework designed to benchmark large language models (LLMs) using challenging, real-world user queries. WildBench consists of 1,024 tasks carefully selected from over one million human-chatbot conversation logs. For automated evaluation with WildBench, we have developed two metrics, WB-Reward and WB-Score, which are computable using advanced LLMs such as GPT-4-turbo. WildBench evaluation uses task-specific checklists to evaluate model outputs systematically and provides structured explanations that justify the scores and comparisons, resulting in more reliable and interpretable automatic judgments. WB-Reward employs fine-grained pairwise comparisons between model responses, generating five potential outcomes: much better, slightly better, slightly worse, much worse, or a tie. Unlike previous evaluations that employed a single baseline model, we selected three baseline models at varying performance levels to ensure a comprehensive pairwise evaluation. Additionally, we propose a simple method to mitigate length bias, by converting outcomes of ``slightly better/worse'' to ``tie'' if the winner response exceeds the loser one by more than $K$ characters. WB-Score evaluates the quality of model outputs individually, making it a fast and cost-efficient evaluation metric. WildBench results demonstrate a strong correlation with the human-voted Elo ratings from Chatbot Arena on hard tasks. Specifically, WB-Reward achieves a Pearson correlation of 0.98 with top-ranking models. Additionally, WB-Score reaches 0.95, surpassing both ArenaHard's 0.91 and AlpacaEval2.0's 0.89 for length-controlled win rates, as well as the 0.87 for regular win rates.

arXiv:2406.03510

Speech-based Clinical Depression Screening: An Empirical Study

This study investigates the utility of speech signals for AI-based depression screening across varied interaction scenarios, including psychiatric interviews, chatbot conversations, and text readings. Participants include depressed patients recruited from the outpatient clinics of Peking University Sixth Hospital and control group members from the community, all diagnosed by psychiatrists following standardized diagnostic protocols. We extracted acoustic and deep speech features from each participant's segmented recordings. Classifications were made using neural networks or SVMs, with aggregated clip outcomes determining final assessments. Our analysis across interaction scenarios, speech processing techniques, and feature types confirms speech as a crucial marker for depression screening. Specifically, human-computer interaction matches clinical interview efficacy, surpassing reading tasks. Segment duration and quantity significantly affect model performance, with deep speech features substantially outperforming traditional acoustic features.

arXiv:2406.03339

The Challenges of Evaluating LLM Applications: An Analysis of Automated, Human, and LLM-Based Approaches

Chatbots have been an interesting application of natural language generation since its inception. With novel transformer based Generative AI methods, building chatbots have become trivial. Chatbots which are targeted at specific domains for example medicine and psychology are implemented rapidly. This however, should not distract from the need to evaluate the chatbot responses. Especially because the natural language generation community does not entirely agree upon how to effectively evaluate such applications. With this work we discuss the issue further with the increasingly popular LLM based evaluations and how they correlate with human evaluations. Additionally, we introduce a comprehensive factored evaluation mechanism that can be utilized in conjunction with both human and LLM-based evaluations. We present the results of an experimental evaluation conducted using this scheme in one of our chatbot implementations which consumed educational reports, and subsequently compare automated, traditional human evaluation, factored human evaluation, and factored LLM evaluation. Results show that factor based evaluation produces better insights on which aspects need to be improved in LLM applications and further strengthens the argument to use human evaluation in critical spaces where main functionality is not direct retrieval.

arXiv:2406.01280공식 코드 · Papers with Code

Demo: Soccer Information Retrieval via Natural Queries using SoccerRAG

The rapid evolution of digital sports media necessitates sophisticated information retrieval systems that can efficiently parse extensive multimodal datasets. This paper demonstrates SoccerRAG, an innovative framework designed to harness the power of Retrieval Augmented Generation (RAG) and Large Language Models (LLMs) to extract soccer-related information through natural language queries. By leveraging a multimodal dataset, SoccerRAG supports dynamic querying and automatic data validation, enhancing user interaction and accessibility to sports archives. We present a novel interactive user interface (UI) based on the Chainlit framework which wraps around the core functionality, and enable users to interact with the SoccerRAG framework in a chatbot-like visual manner.

arXiv:2406.01428

Superhuman performance in urology board questions by an explainable large language model enabled for context integration of the European Association of Urology guidelines: the UroBot study

Large Language Models (LLMs) are revolutionizing medical Question-Answering (medQA) through extensive use of medical literature. However, their performance is often hampered by outdated training data and a lack of explainability, which limits clinical applicability. This study aimed to create and assess UroBot, a urology-specialized chatbot, by comparing it with state-of-the-art models and the performance of urologists on urological board questions, ensuring full clinician-verifiability. UroBot was developed using OpenAI's GPT-3.5, GPT-4, and GPT-4o models, employing retrieval-augmented generation (RAG) and the latest 2023 guidelines from the European Association of Urology (EAU). The evaluation included ten runs of 200 European Board of Urology (EBU) In-Service Assessment (ISA) questions, with performance assessed by the mean Rate of Correct Answers (RoCA). UroBot-4o achieved an average RoCA of 88.4%, surpassing GPT-4o by 10.8%, with a score of 77.6%. It was also clinician-verifiable and exhibited the highest run agreement as indicated by Fleiss' Kappa (k = 0.979). By comparison, the average performance of urologists on board questions, as reported in the literature, is 68.7%. UroBot's clinician-verifiable nature and superior accuracy compared to both existing models and urologists on board questions highlight its potential for clinical integration. The study also provides the necessary code and instructions for further development of UroBot.

arXiv:2405.20485

Phantom: General Trigger Attacks on Retrieval Augmented Language Generation

Retrieval Augmented Generation (RAG) expands the capabilities of modern large language models (LLMs), by anchoring, adapting, and personalizing their responses to the most relevant knowledge sources. It is particularly useful in chatbot applications, allowing developers to customize LLM output without expensive retraining. Despite their significant utility in various applications, RAG systems present new security risks. In this work, we propose new attack vectors that allow an adversary to inject a single malicious document into a RAG system's knowledge base, and mount a backdoor poisoning attack. We design Phantom, a general two-stage optimization framework against RAG systems, that crafts a malicious poisoned document leading to an integrity violation in the model's output. First, the document is constructed to be retrieved only when a specific trigger sequence of tokens appears in the victim's queries. Second, the document is further optimized with crafted adversarial text that induces various adversarial objectives on the LLM output, including refusal to answer, reputation damage, privacy violations, and harmful behaviors. We demonstrate our attacks on multiple LLM architectures, including Gemma, Vicuna, and Llama, and show that they transfer to GPT-3.5 Turbo and GPT-4. Finally, we successfully conducted a Phantom attack on NVIDIA's black-box production RAG system, "Chat with RTX".

arXiv:2405.20389공식 코드 · Papers with Code

Designing an Evaluation Framework for Large Language Models in Astronomy Research

Large Language Models (LLMs) are shifting how scientific research is done. It is imperative to understand how researchers interact with these models and how scientific sub-communities like astronomy might benefit from them. However, there is currently no standard for evaluating the use of LLMs in astronomy. Therefore, we present the experimental design for an evaluation study on how astronomy researchers interact with LLMs. We deploy a Slack chatbot that can answer queries from users via Retrieval-Augmented Generation (RAG); these responses are grounded in astronomy papers from arXiv. We record and anonymize user questions and chatbot answers, user upvotes and downvotes to LLM responses, user feedback to the LLM, and retrieved documents and similarity scores with the query. Our data collection method will enable future dynamic evaluations of LLM tools for astronomy.

arXiv:2405.20267공식 코드 · Papers with Code

Auto-Arena: Automating LLM Evaluations with Agent Peer Battles and Committee Discussions

As LLMs continuously evolve, there is an urgent need for a reliable evaluation method that delivers trustworthy results promptly. Currently, static benchmarks suffer from inflexibility and unreliability, leading users to prefer human voting platforms like Chatbot Arena. However, human evaluations require significant manual effort. To address this, we propose the Auto-Arena, an innovative framework that automates the entire evaluation process using LLM-powered agents. Firstly, an LLM examiner generates questions. Then, two LLM candidates engage in a multi-round peer battle based on individual questions, aiming at revealing their true performance differences. Finally, a committee of LLM judges collaboratively discusses and decides the winner, reducing bias and enhancing fairness. During the peer battles, we observe intriguing scenarios where the LLM candidates display competitive behaviors and even learn from the opponents. In our extensive experiments involving 15 recent LLMs, Auto-Arena shows a 92.14% correlation with human preferences, surpassing all previous expert-annotated benchmarks without any manual efforts. As a result, Auto-Arena offers a promising alternative to current human evaluation platforms for evaluating LLMs automatically.

arXiv:2405.18636

ChatGPT as the Marketplace of Ideas: Should Truth-Seeking Be the Goal of AI Content Governance?

As one of the most enduring metaphors within legal discourse, the marketplace of ideas has wielded considerable influence over the jurisprudential landscape for decades. A century after the inception of this theory, ChatGPT emerged as a revolutionary technological advancement in the twenty-first century. This research finds that ChatGPT effectively manifests the marketplace metaphor. It not only instantiates the promises envisaged by generations of legal scholars but also lays bare the perils discerned through sustained academic critique. Specifically, the workings of ChatGPT and the marketplace of ideas theory exhibit at least four common features: arena, means, objectives, and flaws. These shared attributes are sufficient to render ChatGPT historically the most qualified engine for actualizing the marketplace of ideas theory. The comparison of the marketplace theory and ChatGPT merely marks a starting point. A more meaningful undertaking entails reevaluating and reframing both internal and external AI policies by referring to the accumulated experience, insights, and suggestions researchers have raised to fix the marketplace theory. Here, a pivotal issue is: should truth-seeking be set as the goal of AI content governance? Given the unattainability of the absolute truth-seeking goal, I argue against adopting zero-risk policies. Instead, a more judicious approach would be to embrace a knowledge-based alternative wherein large language models (LLMs) are trained to generate competing and divergent viewpoints based on sufficient justifications. This research also argues that so-called AI content risks are not created by AI companies but are inherent in the entire information ecosystem. Thus, the burden of managing these risks should be distributed among different social actors, rather than being solely shouldered by chatbot companies.

arXiv:2405.18542

Automatic detection of cognitive impairment in elderly people using an entertainment chatbot with Natural Language Processing capabilities

Previous researchers have proposed intelligent systems for therapeutic monitoring of cognitive impairments. However, most existing practical approaches for this purpose are based on manual tests. This raises issues such as excessive caretaking effort and the white-coat effect. To avoid these issues, we present an intelligent conversational system for entertaining elderly people with news of their interest that monitors cognitive impairment transparently. Automatic chatbot dialogue stages allow assessing content description skills and detecting cognitive impairment with Machine Learning algorithms. We create these dialogue flows automatically from updated news items using Natural Language Generation techniques. The system also infers the gold standard of the answers to the questions, so it can assess cognitive capabilities automatically by comparing these answers with the user responses. It employs a similarity metric with values in [0, 1], in increasing level of similarity. To evaluate the performance and usability of our approach, we have conducted field tests with a test group of 30 elderly people in the earliest stages of dementia, under the supervision of gerontologists. In the experiments, we have analysed the effect of stress and concentration in these users. Those without cognitive impairment performed up to five times better. In particular, the similarity metric varied between 0.03, for stressed and unfocused participants, and 0.36, for relaxed and focused users. Finally, we developed a Machine Learning algorithm based on textual analysis features for automatic cognitive impairment detection, which attained accuracy, F-measure and recall levels above 80%. We have thus validated the automatic approach to detect cognitive impairment in elderly people based on entertainment content.

arXiv:2405.15818

DuanzAI: Slang-Enhanced LLM with Prompt for Humor Understanding

Language's complexity is evident in the rich tapestry of slang expressions, often laden with humor and cultural nuances. This linguistic phenomenon has become increasingly prevalent, especially in digital communication. However, existing AI models, including ChatGPT-3.5, face challenges in comprehending these nuances, particularly in Chinese slang. In this study, we present DuanzAI, an innovative approach enhancing Large Language Models (LLMs) with deep Chinese slang comprehension. Leveraging curated datasets and advanced techniques, DuanzAI bridges the gap between human expression and AI comprehension, enabling contextually relevant responses. Our experiments contrast LLMs' performance with a custom Punchline Entity Recognition (PER) system, integrating phonetic matching and pinyin2hanzi techniques. Applying these insights, we developed ChatDAI, an advanced chatbot and released our code at \url{https://github.com/YesianRohn/DuanzAI}.

arXiv:2405.15250

Coaching Copilot: Blended Form of an LLM-Powered Chatbot and a Human Coach to Effectively Support Self-Reflection for Leadership Growth

Chatbots' role in fostering self-reflection is now widely recognized, especially in inducing users' behavior change. While the benefits of 24/7 availability, scalability, and consistent responses have been demonstrated in contexts such as healthcare and tutoring to help one form a new habit, their utilization in coaching necessitating deeper introspective dialogue to induce leadership growth remains unexplored. This paper explores the potential of such a chatbot powered by recent Large Language Models (LLMs) in collaboration with professional coaches in the field of executive coaching. Through a design workshop with them and two weeks of user study involving ten coach-client pairs, we explored the feasibility and nuances of integrating chatbots to complement human coaches. Our findings highlight the benefits of chatbots' ubiquity and reasoning capabilities enabled by LLMs while identifying their limitations and design necessities for effective collaboration between human coaches and chatbots. By doing so, this work contributes to the foundation for augmenting one's self-reflective process with prevalent conversational agents through the human-in-the-loop approach.

arXiv:2405.13052

Large Language Models Can Infer Personality from Free-Form User Interactions

This study investigates the capacity of Large Language Models (LLMs) to infer the Big Five personality traits from free-form user interactions. The results demonstrate that a chatbot powered by GPT-4 can infer personality with moderate accuracy, outperforming previous approaches drawing inferences from static text content. The accuracy of inferences varied across different conversational settings. Performance was highest when the chatbot was prompted to elicit personality-relevant information from users (mean r=.443, range=[.245, .640]), followed by a condition placing greater emphasis on naturalistic interaction (mean r=.218, range=[.066, .373]). Notably, the direct focus on personality assessment did not result in a less positive user experience, with participants reporting the interactions to be equally natural, pleasant, engaging, and humanlike across both conditions. A chatbot mimicking ChatGPT's default behavior of acting as a helpful assistant led to markedly inferior personality inferences and lower user experience ratings but still captured psychologically meaningful information for some of the personality traits (mean r=.117, range=[-.004, .209]). Preliminary analyses suggest that the accuracy of personality inferences varies only marginally across different socio-demographic subgroups. Our results highlight the potential of LLMs for psychological profiling based on conversational interactions. We discuss practical implications and ethical challenges associated with these findings.

arXiv:2405.14388

Evaluation of the Programming Skills of Large Language Models

The advent of Large Language Models (LLM) has revolutionized the efficiency and speed with which tasks are completed, marking a significant leap in productivity through technological innovation. As these chatbots tackle increasingly complex tasks, the challenge of assessing the quality of their outputs has become paramount. This paper critically examines the output quality of two leading LLMs, OpenAI's ChatGPT and Google's Gemini AI, by comparing the quality of programming code generated in both their free versions. Through the lens of a real-world example coupled with a systematic dataset, we investigate the code quality produced by these LLMs. Given their notable proficiency in code generation, this aspect of chatbot capability presents a particularly compelling area for analysis. Furthermore, the complexity of programming code often escalates to levels where its verification becomes a formidable task, underscoring the importance of our study. This research aims to shed light on the efficacy and reliability of LLMs in generating high-quality programming code, an endeavor that has significant implications for the field of software development and beyond.

arXiv:2405.14734공식 코드 · Papers with Code

SimPO: Simple Preference Optimization with a Reference-Free Reward

Direct Preference Optimization (DPO) is a widely used offline preference optimization algorithm that reparameterizes reward functions in reinforcement learning from human feedback (RLHF) to enhance simplicity and training stability. In this work, we propose SimPO, a simpler yet more effective approach. The effectiveness of SimPO is attributed to a key design: using the average log probability of a sequence as the implicit reward. This reward formulation better aligns with model generation and eliminates the need for a reference model, making it more compute and memory efficient. Additionally, we introduce a target reward margin to the Bradley-Terry objective to encourage a larger margin between the winning and losing responses, further improving the algorithm's performance. We compare SimPO to DPO and its latest variants across various state-of-the-art training setups, including both base and instruction-tuned models such as Mistral, Llama 3, and Gemma 2. We evaluate on extensive chat-based evaluation benchmarks, including AlpacaEval 2, MT-Bench, and Arena-Hard. Our results demonstrate that SimPO consistently and significantly outperforms existing approaches without substantially increasing response length. Specifically, SimPO outperforms DPO by up to 6.4 points on AlpacaEval 2 and by up to 7.5 points on Arena-Hard. Our top-performing model, built on Gemma-2-9B-it, achieves a 72.4% length-controlled win rate on AlpacaEval 2, a 59.1% win rate on Arena-Hard, and ranks 1st on Chatbot Arena among <10B models with real user votes.

arXiv:2405.14006공식 코드 · Papers with Code

Evaluating Large Language Models with Human Feedback: Establishing a Swedish Benchmark

In the rapidly evolving field of artificial intelligence, large language models (LLMs) have demonstrated significant capabilities across numerous applications. However, the performance of these models in languages with fewer resources, such as Swedish, remains under-explored. This study introduces a comprehensive human benchmark to assess the efficacy of prominent LLMs in understanding and generating Swedish language texts using forced choice ranking. We employ a modified version of the ChatbotArena benchmark, incorporating human feedback to evaluate eleven different models, including GPT-4, GPT-3.5, various Claude and Llama models, and bespoke models like Dolphin-2.9-llama3b-8b-flashback and BeagleCatMunin. These models were chosen based on their performance on LMSYS chatbot arena and the Scandeval benchmarks. We release the chatbotarena.se benchmark as a tool to improve our understanding of language model performance in Swedish with the hopes that it will be widely used. We aim to create a leaderboard once sufficient data has been collected and analysed.

arXiv:2405.12712

From Human-to-Human to Human-to-Bot Conversations in Software Engineering

Software developers use natural language to interact not only with other humans, but increasingly also with chatbots. These interactions have different properties and flow differently based on what goal the developer wants to achieve and who they interact with. In this paper, we aim to understand the dynamics of conversations that occur during modern software development after the integration of AI and chatbots, enabling a deeper recognition of the advantages and disadvantages of including chatbot interactions in addition to human conversations in collaborative work. We compile existing conversation attributes with humans and NLU-based chatbots and adapt them to the context of software development. Then, we extend the comparison to include LLM-powered chatbots based on an observational study. We present similarities and differences between human-to-human and human-to-bot conversations, also distinguishing between NLU- and LLM-based chatbots. Furthermore, we discuss how understanding the differences among the conversation styles guides the developer on how to shape their expectations from a conversation and consequently support the communication within a software team. We conclude that the recent conversation styles that we observe with LLM-chatbots can not replace conversations with humans due to certain attributes regarding social aspects despite their ability to support productivity and decrease the developers' mental load.