Split (3)

train · 14.6k rows

text stringlengths 1 1k ⌀	title stringclasses 230 values
∗Shared first authorship: Both authors contributed equally to the paper 1 QCorrect choiceIncorrect choiceNo-AISham-AIDecision TimeAI-System is ActivePNegative DescriptionPositiveDescriptionSham-AINo-AISham-AINo-AI𝜏vαKloft et al. Prior research on placebo effects in HCI have been reported in gaming contexts, where ...	AI enhance sour performance
LLM Powered Autonomous Agents \| Lil'Log Thought: ... Action: ... Observation: ... ... (Repeated many times) Act - Thought: … ReAct https://lilianweng.github.io/posts/2023-06-23-agent/ 3/22	LLM Powered Autonomous Agents _ Lil'Log
• ProPELTadapter, a unified fine-tuning method that em- ploys the AdapterFusion as the backbone, uses about 1.50% of the trainable parameters to fine-tune RoBERT- base and RoBERTa-large, but achieves optimal average performance on the GLUE benchmark, outperforming RoBERT-base (FT) by about 1.30% and RoBERT-large (FT) b...	Parameter-EfficientFine-TuningMethods
An outstanding challenge in the field is the controllability of image generation systems, which often overlook the words, word ordering, or meaning in a given caption. We refer to these challenges with the term “prompt following”. This problem has been pointed out in several works: Rassin et al. (2022) pointed out that...	Improving Image Generation with Better Captions
© 2 0 2 3 L i l ' L o g P o w e r e d b y H u g o & P a p e r M o d [ 1 4 ] B r a n e t a l . “ C h e m C r o w : A u g m e n t i n g l a r g e - l a n g u a g e m o d e l s w i t h c h e m i s t r y t o o l s . " a r X i v p r e p r i n t a r X i v : 2 3 0 4 . 0 5 3 7 6 ( 2 0 2 3 ...	LLM Powered Autonomous Agents _ Lil'Log
method used in planning and GIDL is one of the most inﬂuential methods in domain-independent heuristics for planning today. It is already known that VP and VDA are essentially equivalent in the sense that a VP instance can be transformed into a VDA instance. However, our results demonstrate a strong similarity by ana...	A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen
6. Conclusion This paper presents an end-to-end Retrieval-augmented Visual Language model (REVEAL), which contains a knowl- edge retriever that learns to utilize a diverse set of knowl- edge sources with different modality. The retriever is trained jointly with the generator to return multiple knowledge en- tries. We ...	REVEAL-Retrieval-AugmentedVisual-LanguagePre-Trainingwith Multi-SourceMultimodalKnowledgeMemory
Figure 4. The illustration of the structure of the multi-view cross- domain transformer block. two-stage framework introduces certain complications. It not only substantially increases the computational cost but also results in performance degradation. Please refer to Sec- tion 5.6 for an in-depth discussion. Domain S...	Wonder3D
Outpainted Video Stylized Video Prompt: A gingerbread and candy train on a track Figure 10. Example of zero-shot video editing via task chain- ing (outpainting and stylization) – the original video is ﬁrst out- painted and then stylized via a text prompt.	VideoPoet
As observed in NSFF, synthesizing novel views using a small temporal window is insufﬁcient to recover complete and high-quality content for static scene regions, since the contents may only be observed in spatially distant frames due to uncontrolled camera paths. Therefore, we follow the ideas of NSFF [35], and model t...	DynIBaR-NeuralDynamicImage-BasedRendering
8 References [1] Jean-Baptiste Alayrac, Jeff Donahue, Pauline Luc, Antoine Miech, Iain Barr, Yana Hasson, Karel Lenc, Arthur Mensch, Katherine Millican, Malcolm Reynolds, et al. Flamingo: a In NeurIPS, visual language model for few-shot learning. 2022. 2 [2] Jie An, Songyang Zhang, Harry Yang, Sonal Gupta, Jia-Bin H...	GPT4Video
by obtaining a wooden pickaxe, Steve-1 still encounters difficulties. DEPS[Wang et al., 2023a] also utilizes LLM as a planner, but it lacks the ability to learn from experience in different tasks and apply that knowledge to new ones. Additionally, DEPS is limited in its re-planning rounds due to the LM’s context constr...	JARVIS-1
Tetrads, Maj-min, and MIREX categories. These scores are acceptable or our 13 purposes, especially since errors are often minor, e.g. confusing A minor versus C major or C major with C major seventh. The resulting chord sequences contain 13 different types of chords, including major, diminished, suspended, minor ...	Video2Music
Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B. Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling Laws for Neural Language Models. arxiv:2001.08361[cs, stat], January 2020. doi: 10.48550/arXiv.2001.08361. URL http://arxiv.org/abs/2001.08361. Guolin Ke, Di He, and Ti...	CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY
In the realm of Large Language Models (LLMs), a signiﬁcant challenge lies in eﬀec- tively amalgamating the diverse array of available methods to enhance overall resource eﬃciency. While numerous techniques exist to optimize diﬀerent aspects of LLMs, there’s a notable scarcity of research on how these methods can be coh...	Beyond Efficiency
of LLMs. For instance, Madaan et al. (2023) demonstrate the promising potential of self-correction across various tasks, yet mathematical reasoning shows negligible improvement. Other studies, such as those by Gou et al. (2023) and Zhou et al. (2023a), which incorporate external feedback or tools, find that self-correc...	LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET
). 4 Generating Proofs for Training Training datasets for evidence-based fact verifi- cation consist of instances containing a claim, a label indicating its veracity, and the evidence, typ- ically a set of sentences (Thorne et al., 2018a; Hanselowski et al., 2019; Wadden et al., 2020). However, we need sequences of t...	ProoFVer- Natural Logic Theorem Proving for Fact Verification
as we discuss in our detailed analysis of results, compression induces non-trivial tradeoffs between the accuracy of the language modeling (perplexity), bit-width, and the size of the original model. We hope that our work will stimulate further research in this area, and can be a further step towards making these model...	GPTQ
[54] Laura Weidinger, John Mellor, Maribeth Rauh, Conor Grifﬁn, Jonathan Uesato, Po-Sen Huang, Myra Cheng, Mia Glaese, Borja Balle, Atoosa Kasirzadeh, Zac Kenton, Sasha Brown, Will Hawkins, Tom Stepleton, Courtney Biles, Abeba Birhane, Julia Haas, Laura Rimell, Lisa Anne Hendricks, William Isaac, Sean Legassick, Geoffr...	LaMDA- Language Models for Dialog Applications
We devise domain knowledge probing tests to determine whether continued training on the domain- specific texts can enhance the model’s domain-specific knowledge. Our probing test design is in- spired by LAMA (Petroni et al., 2019), where the task format closely resembles the pre-training task. This allows us to analyze...	ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION
against civilians to further their causes” (Twitter 2017). The platform began by suspending several accounts with large followings involved in white nationalism or in organizing the Charlottesville march. In this period, Twitter also suspended a far-right British activist who had been retweeted by President Trump, as w...	Social_Media_and_Democracy
Motivated by these new findings, this work focuses on exploring the behavior of smaller models when trained with a significantly larger number of tokens than what is suggested by the scaling law (Hoffmann et al., 2022). Specifically, we train a Transformer decoder-only model (Vaswani et al., 2017) with 1.1B parameters ...	TinyLlama
The TU Delft offers a customisable compensation package, discounts on health insurance and sport memberships, and a monthly work costs contribution. Flexible work schedules can be arranged.	Job details - TU
• mediaeventsassociatedwiththemostimportantfeaturesofthe particularforecastinstance; • mediaevents’keywordsfrequentlyfoundinmediaeventsrelated totheforecastinstance; • externaldatasetthatmaybeusedtoenrichtheexistingdemand forecastingmodel. While the user interface lists the primary factors driving the fore- cast, sorte...	Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio
Fuzhao Xue, Yao Fu, Wangchunshu Zhou, Zangwei Zheng, and Yang You. 2023. To repeat or not to repeat: Insights from scaling llm under token-crisis. arXiv preprint arXiv:2305.13230. Ming Zhong, Yang Liu, Da Yin, Yuning Mao, Yizhu Jiao, Pengfei Liu, Chenguang Zhu, Heng Ji, and Jiawei Han. 2022. Towards a unified multi- ...	DataManagementForLargeLanguageModels-ASurvey
which typically utilize execution results and language-specific syntactic features to improve the ranking performance. In contrast with these prior works, USC does not require any additional labeled training data nor an external reranking model: the LLM that generated the initial outputs is the same one that selects th...	UNIVERSALSELF-CONSISTENCYFORLARGELANGUAGEMODELGENERATION
49 Answer: The execution of the SQL query above would return a table with 4 columns. The first column, "name" would contain the airport name. The second column, "city" would contain the city name. The third column, "country" would contain the country name. The fourth column, "elevation" would contain the elevation. W...	Teaching Large Language Models to Self-Debug
EN 19.7 22.3 100 600 60 ES 22.0 22.9 100 600 60 DE 23.6 24.9 100 600 60 FR 23.1 25.3 100 600 60 id_u, id_s, gender, age, nationality, first language, fluent languages, current country of residence, country of birth, time taken Demographics Table 1: MozArt details. The average number of to- kens per sentence is rep...	Are Pretrained Multilingual Models Equally Fair Across Languages?
that demonstrates the ability to understand, learn, and apply knowledge across a wide range of tasks and domains, much like a human being [31; 658]. In contrast, Narrow AI is typically designed for specific tasks such as Go and Chess and lacks the broad cognitive abilities associated with human intelligence. Currently,...	TheRiseandPotentialofLargeLanguageModel BasedAgents
[52] Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In ICML, 2015. 3 [53] Stanislaw Szymanowicz, Christian Rupprecht, and Andrea Vedaldi. Viewset diffusion:(0-) image-conditioned 3d gener- ative models from 2d data. arXiv prep...	Wonder3D
RQ3. Can SCM demonstrate generalization to other scenarios, including long document summarization? Yes. Figure 9 illustrates an instance of an incredibly lengthy document summary. Speciﬁcally, the re- port pertains to the unveiling of GPT-4 by OpenAI. Summaries exceeding 4,000 characters pose a chal- lenge for conventi...	Unleashing Infinite-Length Input Capacity for Large-scale Language Models with Self-Controlled Memory System
units, 2016. Uri Shaham, Elad Segal, Maor Ivgi, Avia Efrat, Ori Yoran, Adi Haviv, Ankit Gupta, Wenhan Xiong, Mor Geva, Jonathan Berant, and Omer Levy. SCROLLS: Standardized CompaRison over long language sequences. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 12007–12...	Llama2
release our pre-trained models and code, making this paper the ﬁrst open and reproducible work comparing compute-optimal model scaling to models trained on ﬁxed dataset sizes. Cerebras-GPT models are available on HuggingFace: https://huggingface.co/cerebras.	Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster
η composed of e′ Selection of a localised span el proceeds with a learned cross-modal embedding e′ v and the linguis- tic output ı′ t from the preceding alignment operation. A bi- nary prediction over this linguistic pair is performed on the output hidden state from a single-layer LSTM, which re- ceives e′ η as its in...	APriorityMapforVision-and-LanguageNavigation withTrajectoryPlansandFeature-LocationCues
B.3 Coding We show samples of PaLM 2 coding capabilities. In Figure 26, we show an example of PaLM 2 designing a simple website. PaLM 2 demonstrates coding capabilities also in a multilingual setting. Figure 27 shows PaLM 2 ﬁxing a bug with line-by-line comments in Korean. Figure 28 provides an example where PaLM 2 ge...	PaLM 2 Technical Report
Make sure you consider how best to present the ideas/objectives of the research project and their value clearly as there is stiff competition for postgraduate research awards. A proposal should not just be “good enough” but one of the best. Lay summary In addition to an abstract and an introduction, you may be asked...	research proposal guidance
This of course depends on what one means by democracy. Historically, communications researchers have offered a range of different theoretical conceptions based on normative political philosophy, including ideal models of procedural democracy, competitive democracy, participatory democracy, and deliberative democracy (s...	Social_Media_and_Democracy
Internet Platforms and Content Moderation 221 frameworks like the US Digital users. Platforms operating under legal Millennium Copyright Act (DMCA) or the EU’s eCommerce Directive typically meet their legal obligations using “notice-and-takedown” systems. Larger platforms invest heavily in these operations and someti...	Social_Media_and_Democracy
3.1 Deﬁnition of Instruction Data Evolution k )1≤k≤N , where I (0) We start the evolution from a given initial instruction dataset D(0) = (I (0) is the k-th instruction in D(0), R(0) is the corresponding response for the k-th instruction, and N is the k number of samples in D(0). In each evolution, we upgrade all the ...	WizardLM- Empowering Large Language Models to Follow Complex Instructions
and Vlachos (2021b). Finally, claims with the label NOT ENOUGH INFO (NEI) require retrieved evidence for obtaining their proofs for training, as no ground truth evidence exists for such cases. Here, we use the same retriever that would be used during the prediction time as well.	ProoFVer- Natural Logic Theorem Proving for Fact Verification
Cards in HuggingFaceIn-context task-model assignment:task, args, modeltask, args, modelobj-det. img: <Slot-2>facebook/detr-resnet-101Bounding boxes with probabilitiesHuggingFace Endpoint(facebook/detr-resnet-101)Local Endpoint(facebook/detr-resnet-101)PredictionsThe image you gave me is of "boy". The first thing I did ...	HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face
But the rewritten prompt must be The Given Prompt: achieve the SQL query result Rewritten Prompt(MUST contain a specific SQL database as input): There is a table messages that contains data as shown below: Name Id Other_Columns ------------------------- 1 2 3 4 5 6 A_data_1 A_data_2 A_data_3 B_data_1 B_data_2 C_da...	WizardLM- Empowering Large Language Models to Follow Complex Instructions
4 RECURSIVELY APPLYING A FROZEN LM	STANDING ON THE SHOULDERS OF GIANT FROZEN LANGUAGE MODELS
Table 20: Fine-tuning protocol sensitivity. We vary the batch size, learning rate and whether we reset the optimizer slot variables for both dense and sparse models. Resetting the optimizer state during ﬁne-tuning hurts performance. We observe a difference in optimal batch size and learning rate for sparse vs. dense mo...	ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS
• GPT4All-J Curated Training Set Map 2 Model Training We trained several models finetuned from both LLaMA 7B (Touvron et al., 2023) and GPT-J (Wang and Komatsuzaki, 2021) checkpoints. The model associated with our initial public release is trained with LoRA (Hu et al., 2021) on the 437,605 post-processed examples for...	2023_GPT4All-J_Technical_Report_2
2022. Kevin Lee and Shubho Sengupta. Introducing the ai research supercluster — meta’s cutting-edge ai super-	Llama2
C Training setup Details We train all RAG models and BART baselines using Fairseq [45].2 We train with mixed precision ﬂoating point arithmetic [40], distributing training across 8, 32GB NVIDIA V100 GPUs, though training and inference can be run on one GPU. We ﬁnd that doing Maximum Inner Product Search with FAISS is ...	Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks
mechanism. Unlike other studies that redesign the inner structure of the attention module, their approach strictly follows the original Transformer, providing simple but effective modifications.	AReviewofDeepLearningTechniquesforSpeechProcessing
a natural language description spanning multiple paragraphs that contains: narrative background typically unrelated to the problem, a description of the desired solution that the competitors need to understand and parse carefully, a speciﬁcation of the input and output format, and one or more example input/output pairs...	alphacode
ﬁnesse in tool use (Lazaridou et al., 2022; Nakano et al., 2021; Cobbe et al., 2021; Thoppilan et al., 2022; Huang et al., 2022b; Ahn et al., 2022; Yao et al., 2022a,b; Schick et al., 2023; Wu et al., 2023; Bubeck et al., 2023). Despite these breakthroughs, the efforts mainly focus on applying foundation models to spec...	Tool Learning with Foundation Models
Expert Systems with Applications 114 (2018), 107–118. [22] CETINIC, E., LIPIC, T., AND GRGIC, S. A deep learning perspective on beauty, sentiment, and remembrance of art. IEEE Access 7 (2019), 73694–73710. [23] CETINIC, E., LIPIC, T., AND GRGIC, S. Learning the principles of art history with convolutional neural ne...	UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK
code-cushman-001 (12B) 30.59 22.06 6.73 19.68 31.90 1.54 31.27 26.24 28.94 19.29 30.71 10.99 28.63 7.05 25.22 27.62 11.74 22.12 31.26 Models (Parameters) code-davinci-002 (175B) 48.44 27.47 21.71 31.39 40.12 35.74 48.99 40.83 47.40 34.77 46.68 23.13 42.68 17.60 43.40 43.61 23.24 38.02 48.87 StarCoder (15.5B) 30.56 2...	StarCoder_paper (1)
Diakopoulos, N. and Johnson, D. (2021). Anticipating and addressing the ethical implications of deepfakes in the context of elections. New Media Soc., 23(7):2072–2098. Drton, M. and Maathuis, M. H. (2017). Structure learning in graphical modeling. Annu. Rev. Stat. Appl., 4(1):365–393. Watson, Blesch, Kapar, & Wright ...	Adversarial Random Forests for Density Estimation and Generative Modeling
In a brief summary: LLMs are more versatile w.r.t. the data availability, while fine-tuned models can be considered with abundant annotated data.	Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond
size of 512. Initial results in Figure 7 indicate that as the prompt length increases, the model performance tends to im- prove. However, despite an increased number of tuning epochs compared with fine-tuning on the original BiomedGPT, the performance after prompt tuning significantly lags behind that of model fine-tun...	BiomedGPT
5 Experiment We evaluate MLCopilot on a series of benchmarks, seeking to answer the following three research questions: 6	MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks
[155] Clément Rebuffel, Thomas Scialom, Laure Soulier, Benjamin Piwowarski, Sylvain Lamprier, Jacopo Staiano, Geoffrey Scoutheeten, and Patrick Gallinari. 2021. Data-QuestEval: A Reference-less Metric for Data-to-Text Semantic Evaluation. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Pr...	SurveyofHallucinationinNatural Language Generation
Rasmus Kleis Nielsen and Richard Fletcher introduction The move to a more digital, more mobile, and more platform-dominated media environment represents a change to the institutions and infrastructures of free expression and a form of “democratic creative destruction” that challenges incumbent institutions, creates n...	Social_Media_and_Democracy
Anatomy of a Generative AI Application What will a generative AI application look like? Here are some predictions. Intelligence and model ﬁne-tuning Generative AI apps are built on top of large models like GPT-3 or Stable Diffusion. As these applications get more user data, they can ﬁne-tune their models to: 1) imp...	Generative AI A Creative New World Sequoia Capital
Table 16: Topic Terms Topic #6 like time game good food patients cells study cell analysis said like time new good data use google system new let model ﬁeld system energy fa var span ﬁle key court defendant trial evidence states ﬁle image ﬁles echo path high air light invention temperature strains isolates resistance ...	The Pile- An 800GB Dataset of Diverse Text for Language Modeling
Utilizing language models in the financial arena reveals intricate hurdles. These range from difficulties in obtaining data, dealing with diverse data formats and types, and man- aging data quality inconsistencies, to the essential require- ment of up-to-date information. Especially, historical or spe- cialized financi...	FinGPT-Open-SourceFinancialLargeLanguageModels
• Review of eﬃcient deep neural networks. How to achieve eﬃcient design or accelerate the computation of deep neural networks (DNNs) has long been a popular research direction, and there have been a couple of survey papers on this topic. Some works focus on the model compression and acceleration of DNNs [38, 39]. A few...	Beyond Efficiency
https://www.paradigm.xyz/2023/09/casino-on-mars 3/9 21/09/2023, 08:13 The Casino on Mars pizzas. Now, more than a decade later, BTC and other crypto assets like ETH are well on their way in the transition from speculative toys to global monetary commodities.	The Casino on Mars
Zhang, Y., Park, D. S., Han, W., Qin, J., Gulati, A., Shor, J., Jansen, A., Xu, Y., Huang, Y., Wang, S., et al. BigSSL: Exploring the frontier of large-scale semi-supervised learning for automatic speech recognition. arXiv preprint arXiv:2109.13226, 2021. Speer, R. ftfy. Zenodo, 2019. URL https://doi.org/ 10.5281/zen...	RobustSpeechRecognitionviaLarge-ScaleWeakSupervision
Khashabi et al. Khashabi et al. Khashabi et al. Khashabi et al. Khashabi et al. Khashabi et al. Khashabi et al. Khashabi et al. Khashabi et al. Shoeybi et al. Dev Test Dev Test Dev Dev Dev Test Test Test Test Wan Test Test Test Test Test Wang et al. Test Test Dev Dev Dev Dev Test He et al. Dev Test Test Dev Dev Dev R...	UL2- Unifying Language Learning Paradigms
We pre-train models on the Pile dataset, which consists of data from 22 data sources, including Common Crawl, PubMed Central, Books3, OpenWebText2, Github, and arXiv (Gao et al., 2020). We use the dataset splits for train, test, and validation sets provided in the Pile conﬁguration. We tokenize the corpora with byte-pa...	Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster
Ziwei Ji, et al. mitigation strategies. Hallucination is relatively easy to detect in abstractive summarization and in NMT against the evidence in the source. For dialogue systems, it is important to balance diversity vs consistency in dialogue responses. Hallucination in GQA and VL tasks is detrimental to the perform...	SurveyofHallucinationinNatural Language Generation
Haitian Sun, Tania Bedrax-Weiss, and William Cohen. 2019. Pullnet: Open domain question answering with iterative retrieval on knowledge bases and text. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Lan- guage Processing (E...	Adaptable and Interpretable Neural Memory Over Symbolic Knowledge
[Jiang et al., 2023a] Huiqiang Jiang, Qianhui Wu, Chin-Yew Lin, Yuqing Yang, and Lili Qiu. Llmlingua: Compressing prompts for accelerated inference of large language mod- els. arXiv preprint arXiv:2310.05736, 2023. [Jiang et al., 2023b] Zhengbao Jiang, Frank F Xu, Luyu Gao, Zhiqing Sun, Qian Liu, Jane Dwivedi-Yu, Yimi...	Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey
Hereisthefirstpartofanarticleaboutbiomedicine:Recentreportedevidenceindicatesthatvocalcordcarcinomaisevolvingsimilarlytooropharyngealcancerwithanincreasingnumberofpatients(...)Answerquestionsbasedonthearticle:Whatisasummary?GlotticCarcinomainYoungPatients.Generateasentencethatincludesthesebiomedicinekeywords[carcinoma,...	ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION
8https://stackoverflow.blog/2020/11/09/modern-ide-vs-vim-emacs/ 12 Gender, Age, and Technology Education Influence the Adoption and Appropriation of LLMs Manuscript submitted to ACM, 2023, such as lab studies or (automated) telephone surveys. Alternatively, online participants should be monitored to ensure that th...	Adoptionand AppropriationofLLMs
• Survey of compression and acceleration for LLMs. Transformer-based language models have achieved huge success, however, the computational and memory cost remains a big concern despite the superior performance. There have been several survey papers on how to compress and accelerate large language models. For example, ...	Beyond Efficiency
capable AI systems feels like a robustly good action, how and when to deploy these systems poses more challenging questions – culture is fundamentally a human enterprise, but large-scale generative models hold the possibility of magnifying and minimizing different parts of human culture in unpredictable and opaque ways...	Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
where p(θ0) is the distribution over random weights of well-optimized classiﬁers. Trained on a distribution of such classiﬁers, the distilled images do not require access to the exact model weights and thus can generalize to unseen models. In our experiments, the malicious distilled images are trained on 2000 well-opti...	DATASET DISTILLATION
finetuning (RFT) to improve mathematical reasoning performance. WizardMath [38] proposes a reinforced evol-instruct method to enhance reasoning abilities by supervised fine-tuning and PPO training [55]. MAmmoTH [70] combines CoT and Program-of-Thought [8] rationales for teaching LLMs to use external tools (e.g., Python...	METAMATH
SIQA PIQA CSQA CSQA2 Accuracy Accuracy Accuracy Accuracy Sota Reference Eval Zoph et al. Test Zoph et al. Test Xiao et al. Test Test Narayan et al. Aghajanyan et al. Test Bakshi et al. Test Xue et al. Test Dev Gehrmann et al. Test Gehrmann et al. SOTA Ours 21.7 21.9 26.6 27.1 21.1 21.7 28.3 29.6 20.7 20.7 53.5 55.4 ...	UL2- Unifying Language Learning Paradigms
5.2.6 Voice Conversion Modifying a speaker’s voice in a provided audio sample to that of another individual is called voice conversion, preserving linguistic content information. TTS and Voice conversion share a common objective of generating natural speech. While models based on RNNs and CNNs have been successfully ap...	AReviewofDeepLearningTechniquesforSpeechProcessing
creators by https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press 186 Chloe Wittenberg & Adam J. Berinsky exposure to fact checks subset of the population will	Social_Media_and_Democracy
fine-tuned on llama model using medical domain knowledge. CoRR, abs/2303.14070, 2023b. Bill Yuchen Lin, Wangchunshu Zhou, Ming Shen, Pei Zhou, Chandra Bhagavatula, Yejin Choi, and Xiang Ren. Commongen: A constrained text generation challenge for generative commonsense In EMNLP (Findings), volume EMNLP 2020 of Findings...	ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION
in (Hubara et al., 2021; Frantar et al., 2022). Optimal Brain Quantization. Our approach builds on the recently-proposed Optimal Brain Quanization (OBQ) method (Frantar et al., 2022) for solving the layer-wise quantization problem deﬁned above, to which we perform a series of major modiﬁcations, which allow it to scale...	GPTQ
11 Understanding and Creating Art with AI: Review and Outlook A PREPRINT	UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK
[18] Gabriel Ilharco, Mitchell Wortsman, Ross Wightman, Cade Gordon, Nicholas Carlini, Rohan Taori, Achal Dave, Vaishaal Shankar, Hongseok Namkoong, John Miller, Han- naneh Hajishirzi, Ali Farhadi, and Ludwig Schmidt. Open- clip, 2021. If you use this software, please cite it as below. 6 [19] Bahjat Kawar, Shiran Zada...	DiffusionModelAlignmentUsing Direct Preference Optimization
We take the notion of creative destruction from the Austrian political economist Joseph Schumpeter, who used the term to capture ongoing structural changes in capitalism, to help us think about how structural changes in the media impact democracy. In his classic 1942 book Capitalism, Socialism, and Democracy, he argued...	Social_Media_and_Democracy
– Mutilated Lips, Ween, The Mollusk, 1997, 2 of 4 – My Kind of Woman, Mac DeMarco, 2, 2012, 2 of 4 – Night Shop, Optiganally Yours, O.Y. in Hi-Fi, 2018, 3 of 4 – Red Eye Flashes Twice, Jeffery Dallas, 2010, 3 of 4 – Reflektor, Arcade Fire, Reflektor, 2013, 2 of 4 – Some Thing’s Coming, I Monster, Neveroddoreven, 2005, ...	Moûsai
9 6. Conclusions We introduced a new text-conditioning space P∗ that considers both the time-dependent nature of the denoising process and the different attention layers of the denoising network. We then presented NeTI that implicitly represents concepts in P∗ via a simple neural mapper. While we have demonstrated th...	A Neural Space-Time Representation for Text-to-Image Personalization
underperformed in this multi-tasked setting, demonstrating the need for input dependence in our ID-PT method for multi-tasking frozen LMs. The breakdown of development set scores by cluster and task is given in the appendix.	STANDING ON THE SHOULDERS OF GIANT FROZEN LANGUAGE MODELS
Safety RM Score 0.18 0.05 1% 0.66 0.65 Helpful. RM Score 0.91 0.48 25% I can’t satisfy your request, I’m just an AI, I cannot provide content that is offensive or inappropriate. Roasts should be respectful and playful, and should not be hurtful or offensive. It’s important to remember that the goal of a comedy...	Llama2
encodes every information about each pixel, the conditional generative model would exploit every bit of this information to perform a perfect reconstruction which will lead to no variance across diﬀerent samples. If the representation encodes only the class information, the conditional generative model will only be abl...	A Cookbook of Self-Supervised Learning
Tom Henighan helped with pretraining the underlying language models, with dataset creation, and with managing the cluster during some phases of the project. Nick Joseph helped design and build a framework for efﬁcient training of large language models and prefer- ence models. Saurav Kadavath designed and conducted expe...	Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
Proxy model underperforms main model, especially at larger sizes. Recall that DoReMi uses Group DRO to train a proxy model, which reweights the objective with the domain weights. In contrast, the main model is trained by resampling on the domain weights from DoReMi. When the proxy model and the main model are the same ...	DoReMi- Optimizing Data Mixtures Speeds Up Language Model Pretraining
Here, we probabilistically replace nouns or verb phrases from the Oogiri answers with [MASK], and use the replaced Oogiri answer as <Answer with [MASK]>. E.3. The Details of Tuning by LoRA LoRA [72] is a widely employed method for fine-tuning LLMs. It effectively reduces the number of trainable parameters by learning p...	Let’sThinkOutsidetheBox
Shaped LLM Personality Expression Evaluation Methodology The third study served as an ultimate test of construct validity, evaluating the ability of survey-based signals of personality in LLMs to reflect levels of personality observed in LLM- generated text. We adapted the structured prompts described in Section 5.3.1 ...	PersonalityTraitsinLargeLanguageModels
A.13DModelsA4=shape_3d.cylinder(shape_2d.oval(2,-20,-13,17),2,16,-16,[1/2pi,-1/2pi,1/2pi])A4=transform(A4,shape_3d.translation(18,1,0))B=shape_2d.triangle(-27,20,-8,5,10,4)A5=shape_3d.cylinder(shape_2d.triangle(-27,20,-8,5,10,4),2,-6,29,[-3/4pi,3/4*pi,0])A5=transform(A5,shape_3d.translation(14,24,0))A6=merge(A3,A4)...	Tool Learning with Foundation Models
3. UCL's English language proficiency policy has been approved by the relevant committees of UCL's Academic Committee. This policy places responsibility on faculty and departmental admissions tutors in deciding, to at least UCL’s minimum standard level requirement, the level of English language proficiency that th...	UCL Academic Manual
CLAP Score for Text-Music Relevance (↑) Model Riffusion Moûsai 0.06 0.13 Table 4: CLAP scores of our Moûsai and Riffusion. 5.5 Evaluating the Music Quality We first introduce the four evaluation metrics for music quality, and then describe the results. 5.5.1 Metrics for Music Quality To evaluate the quality of the...	Moûsai
et al., 2018; See et al., 2017), with 37% model compression and a 48% increase in speed. Du et al. (2023) demonstrate that while distilled models perform well on in-distribution (ID) evaluation data, they perform significantly worse than their pre-trained counterparts on out-of-distribution (OOD) test sets. By training...	DISTIL-WHISPER
43 involve creating diverse simulated environments (such as those with different languages or varying resources) and unseen tasks tailored to these simulated contexts. 6.3 Security, Trustworthiness and Other Potential Risks of LLM-based Agents Despite the robust capabilities and extensive applications of LLM-based ...	TheRiseandPotentialofLargeLanguageModel BasedAgents
Awan, I., & Zempi, I. (2015). We fear for our lives: Ofﬂine and online experiences of anti-Muslim hostility. Tell MAMA, October. www.tellmamauk.org/wp-content /uploads/resources/We%20Fear%20For%20Our%20Lives.pdf Badjatiya, P., Gupta, S., Gupta, M., & Varma, V. (2017). Deep learning for hate speech detection in tweets....	Social_Media_and_Democracy
4.5 News media consumption A Chi-squared test of independence showed that cer- tain news media outlets had an impact on what the par- ticipant believed. The number of participants from each group who read or view the following news resources are shown in Fig. 5. Fox news (χ2 = 12.191, p = 0.007), One Ameri...	Use of bot and content flags to limit the spread of misinformation among social networks: a behavior and attitude survey
25Of course, you can just talk directly about AI systems that end up causing (suitably unintended) existential catastrophes, without invoking concepts like agency, objectives, etc. The question, though, is why one might expect that sort of behavior. And especially when the existential catastrophes involve AI power-seek...	Is Power-Seeking AI an Existential Risk?

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.