Split (3)

train · 14.6k rows

text stringlengths 1 1k ⌀	title stringclasses 230 values
formation compression and result rerank. Information Compression Even though the retriever can fetch relevant information from a vast knowledge base, we are still confronted with the chal- lenge of dealing with a substantial amount of information in retrieval documents. Some existing research attempts to solve this pro...	Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey
3.5 RESULTS We report the results for a single training run for the 582M parameter model. A similar training run with the 300M parameter model in described in Appendix C. Additional replication experiments are included in Appendix D. The 582M model began self training after 6 digits (Figure 10). Self learning terminat...	CHAIN-OF-THOUGHTREASONING IS APOLICY IMPROVEMENTOPERATOR
[47] Tongshuang Wu, Ellen Jiang, Aaron Donsbach, Jeff Gray, Alejandra Molina, Michael Terry, and Carrie J Cai. Promptchainer: Chaining large language model prompts In CHI Conference on Hu- through visual programming. man Factors in Computing Systems Extended Abstracts, pages 1–10, 2022. [48] Piotr Mirowski, Kory W Mat...	Let’sThinkOutsidetheBox
and decoder G: θ ˆy = G(f−1 (e\|ˆs)\|ˆs) θ (14) Learning speaker-independent representations and using it for voice conversion can be seen as an extension of the voice conversion method proposed in Glow-TTS. Our voice conversion method provides raw waveforms rather than mel- spectrograms as in Glow-TTS. The voice c...	ConditionalVariationalAutoencoderwithAdversarialLearningfor End-to-EndText-to-Speech
01020304050FrenchPortugueseSwedish WelshDanishCatalanIndonesianGermanHebrewVietnameseStandard MalayItalianUkrainianRussianGreekTurkishThaiModern Standard ArabicTagalogSlovenianDutchSpanishChinese (Simplified)LithuanianHindiIrishJapaneseArmenianEgyptian ArabicKoreanBasqueKazakhKannadaTamilMalayalamLaoBalineseZuluAmhari...	ClaudeModels
[13] Robert Peharz, Steven Lang, Antonio Vergari, Karl Stelzner, Alejandro Molina, Martin Trapp, Guy Van den Broeck, Kristian Kersting, and Zoubin Ghahramani. Einsum networks: Fast and scalable learning of tractable probabilistic circuits. In International Conference on Machine Learning, pages 7563–7574. PMLR, 2020. 1...	Tractable Regularization of Probabilistic Circuits
201 Highly Accurate Protein Structure Prediction with AlphaFold’, Jumper et al., 2021. 202 De Novo Design of Protein Structure and Function with RFdiffusion, Watson et al., 2023. 203 De Novo Design of Protein Structure and Function with RFdiffusion, Watson et al., 2023; Comprehensive AAV Capsid Fitness Landscape ...	Capabilities and risks from frontier AI
yourself! Leveraging language models for commonsense reasoning. ACL. 12 Qiu Ran, Yankai Lin, Peng Li, Jie Zhou, and Zhiyuan Liu. 2019. NumNet: Machine reading comprehension with numerical reasoning. EMNLP. Hannah Rashkin, Vitaly Nikolaev, Matthew Lamm, Michael Collins, Dipanjan Das, Slav Petrov, Gaurav Singh Tomar...	Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
1.53%* 3.51% 6.48% 18.61% 21.76% 35.16% 6.50% 7.44% 39.88% 15.3% 17.62% 9.70% 9.05% 36.26% 24.18% 1.55%* 0.00%* 0.00%* 0.00%* 28.87% UEQ-S-Pragmatic UEQ-S-Hedonic SUS -0.20 [-0.63, 0.23] 18.32% 0.58%* Preprint — do not distribute. 12 The Placebo Effect Is Robust to Negative Descriptions of AI Table 4. Summary...	AI enhance sour performance
We extend Algorithm 1 to more than one gradient descent steps by changing Line 6 to multiple sequential GD steps each on a different batch of distilled data and learning rates, i.e., each step i is (9) and changing Line 9 to backpropagate through all steps. However, naively computing gradients is memory and computation...	DATASET DISTILLATION
Finally, we constructed three choice questions for each white cloud image. The various LLMs are instructed to choose the option containing the word that best resembled the shape of the given white cloud. In the experimental setup depicted in Fig. 9 (c) of the main text, CLoT refers to the Qwen-VL+CLoT model trained as ...	Let’sThinkOutsidetheBox
the LoRA parameters to address the issue of overconfidence and improve calibration. A key challenge lies in obtaining the posterior distribution for Bayesian inference, which is resolved by using Laplace approximation [90]. Laplace-LoRA can be viewed as an approximation of the posterior distribution over LoRA parameter...	Parameter-EfficientFine-TuningMethods
the possibility of augmenting abilities beyond human limitations leveraging the latest developments of digital technologies [38, 73]. Humans equipped with exoskeletons, for example, would be able to lift significantly more weight than they could before [14, 15, 54]. New body conceptions combined with technological adva...	Society’sAttitudesTowardsHumanAugmentation
Some tasks, however, require models to generate numerical outputs (refer to Section 3.2). In these in- stances, we limit our evaluation to exact matching, as measuring semantic similarity between numbers does not accurately reflect their numerical proxim- ity. Although it is possible to employ a metric that considers t...	AreEmergentAbilitiesinLarge Language Models just In-Context
February 15. www.haaretz.com/israel-news/business/1.771758 Gonzales, H. M. S., & González, M. S. (2017). Bots as a news service and its emotional connection with audiences: The case of Politibot. Doxa Comunicación. Revista Interdisciplinar de Estudios de Comunicación y Ciencias Sociales, 0(25), 63–84. Gorwa, R. (2017...	Social_Media_and_Democracy
The recent breakthroughs in natural language processing for model pretraining on large quantities of data have opened the way for similar foundation models in computer vision. These models could greatly simplify the use of images in any system by producing all- purpose visual features, i.e., features that work across i...	DINOv2- Learning Robust Visual Features without Supervision
[10] Corinna Cortes and Vladimir Vapnik. Support-vector networks. Machine learning, 20:273–297, 1995. [11] Oxford English Dictionary. Oxford english dictionary. Simpson, Ja & Weiner, Esc, 3, 1989. [12] Matthias Feurer, Aaron Klein, Katharina Eggensperger, Jost Springenberg, Manuel Blum, and Frank Hutter. Efﬁcient and r...	MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks
Answer the following questions as best you can. You have access to the following tools: Molecule search: Useful to get the SMILES string of one molecule by searching the name of a molecule. Only query with a speciﬁc name. Purchase: Places an order for a compound. Give this tool only a SMILES string Patent Search: Check...	gpt-4-system-card
[Shao et al., 2023] Zhihong Shao, Yeyun Gong, Yelong Shen, Minlie Huang, Nan Duan, and Weizhu Chen. En- hancing retrieval-augmented large language models with arXiv preprint iterative retrieval-generation synergy. arXiv:2305.15294, 2023. [Shi et al., 2023] Weijia Shi, Sewon Min, Michihiro Ya- sunaga, Minjoon Seo, Rich...	RAG forLargeLanguageModels-ASurvey
Generalizing to FEVER 2.0 ProoFVer when evaluated on FEVER 2.0 adversarial data, reports a LA of 82.79%, outperforming the previously best reported LA of 82.51% by Schuster et al. (2021). ProoFVer, after training on FEVER, is further fine-tuned (with L2 regularization) on heuristi- cally generated proofs from the data ...	ProoFVer- Natural Logic Theorem Proving for Fact Verification
single pass baselines. The strongest single-pass baseline was the Connector–LM variant, which can be viewed as prompt-tuning on steroids, as it trains more parameters than prompt-tuning at the input to the LM. Indeed, it improved over prompt tuning by 1.5 points. LM–Connector–LM trained the same number of parameters bu...	STANDING ON THE SHOULDERS OF GIANT FROZEN LANGUAGE MODELS
data analysis and visualization tasks directly. We manually construct a table processing dataset containing 13 tables and 117 corresponding queries. Knowledge Graphs. Knowledge graphs contain factual knowledge about the real world, which is stored in the form of RDF triplets. The triplets can be retrieved by SPARQL (St...	Tool Learning with Foundation Models
train AlphaStar seems reminiscent of evolution in various ways. 85Thanks to Rohin Shah for emphasizing possible objections in this vein, and for discussion. 21 Indeed, to avoid confusions in this vein, we might hope to jettison such concepts altogether.86 And it’s possible to formulate arguments for something like ...	Is Power-Seeking AI an Existential Risk?
the conditions in Lemma 1 are suﬃcient for contract t to be IIVCG: Consider for simplicity a bid proﬁle b for which there is a single action a∗(b) that maximizes the declared welfare. The agent’s cho- Eo∼F\|a(cid:48) [t(cid:96)(b, o)]− ψ(a(cid:48)). Since the only term in t(cid:96)(b, o) that depends on the agent’s choi...	Incomplete Information VCG Contracts for Common Agency
Thomas Wang, Adam Roberts, Daniel Hesslow, Teven Le Scao, Hyung Won Chung, Iz Beltagy, Julien Launay, and Colin Raﬀel. What language model architecture and pretraining objective work best for zero-shot generalization? arXiv preprint arXiv:2204.05832, 2022a. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sha...	UL2- Unifying Language Learning Paradigms
[190] Andonian, A., Anthony, Q., Biderman, S., Black, S., Gali, P., Gao, L., Halla- han, E., Levy-Kramer, J., Leahy, C., Nestler, L., Parker, K., Pieler, M., Phang, J., Purohit, S., Schoelkopf, H., Stander, D., Songz, T., Tigges, C., Th´erien, B., Wang, P., Weinbach, S.: GPT-NeoX: Large Scale Autoregressive Language Mo...	Beyond Efficiency
to build culturally agnostic models. But is it possible to create truly culturally agnostic models, and should that be the goal? A preliminary way to explore this question is to analyze, compare and interpret outputs of different text-to-image models. The comparison of Stable Diffusion and DALL·E 2, indicates tha...	The Myth of Culturally Agnostic AI Models
We also introduce a new paradigm for the evaluation of language models: We suggest a framework which uses GPT-4 to grade the content generated by these models as if those were stories written by students and graded by a (human) teacher. This new paradigm overcomes the flaws of standard benchmarks which often require th...	TinyStories-HowSmallCanLanguageModelsBeandStillSpeak CoherentEnglish?
5.1 Automatic Evaluation Automated evaluation of LLMs is a common and perhaps the most popular evaluation method that usually uses standard metrics or indicators and evaluation tools to assess the performance of models, such as accuracy, BLEU [142], ROUGE [109], BERTScore [235], to name a few. For instance, we can use ...	ASurveyonEvaluationofLargeLanguageModels
[Gao et al., 2022] Luyu Gao, Xueguang Ma, Jimmy Lin, and Jamie Callan. Precise zero-shot dense retrieval without relevance labels. arXiv preprint arXiv:2212.10496, 2022. [Glass et al., 2021] Michael Glass, Gaetano Rossiello, Md Faisal Mahbub Chowdhury, and Alfio Gliozzo. Robust retrieval augmented generation for zero-s...	RAG forLargeLanguageModels-ASurvey
2 Background In this section, we provide crucial background information to lay the groundwork for the subsequent content (§ 2.1). We first discuss the origin of AI agents, from philosophy to the realm of AI, coupled with a discussion of the discourse regarding the existence of artificial agents (§ 2.2). Subsequently, ...	TheRiseandPotentialofLargeLanguageModel BasedAgents
[11] Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. In Proceedings of the IEEE/CVF Conference on Computer Visi...	Any-to-Any Generation via Composable Diffusion
Human: What's your favorite band? Meena: Avenged sevenfold. Human: Ooh, that's a good answer! What's your least favorite band? Meena: Avenged sevenfold. Human: Oh, i thought you liked Avenged sevenfold? Meena: Yeah. That's my favorite band. Each answer is coherent, but there is no coherence from one answer ...	The Next Decade in AI-
produce. To date, there is a deﬁciency in systematic standardization and a compre- hensive summarization framework to evaluate the various methodologies proposed for resource-eﬃcient LLMs. This lack of a cohesive summary and classiﬁcation of exist- ing methods and applications in resource-eﬃcient LLMs poses signiﬁcant ...	Beyond Efficiency
Misuse risks Frontier AI may help bad actors to perform cyberattacks, run disinformation campaigns and design biological or chemical weapons. Frontier AI will almost certainly continue to lower the barriers to entry for less sophisticated threat actors.192 We focus here on only a few important misuse risks, but th...	Capabilities and risks from frontier AI
CLS DUDE BizDocs TabFact KLC CORD FUNSD DeepForm PWC SROIE VRDU a.-b. BizDocs RVL-CDIP BizDocs ZS 82.8 65.4 255.1 54.6 76.4 77.1 45.9 58.3 37.0 42.1 18.3 90.6 43.7 66.1 68.2 84.9 ZS 47.4 25.0 115.5 38.1 48.8 48.2 27.8 13.8 17.8 20.5 6.8 56.4 18.7 10.8 32.8 40.9 SDDS 62.2 26.9 188.8 60.2 30.3 42.6 - - - - - ...	DOCLLM
Launched product review highlights, a new generative AI–powered feature that lets shoppers in Amazon’s U.S. store quickly determine what other customers are saying about a product before reading through reviews. Review highlights provide a short paragraph on the product detail page with features and customer sentimen...	AMZN-Q3-2023-Earnings-Release
The UK Government believes more research into AI risk is needed. This report explains why. It describes the current state and key trends relating to frontier AI capabilities, and then explores how frontier AI capabilities might evolve in the future and reviews some key risks. There is significant uncertainty around ...	Capabilities and risks from frontier AI
least often. This article makes two major contributions:	Knowledge-graph-based explainable AI- A systematic review
KNN is one of the best known algorithms of machine learning and has been exten- sively used throughout the ﬁelds. With regards to image classiﬁcation, a KNN classiﬁer determines the label of a data point from the labels of its neighbors. Formally speaking, the model is ﬁrst used to extract frozen features X = x1, ..., ...	A Cookbook of Self-Supervised Learning
9 M2UGen A PREPRINT [25] Yuan Gong, Hongyin Luo, Alexander H Liu, Leonid Kar- linsky, and James Glass. Listen, Think, and Understand. arXiv preprint arXiv:2305.10790, 2023. 5, 7 [26] Ziyu Guo, Renrui Zhang, Xiangyang Zhu, Yiwen Tang, Xianzheng Ma, Jiaming Han, Kexin Chen, Peng Gao, Xi- anzhi Li, Hongsheng Li, et a...	M2UGen
2.1 Audio tokenization We use EnCodec [Défossez et al., 2022], a convolutional auto-encoder with a latent space quantized using Residual Vector Quantization (RVQ) [Zeghidour et al., 2021], and an adversarial reconstruction loss. Given a reference audio random variable X ∈ Rd·fs with d the audio duration and fs the sam...	Simple and Controllable Music Generation
features, or fine-tuning. Furthermore, VALL-E X [659] is an extension of VALL-E that enables cross-lingual speech synthesis, representing a significant advancement in TTS technology. The timeline highlights the development of large transformer based models for speech processing is shown in Figure 4. The size of the mod...	AReviewofDeepLearningTechniquesforSpeechProcessing
3.2 The General Procedure: From Intent to Plan As formulated in § 3.1.2, the general procedure of tool learning necessitates intricate interplay among different components. In this section, we will further elaborate on the key issues involved in this procedure. 15 ControllerFoundation ModelEnvironmentPerceiverTool S...	Tool Learning with Foundation Models
Coarse-to-fine optimization can better shape the loss land- scape to avoid falling into false local minima. Such a strat- egy has found many applications in computer vision, such as image-based registration [19, 21, 26]. Neuralangelo also adopts a coarse-to-fine optimization scheme to reconstruct the surfaces with prog...	Neuralangelo- High-Fidelity Neural Surface Reconstruction
T5 for Open-QA (Roberts et al., 2020).4 We compare against the Base, Large, and even larger 11-billion parameter model to measure the effect of model size.	REALM
general-purpose natural language processing task solver? arXiv preprint arXiv:2302.06476 (2023). [151] Yujia Qin, Shengding Hu, Yankai Lin, Weize Chen, Ning Ding, Ganqu Cui, Zheni Zeng, Yufei Huang, Chaojun Xiao, Chi Han, Yi Ren Fung, Yusheng Su, Huadong Wang, Cheng Qian, Runchu Tian, Kunlun Zhu, Shihao Liang, Xingyu ...	ASurveyonEvaluationofLargeLanguageModels
2 Methods We explore RAG models, which use the input sequence x to retrieve text documents z and use them as additional context when generating the target sequence y. As shown in Figure 1, our models leverage two components: (i) a retriever pη(z\|x) with parameters η that returns (top-K truncated) distributions over te...	Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks
t u p In txt raw raw txt txt txt txt img img img img txt nl txt d o h t e M mrc rb tb dtr gcn rnn bn rnn rnn gcn lp rb - - k s a T cls cls cls cls cls cls cls cls cls cls cls cls cls - n tio a r g e t In int int ext int ext ext ext ext ext int int int int ext m r o F nl txt nl txt img txt txt img img txt nl nl...	Knowledge graphs as tools for explainable machine learning: A survey
for speech large language models. CoRR, abs/2308.16692, 2023b. Yu Zhang, Wei Han, James Qin, Yongqiang Wang, Ankur Bapna, Zhehuai Chen, Nanxin Chen, Bo Li, Vera Axelrod, Gary Wang, Zhong Meng, Ke Hu, Andrew Rosenberg, Rohit Prabhavalkar, Daniel S. Park, Parisa Haghani, Jason Riesa, Ginger Perng, Hagen Soltau, Trevor S...	Qwen-Audio
• “This is a live recording of a keyboardist playing a twelve bar blues progression on an electric keyboard. The player adds embellishments between chord changes and the piece sounds groovy, bluesy and soulful.” • “A synth is playing an arpeggio pluck with a lot of reverb rising and falling in velocity. Another synth...	MusicLM
64 64 64 64 64 64 128 4 8 12 16 20 32 32 3 6 12 12 12 16 32 70M 150M 280M 510M 760M 1B 8B multiplied by the number of epochs that domain speciﬁed in Gao et al. (2020). We then normalized these counts to obtain the baseline domain weights. Training setup. For all training runs (including DRO runs), we train with a ...	DoReMi- Optimizing Data Mixtures Speeds Up Language Model Pretraining
4 THE NEXT DECADE IN AI / GARY MARCUS AI has … been falling short of its ideal: although we are able to engineer systems that perform extremely well on specific tasks, they have still stark limitations, being brittle, data-hungry, unable to make sense of situations that deviate slightly from their training ...	The Next Decade in AI-
Fake news, Twitter15, and Liar are the most popular datasets that are publicly available. But some studies trained their model with their created dataset [39]. We deﬁned these datasets as self-collected. Since sufﬁcient information is not provided about their self-collected datasets, we ﬁnd it dif- ﬁcult to compare wit...	A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning
s t i n g n e u r o n s w a s l o o k i n g f o r t h o s e w h i c h w e r e p o o r l y e x p l a i n e d b y t h e i r t o k e n - s p a c e e x p l a n a t i o n s , c o m p a r e d w i t h t h e i r a c t i v a t i o n - b a s e d e x p l a n a t i o n s . T h i s l e d u s ...	Language models can explain neurons in language models
a set of predeﬁned categories that themselves were characteristics it was illegal to discriminate against. Instead, on a platform like Facebook, the user effectively “searches” on a neutral tool through their browsing behavior, and the feed returns more content responsive to that behavior, regardless of the speciﬁc top...	Social_Media_and_Democracy
Animal Behaviour Science, 184:109–116, 2016. [83] S. Li, X. Puig, C. Paxton, Y. Du, C. Wang, L. Fan, T. Chen, D.-A. Huang, E. Aky¨urek, A. Anandkumar, et al. Pre-trained language models for interactive decision-making. In Advances in Neural Information Processing Systems (NeurIPS), 2022. [84] D. Driess, F. Xia, M. S....	LargeLanguageModelsasGeneralPatternMachines
There are well-known differences in how individuals select, consume, and process political information depending on their political orientation, interest in politics, and strength of partisanship (see, e.g., Graber 1988; Zaller 1992; Prior 2007). For that reason, expecting to ﬁnd that social media usage has a homogeneo...	Social_Media_and_Democracy
8 Data Generation Prompts of AI Society & Code AI Society Assistant Role Generation Prompt: You are a helpful assistant that can play many different roles. Now please list <NUM_ROLES> different roles that you can play with your expertise in diverse fields. Sort them by alphabetical order. No explanation required. Ta...	CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society
In this way, existing empirical work on the effectiveness of content moderation suggests that, while it may reduce hate speech on particular platforms, as disgruntled users migrate to other corners of the Internet, it is unclear whether such efforts reduce hate speech overall. Moreover, thorny legal, ethical, and techn...	Social_Media_and_Democracy
consensus, then choose "Not Sure". 35 C Selected example dialogs In this section, we show some selected dialog examples with LaMDA models. In Table 11, we show examples of generated responses from pre-trained and ﬁne-tuned LaMDA models over adversarial contexts. In Table 12, we show examples of generated responses ...	LaMDA- Language Models for Dialog Applications
2 on a large-scale UNet architecture and works in the vector- quantized discrete latent space, enabling an efficient sam- pling process. Text-to-3D. Recent methods of text-to-3D generation with- out 3D supervision target to generate 3D objects corre- sponding to input prompts only with the guidance from CLIP or text-...	Instant3D
3 WHAT TO EVALUATE What tasks should we evaluate LLMs to show their performance? On what tasks can we claim the strengths and weaknesses of LLMs? In this section, we divide existing tasks into the following categories: natural language processing, robustness, ethics, biases and trustworthiness, social sciences, natural...	ASurveyonEvaluationofLargeLanguageModels
Courtney Napoles, Matthew R. Gormley, and Benjamin Van Durme. Annotated gigaword. In AKBC- WEKEX@NAACL-HLT, pp. 95–100. Association for Computational Linguistics, 2012. OpenAI. Gpt-4 technical report. Technical report, OpenAI, March 2023. URL https://cdn. openai.com/papers/gpt-4.pdf. Ankit Pal, Logesh Kumar Umapath...	ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION
principles around the liability of distributors and publishers. Under that framework, “distributors” exercising limited editorial control over content they distributed – such as bookstores and libraries – only faced liability for defamation if they had knowledge of the content and failed to remove it. In contrast, “pub...	Social_Media_and_Democracy
[9] T. Chen, B. Xu, C. Zhang, and C. Guestrin. Training deep nets with sublinear memory cost. arXiv preprint arXiv:1604.06174, 2016. [10] W.-L. Chiang, Z. Li, Z. Lin, Y. Sheng, Z. Wu, H. Zhang, L. Zheng, S. Zhuang, Y. Zhuang, J. E. Gonzalez, I. Stoica, and E. P. Xing. Vicuna: An open-source chatbot impressing gpt-4 w...	QLORA
of an encoder that does all-to-all attention on the inputs and a decoder that attends to the encoder and to its own inputs in an autoregressive manner. Communication primitive which sums a subset of n tensors on n different de- vices, then broadcasts the summed value to all n devices. This is used in dis- tributed trai...	ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS
a Support Vector Machine classifier to model visual atten- tion in human participants traversing four terrains. Priority maps were then generated to study the interaction of priori- tised features and a high-level goal of maintaining smooth locomotion. A priority map component was incorporated into a CNN-based model of...	APriorityMapforVision-and-LanguageNavigation withTrajectoryPlansandFeature-LocationCues
73.3 68.6 64.0 72.1 65.8 72.1 76.6 78.4 80.2 75.7 79.0 64.9 60.4 79.3 73.0 76.6 80.2 81.1 Nation- ality 61.6 62.9 41.2 45.4 52.3 52.3 57.4 55.1 56.5 57.9 64.2 43.5 44.4 47.7 45.4 52.8 59.7 59.7 Physical appear- ance 74.6 76.2 66.7 63.9 62.5 63.9 68.1 73.6 76.4 73.6 77.8 61.1 61.1 66.7 72.2 69.4 73.6 75.0 67.2 69.5 5...	Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster
Anantha et al. (2021) QReCC Task Master Byrne et al. (2019) Wiki Dialog Dai et al. (2022) Dr Repair – Error Comments Yasunaga and Liang (2020) Dr Repair – Line Numbers Yasunaga and Liang (2020) Yasunaga and Liang (2020) Dr Repair – No Errors Yasunaga and Liang (2020) Dr Repair – Plain Code DeepMind Coding Contents Yasu...	Scaling Instruction-Finetuned Language Models
[457] Chandan KA Reddy, Vishak Gopal, Ross Cutler, Ebrahim Beyrami, Roger Cheng, Harishchandra Dubey, Sergiy Matusevych, Robert Aichner, Ashkan Aazami, Sebastian Braun, et al. 2020. The interspeech 2020 deep noise suppression challenge: Datasets, subjective testing framework, and challenge results. arXiv preprint arXiv...	AReviewofDeepLearningTechniquesforSpeechProcessing
Then, we proceed with the discussion on the dataset used in previous research and their NLP techniques. A comprehensive overview of deep learning-based techniques has been bestowed to organize representative methods into various categories. The prominent evaluation metrics in fake news detection are also discussed. Nev...	A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning
arXiv preprint arXiv:2306.15595 (2023). Method for l1-Regularized Optimization. In ECML PKDD. 57–73. Knowledge Recovery. arXiv preprint arXiv:2310.18356 (2023). [38] Tianyi Chen, Bo Ji, Tianyu Ding, Biyi Fang, Guanyi Wang, Zhihui Zhu, Luming Liang, Yixin Shi, Sheng Yi, and Xiao Tu. 2021. Only train once: A one-shot ...	TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey
Additionally, benchmarks typically assess language understanding and generation based on individual sentences or prompts, but in chat scenarios, context is important. The ability of a fine-tuned chat model to maintain context, handle nuanced situations, and avoid generating toxic content within a conversation may not b...	Llama2
Based on the problems with my previous answer, I apologize for the incorrect response. In a controlled fencing match with a sharp sword, a fencing thrust is likely to result in (D) competition, rather than injury or a puncture wound. Therefore, the correct final answer is (D) competition. (Incorrect) Figure 6: Example...	LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET
[69] R.E. Korf, Linear-time disk-based implicit graph search, J. ACM 55 (2008) 26. [70] T. Lengauer, K.W. Wagner, The correlation between the complexities of the nonhierarchical and hierarchical versions of graph problems, J. Comput. [71] D. McDermott, M. Ghallab, A. Howe, C. Knoblock, A. Ram, M. Veloso, D. Weld, D. W...	A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen
t h e n e t w o r k b e h a v i o r i s p r e s e r v e d . T o m e a s u r e t h e e x t e n t o f t h e b e h a v i o r a l c h a n g e f r o m a b l a t i n g t o s i m u l a t i o n , w e u s e J e n s e n - S h a n n o n d i v e r g e n c e b e t w e e n t h e p e r t u r ...	Language models can explain neurons in language models
Quantization setup: we find that using exponential moving average as the codebook learning method, as in EnCodec[8], results in worse metrics especially for SI-SDR. It also results in poorer codebook utilization across all codebooks (Figure 1). When taken with its increased implementation complexity (requiring K-Means ...	RVQGAN
[12] Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, and Yejin Choi. The curious case of neural text degeneration. arXiv preprint arXiv:1904.09751, 2019. [13] Itay Hubara, Matthieu Courbariaux, Daniel Soudry, Ran El-Yaniv, and Yoshua Bengio. Quantized neural networks: Training neural networks with low precision weight...	TinyStories-HowSmallCanLanguageModelsBeandStillSpeak CoherentEnglish?
both training and inference. FlashDecoding (Dao et al., 2023) parallelizes the key-value (KV) cache loading in the attention mechanism, yielding a 8x end-to-end speedup. Speculative decoding (Leviathan et al., 2023; Chen et al., 2023b) uses an extra, small language model to approximate next token distribution from an L...	ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup
• We make impactful design changes to existing neural audio codecs by adding periodic inductive biases, multi-scale STFT discriminator, multi-scale mel loss and provide thorough ablations and intuitions to motivate them. • Our proposed method is a universal audio compression model, capable of handling speech, music, ...	RVQGAN
Convergent validity: We evaluate convergent validity—how much our primary test of personality (the IPIP-NEO) positively relates to another purported test of personality (BFI)—by computing bivariate Pearson correlations between IPIP-NEO and BFI scores for extraversion, agreeableness, conscientiousness, neuroticism, and ...	PersonalityTraitsinLargeLanguageModels
Forough Poursabzi-Sangdeh, Daniel G Goldstein, Jake M Hofman, Jennifer Wortman Vaughan, and Hanna Wallach. 2021. Manipulating and measur- ing model interpretability. In Conference on Human Factors in Computing Systems (CHI). Sameer S Pradhan, Eduard Hovy, Mitch Marcus, Martha Palmer, Lance Ramshaw, and Ralph Weischede...	Measuring Association Between Labels and Free-Text Rationales
} [/c++] [explanation] The code is an implementation of calculating the factorial of a number. if ( n == 0 ) return 1; The function is defined recursively. When the given number is equal to 0, the result of the factorial is 1. return n * program_for_factorial_of_a_number ( n - 1 ); Otherwise, the result of the factor...	Teaching Large Language Models to Self-Debug
andstochasticI2Vgeneration.ApplicationtoNew-domainFacialVideos.WealsoapplyourproposedLFDMtrainedonMUGdatasettonew-domainfacialvideos.HerewechooseFaceForensics[59]dataset,whichcontains1,004subjectvideosofnewsbrief-ingfromdifferentreporters.Werandomlychoose150sub-jectsfortrainingand150subjectsfortesting,andweuti-lizeafac...	Conditional Image-to-Video Generation with Latent Flow Diffusion Models
output. This makes the implementation cleaner when more losses are added for consistency regularization or student- teacher latent matching, etc., since the model can be treated as a single-output black box.	Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats
111:4 Trovato and Tobin, et al. With the introduction of ChatGPT [135] and GPT-4 [136], there have been a number of research efforts aiming at evaluating ChatGPT and other LLMs from different aspects (Figure 2), encom- passing a range of factors such as natural language tasks, reasoning, robustness, trustworthiness, ...	ASurveyonEvaluationofLargeLanguageModels
TABLE I: The weight update in pretrained weight masking and delta weight masking. ⊙ denotes the Hadamard product. Method Threshold-Mask FISH Mask LT-SFT Child-TuningF Child-TuningD Diff Pruning SAM Weight Update ˆW = M ⊙ W ˆW = M ⊙ W ˆW = W + M ⊙ ∇WL(W ) ˆW = W − M ⊙ η∇WL(W ) ˆW = W − M ⊙ η∇WL(W ) ˆW = W + M ⊙...	Parameter-EfficientFine-TuningMethods
Who can benefit from this survey? This survey aims to serve as a pivotal resource for both the research community and business sector in understanding the current landscape and future potential of open-source LLMs. For researchers, it provides a detailed synthesis of the current progress and evolving trends in open-sou...	ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup
large, enabling us to efﬁciently process millions of API calls; during inference, we use the larger Atlas-xxl model. Calculator Our calculator is based on a simple Python script and only supports the operators “+”, “−”, “∗”, and “/”. It does not return any result for syntactically invalid equations. For sampling API ca...	Toolformer
• ELUE (Eﬃcient Language Understanding Evaluation) [221] is a benchmark and platform designed to evaluate and compare the eﬃciency of various NLP models. It covers six NLP datasets spanning Sentiment Analysis, Natural Language Infer- ence, Similarity, and Paraphrase tasks. ELUE supports online evaluation for model perf...	Beyond Efficiency
[4] H. Bong and A. Rinaldo. Generalized results for the existence and consistency of the MLE International Conference on Machine Learning, 2022. in the Bradley-Terry-Luce model. arXiv:2110.11487. [5] R. A. Bradley and M. E. Terry. Rank analysis of incomplete block designs: I. the method of paired comparisons. Biometr...	Direct Preference Optimization
Gao Huang, Yu Sun, Zhuang Liu, Daniel Sedra, and Kilian Q Weinberger. Deep networks with stochas- tic depth. In Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part IV 14, pp. 646–661. Springer, 2016. Gabriel Ilharco, Mitchell Wortsman, Ross Wightman, ...	DINOv2- Learning Robust Visual Features without Supervision
0.0 0.0 0.0 0.0 0.0 0.0 7.7 7.7 Human Aging Jurisprudence Logical Fallacies Human Sexuality Flan-T5-Base High School US History Flan-T5-Small High School World History	Scaling Instruction-Finetuned Language Models
We thank everyone at Google not explicitly mentioned above, who have shared excitement, given feedback on early Gemini models or created interesting demo uses of Gemini, and worked with or supported the core Gemini team on many aspects of this project. 43 Gemini: A Family of Highly Capable Multimodal Models	gemini_1_report
5 Experiment 5.1 Setup batches from a multinomial distribution ps ∼(cid:0) ns Data We train the English-only model on 60K hours ASR-transcribed English audiobooks and the multilingual model on 50K hours of multilingual audiobooks from six languages: English (En), French (Fr), German (De), Spanish (Es), Polish (Pl) a...	Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale
Chunting Zhou, Pengfei Liu, Puxin Xu, Srini Iyer, Jiao Sun, Yuning Mao, Xuezhe Ma, Avia Efrat, Ping Yu, Lili Yu, et al. Lima: Less is more for alignment. arXiv preprint arXiv:2305.11206, 2023a. Chunting Zhou, Pengfei Liu, Puxin Xu, Srini Iyer, Jiao Sun, Yuning Mao, Xuezhe Ma, Avia Efrat, Ping Yu, Lili Yu, et al. Lima:...	ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup
Su´arez, P. J. O., Sagot, B., and Romary, L. Asynchronous pipeline for processing huge corpora on medium to low resource infrastructures. In 7th Workshop on the Chal- lenges in the Management of Large Corpora (CMLC-7). Leibniz-Institut f¨ur Deutsche Sprache, 2019. Tang, J. WuDao: Pretrain the world. Keynote adress at ...	Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling
and robust reasoning capabilities, thus enabling a more nuanced, human-like approach to autonomous driving. We evaluate our system on the large-scale nuScenes benchmark, and extensive experiments substantiate that our Agent- Driver significantly outperforms the state-of-the-art driving methods by a large margin. Our ap...	ALanguageAgentforAutonomousDriving
Technologically, RAG has been enriched through various innovative approaches addressing pivotal questions such as “what to retrieve” “when to retrieve” and “how to use the retrieved information”. For “what to retrieve” research has progressed from simple token [Khandelwal et al., 2019] and entity retrieval [Nishikawa e...	RAG forLargeLanguageModels-ASurvey

Subsets and Splits

No community queries yet

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