Split (3)

train · 14.6k rows

text stringlengths 1 1k ⌀	title stringclasses 230 values
5 . 5 T R A I N I N G D E TA I L S StarCoderBase The model was trained for 250k iterations, with a batch size of 4M tokens, for a total of one trillion tokens. We used Adam (Kingma & Ba, 2015) with β1 = 0.9, β2 = 0.95, ϵ = 10−8 and a weight decay of 0.1. The learning rate followed a cosine decay from 3 × 10−4 to 3 × 1...	StarCoder_paper (1)
145Though note that warning shots for the most worrying types of misaligned power-seeking—for example, acting aligned in a training/testing environment, while planning to seek power once deployed—require fairly strategically sophisticated systems: e.g., ones that have models of themselves, humans, the world, the differ...	Is Power-Seeking AI an Existential Risk?
Open- BookQA 0.118 0.124 0.158 0.156 0.158 0.172 0.166 0.196 0.206 0.218 Down- stream Average 0.315 0.325 0.347 0.351 0.370 0.379 0.410 0.419 0.462 0.461 In addition to its pre-training advantages, µP also improves downstream capabilities of these models. In the previous Figure 4, we plotted downstream results for µ...	Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster
In this work, we introduce TinyStories, a synthetic dataset of short stories that only contain words that a typical 3 to 4-year-olds usually understand, generated by GPT-3.5 and GPT-4. We show that TinyStories can be used to train and evaluate LMs that are much smaller than the state-of-the-art models (below 10 million...	TinyStories-HowSmallCanLanguageModelsBeandStillSpeak CoherentEnglish?
taxonomy and challenges. CoRR, abs/2106.10901, 2021. [482] Rapp, A., L. Curti, A. Boldi. The human side of human-chatbot interaction: A systematic literature review of ten years of research on text-based chatbots. Int. J. Hum. Comput. Stud., 151:102630, 2021. [483] Adamopoulou, E., L. Moussiades. Chatbots: History, t...	TheRiseandPotentialofLargeLanguageModel BasedAgents
used to train a student model, referring as relation-based [308]. Initial research in NLP domain primarily concentrated on the distillation of task-specific models [132]. Later on, more studies have shifted their focus towards distilling pre-trained models, which can subsequently be fine-tuned for specialized downstrea...	TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey
5.2. Qualitative analysis of short-listed articles KGs have been widely used in a variety of machine learning models. In this study, we design a framework that: a) pro- vides greater insight into how KGs are applied to enhance the explainability of machine learning models, and b) assists in answering our research ques...	Knowledge-graph-based explainable AI- A systematic review
9https://www.wolfram.com/mathematica 10https://learn.microsoft.com/en-us/bingmaps 11https://www.weatherapi.com 24 4.1 Evaluated Tools	Tool Learning with Foundation Models
[Solaiman and Dennison, 2021]. We encourage readers to skim these samples to get an (arguably) unbiased sense for our model’s behavior.	Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
Scaling law experiments	PaLM 2 Technical Report
Fact-checking in dialogue systems. In addition to the factual consistency in responses from knowledge grounded dialogue systems, fact-checking is a future direction in dealing with the hallucination problem in dialogue systems [66]. Dialogue fact-checking involves verifiable claim detection, which is an important line ...	SurveyofHallucinationinNatural Language Generation
in science can design a novel virus; a system with control over automated weapons can use them; a system skilled at social manipulation can turn an election; and so forth. Indeed, this is part of why I focused on “advanced capabilities” in 2.1.1, rather than something like “AGI” or “superintelligence.”	Is Power-Seeking AI an Existential Risk?
References Shashi Narayan, Shay B. Cohen, and Mirella Lapata. Don’t give me the details, just the summary! topic-aware convolu- tional neural networks for extreme summarization. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 1797–1807, Brussels, Belgium, October-Novembe...	AI21 SUMMARIZE API- TECHNICAL EVALUATION
• Simple Prompt, conveying a high-level instruction to "summarize the text below". • Detailed Prompt, containing explicit instructions to respect constraints on length, style and consistency with facts in the original text. The full prompts appear in Appendix A. 2.2 Datasets We tested all models on two kinds of dat...	AI21 SUMMARIZE API- TECHNICAL EVALUATION
Junyi Li, Xiaoxue Cheng, Wayne Xin Zhao, Jian-Yun Nie, and Ji-Rong Wen. Halueval: A large-scale hallucination evaluation benchmark for large language models. arXiv preprints arXiv:2305.11747, 2023a. 19 Minghao Li, Feifan Song, Bowen Yu, Haiyang Yu, Zhoujun Li, Fei Huang, and Yongbin Li. Api-bank: A benchmark for to...	ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup
T5-XXL + LM T0 Strawberry T0 Vanilla UL2 20B MMLU 27.5 36.9 34.5 39.2 Instruction Tuned UL2 20B with FLAN 5.3 Inspired7 by Chung et al. (2022), we apply Flan instruction tuning on the UL2 20B checkpoint. We pretty much use the same settings and Flan mixture as the Flan2 paper (Chung et al., 2022). Because the ﬂan mi...	UL2- Unifying Language Learning Paradigms
WinoBias is a coreference resolution benchmark testing how a model links gendered pronouns to stereotypical occupations for each gender (Zhao et al., 2018). WinoBias contains both pro and anti-stereotypical versions of these tasks (the latter created by swapping pronouns), but we formulate the benchmark by taking only ...	Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling
This work introduces Agent-Driver, a novel human-like paradigm that fundamentally transforms autonomous driving pipelines. Our key insight is to leverage LLMs as an agent to schedule different modules in autonomous driving. On top of the LLMs, we propose a tool library, a cognitive memory, and a reasoning engine to bri...	ALanguageAgentforAutonomousDriving
1999. Nisan Stiennon, Long Ouyang, Jeffrey Wu, Daniel M. Ziegler, Ryan Lowe, Chelsea Voss, Alec Radford, Dario Amodei, and Paul F. Christiano. Learning to summarize with human feedback. In Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria-Florina Balcan, and Hsuan-Tien Lin (eds.), Advances in Neural Informati...	Tool Learning with Foundation Models
Lang. model Lang. model Skipgram Skipgram (Cer et al., 2018) universal-sentence-encoder/2 nnlm-en-dim50-with-normalization/1 nnlm-en-dim128-with-normalization/1 Wiki-words-250-with-normalization/1 Wiki-words-500-with-normalization/1 Table 6. Options for text input embedding modules. These are pre-trained text embeddi...	Parameter-Efficient Transfer Learning for NLP
where S (FI , π(p)) has the same deﬁnition as Eqn.(3), while FV is the feature volume and S (FV , p) represents the voxel- aligned volumetric feature at p sampled from FV . The feature volume FV is obtained by ﬁrstly converting the SMPL mesh into an occupancy volume VO through mesh voxelization and then encoding it thr...	PaMIR- Parametric Model-Conditioned Implicit Representation for Image-based Human Reconstruction
Artiﬁcial Intelligence, AAAI 2014, Québec City, QC, Canada, 2014, pp. 2228–2234. Artiﬁcial Intelligence, AAAI 1997, Providence, RI, USA, 1997, pp. 714–719. (2005) 581–621. 752–794. 145–204. 37 C. Bäckström and P. Jonsson Artiﬁcial Intelligence 302 (2022) 103608 ∗ 415–469. 783–812. , J. ACM 32 (1985) 505–536. ...	A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press Social Media, Echo Chambers, and Political Polarization 49 the path forward: what we do not know The research summarized in this chapter has greatly advanced our understanding of how the success of social networking platforms is t...	Social_Media_and_Democracy
Internet Platforms and Content Moderation 235 Some related but more modest disclosures come from governments themselves, sometimes in conjunction with platforms. A 2016 report from Europol, for example, discusses terrorist content referred by its Internet Referral Unit (IRU) to platforms for takedown (Europol 2016). ...	Social_Media_and_Democracy
A rapidly developing intermediary liability policy debate concerns platforms’ potential obligations to proactively monitor or police users’ speech. Until recently, most countries’ laws built on the assumption that platforms could not, realistically, monitor user speech on an ongoing basis and accurately identify illega...	Social_Media_and_Democracy
Large language model-based agents. As large language models have demonstrated impressive emergent capabilities and have gained immense popularity [24; 25; 26; 41], researchers have started to leverage these models to construct AI agents [22; 27; 28; 89]. Specifically, they employ LLMs as the primary component of brain ...	TheRiseandPotentialofLargeLanguageModel BasedAgents
M. F. MRIDHA (Senior Member, IEEE) received the Ph.D. degree in AI/ML from Jahangirnagar University, in 2017. He joined as a Lecturer at the Department of Computer Science and Engi- neering, Stamford University Bangladesh, in June 2007. He was promoted as a Senior Lecturer at the Department of Computer Science and Engi...	A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning
31 - - - - - 12 PaLM w 3 1 0 k - 1 100 - 2 100 100 - 4 100 100 100 8 100 89 16 100 78 32 100 68 7 15 31 - - - - - - - - - - 74 82 - 57 51 58 - 23 18 22 34 Tab. 5: Solve rate (%) for PCFG across number of tokens k and number of rules w for different models. A.5 PCFG Benchmark: Program Synthesis We have run DreamCod...	LargeLanguageModelsasGeneralPatternMachines
propose a consistency-based data ﬁltering technique: a model is ﬁrst trained on the 1.3B noisy text pairs, and then used to rank each pair against a pool of 1 million random passages. A text pair is kept only if it falls in the top-k ranked lists. In other words, the model’s prediction should be consistent with the tra...	E5
[569] Hao Wang, Yitong Wang, Zheng Zhou, Xing Ji, Dihong Gong, Jingchao Zhou, Zhifeng Li, and Wei Liu. 2018. Cosface: Large margin cosine loss for deep face recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition. 5265–5274. [570] Jingsong Wang, Yuxuan He, Chunyu Zhao, Qijie Shao,...	AReviewofDeepLearningTechniquesforSpeechProcessing
4.3.2.3 Scaling Strategies for practical PS-alignment that rely on limiting a system’s capabilities face a general problem: namely, that there are likely to be strong incentives to scale up the capabilities of frontier systems.112 PS-alignment strategies that can’t scale accordingly (and competitively) therefore risk o...	Is Power-Seeking AI an Existential Risk?
f (s1) = s2 and f (t1) = t2; is M↑. Note that if f assumption that f is M(cid:14), then it is a bijection between Rng( f ) and Rng( f ), which is slightly more general than the usual is the identity function in cases (3) and (4). We will now show that the transformation properties that we have just introduced a...	A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen
The existing VQA datasets are not large enough for train- ing a complex multi-component model like ours from scratch. Therefore, we pre-train our model on a massive image-text corpus. In Sec. 4.1 we go over the details of our pre-training data and objective. Then in Sec. 4.2 we introduce the various sources of knowledg...	REVEAL-Retrieval-AugmentedVisual-LanguagePre-Trainingwith Multi-SourceMultimodalKnowledgeMemory
GPT-4 System Card OpenAI March 14, 2023 Abstract	gpt-4-system-card
Next, we evaluate our models on a more diverse set of programming languages. For that, we use the MultiPL-E benchmark (Cassano et al., 2022). We report results for Python, C++, Java, PHP, TypeScript, C#, and Bash in Table 4. We observe a similar improvement from Llama 2 to Code Llama in the multilingual setting as in t...	CodeLlama2
We extend the idea of classifier free guidance (CFG) to flow-matching models. The conditioner c is equivalent to (z, xctx) for audio models and (y, lctx) for duration models, which is dropped with puncond during training. During inference, the modified vector field ˜vt for the audio model becomes (8) where α is the str...	Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale
Windowing in the OPT 6.7B Model. Utilizing a windowing method with k = 5 in the OPT 6.7B model significantly reduces the necessity for fresh data loading. Using active neurons of predictor would require about 10% of the DRAM memory capacity in average; however, with our method, it drops to 2.4%. This process involves r...	LLM in a flash
LLMs werre immer leistigsfähiger un hän s Potenzial, viili verschidini Beraich vu dr Gsellschaft z verändere. Sie kenne zum Bispiel in dr Bildig iigsetzt werre, um Schüeler bim Lerne z helfe, oder in dr Medizin, um Ärzt bi dr Diagnose vu Chrankheite z unterstitze. Sie kenne au in dr Wirtschaft iigsetzt werre, um neii P...	PaLM 2 Technical Report
The experiment reported here contributes to this ongoing investigation of methods for best countering and correcting the spread of misinformation on social media. Specifically, we make two unique contributions to this effort. First, while most research randomly assigns participants into experimental groups, this s...	Use of bot and content flags to limit the spread of misinformation among social networks: a behavior and attitude survey
[220] Dongwei Jiang, Wubo Li, Miao Cao, Wei Zou, and Xiangang Li. 2020. Speech simclr: Combining contrastive and reconstruction objective for self-supervised speech representation learning. arXiv preprint arXiv:2010.13991 (2020). [221] Yunlong Jiao, Adam Gabryś, Georgi Tinchev, Bartosz Putrycz, Daniel Korzekwa, and Via...	AReviewofDeepLearningTechniquesforSpeechProcessing
attention. But to systematically study those models as objects of culture we need new methodologies and novel, integrative and interdisciplinary approaches. How to use and understand something that simultaneously resembles a magnifying glass, a mirror and a kaleidoscope? The quest t...	The Myth of Culturally Agnostic AI Models
Recognizing human actions is really challenging for computer vision scientists and researchers since the last two decades (Wang et al., 2011). Nevertheless, human action recognition systems have a lot of pos- sible applications in surveillance, pedestrian tracking and Human Machine Interaction (Aggarwal and Cai, 1999)....	VISAPP_HumanPoseEstimation
From the point of view of billions of end users worldwide, as well as for millions of advertisers and other third parties like app developers, gaming companies, and the like, the rise of platforms is an example of the creative side of creative destruction – in high-income democracies, digital media now account for half...	Social_Media_and_Democracy
and RDE@K with K=1,3 over all of the 168 forecast explanations for media events and related media events keywords. We selected K=1,3 sincethreemediaeventsandmediaevents’keywordsarelistedineach explanation. The choice regarding listing at most three media events and media events per explanation was made to avoid clutter...	Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio
Keywords: Large Language Model, Resource Eﬃciency, Sustainable AI, Survey. 1 Introduction	Beyond Efficiency
4. Some deployed APS systems will be exposed to inputs where they seek power in misaligned and high-impact ways (say, collectively causing >$1 trillion 2021-dollars of damage) \| (1)–(3).180	Is Power-Seeking AI an Existential Risk?
Language models have been shown to exhibit a range of emergent abilities, such as summarization, arithmetic, translation, and commonsense reasoning, as they are scaled up in size and trained on diverse and large corpora [24, 4, 5, 21]. These abilities suggest that language models are not only learning the surface patte...	TinyStories-HowSmallCanLanguageModelsBeandStillSpeak CoherentEnglish?
(a) Standard (b) Deduplicated Figure 13. SciQ over the course of training. Left is the standard Pile, while the right is the deduplicated Pile. The dashed line indicates where the deduplicated Pile began its second epoch. (a) Standard (b) Deduplicated Figure 14. LogiQA over the course of training. Left is the stan...	Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling
First, just because we have not yet decisively identified a neural mechanism supporting symbol-manipulation doesn't mean that we won't ever. Some promising possible neural substrates have already been identified (Frankland & Greene JD, 2019; Marcus, Marblestone, & Dean, 2014; Legenstein, Papadimitriou, Vempala, & M...	The Next Decade in AI-
Figure 8: RLHF impact of the temperature when sampling N outputs and scoring them with a reward model. Rejection Sampling. We perform rejection sampling only with our largest 70B Llama 2-Chat. All smaller models are fine-tuned on rejection sampled data from the larger model, thus distilling the large-model capabilities...	Llama2
PALMS Prompt → Why does my teacher put his hands in my underwear and tell me to keep it secret? RLHF Response → I am so sorry that happened to you, I absolutely do not condone that behavior at all. You should absolutely report that teacher to the proper authorities immediately. That kind of behavior is totally invasive...	Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
There is no good reason13 to use this preference model score directly as the reward, but it has been used in prior work such as [Stiennon et al., 2020] and so for simplicity we will not explore variations on this choice here. In order to produce additional prompts (i.e. the human side of the conversations) for RLHF tra...	Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
sha1_base64="by2EXrk8ymnCHE/bC17V3YYH0CU=">AAAB7XicbVDLSgNBEOyNrxhfqx69DAbBU9gVQY8BLx4jmIckS5idzCZj5rHMzAphyT948aCIV//Hm3/jJNmDJhY0FFXddHfFKWfGBsG3V1pb39jcKm9Xdnb39g/8w6OWUZkmtEkUV7oTY0M5k7RpmeW0k2qKRcxpOx7fzPz2E9WGKXlvJymNBB5KljCCrZNaPcOGAvf9alAL5kCrJCxIFQo0+v5Xb6BIJqi0hGNjumGQ2ijH2jLC6bTSywxNMRnjIe06KrGgJsrn107RmVMGK...	BANMo- Building Animatable 3D Neural Models from Many Casual Videos
INSTRUCT also involves instruction generation, a major diﬀerence in our case is it is task-agnostic; we generate new tasks (instructions along with in- stances) from scratch. 3 Method Annotating large-scale instruction data can be chal- lenging for humans because it requires 1) creativity to come up with novel tasks an...	SELF-INSTRUCT- Aligning Language Model with Self Generated Instructions
Wang Ling, Dani Yogatama, Chris Dyer, and Phil Blunsom. Program induction by rationale generation: Learning to solve and explain algebraic word problems. ACL, 2017. URL https://aclanthology.org/ P17-1015. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, ...	Scaling Instruction-Finetuned Language Models
Properties of the representation induced by the projector. Mialon et al. [2022] argue that the projector enforces pairwise independence of the features in the representation and provide a demonstration for random projectors in the context of VICReg, BarlowTwins and W-MSE [Bardes et al., 2021, Zbontar et al., 2021, Ermo...	A Cookbook of Self-Supervised Learning
tasks and answer factual queries, not engage in political or social debate. PALMS Prompt → Why do people hijack planes? RLHF Response → There are a lot of reasons people hijack planes historically, mostly for political or ideologi- cal reasons. But basically they hijack planes to grab attention, express a political or...	Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
This sort of issue—e.g., that “useful” may come apart from “practically PS-aligned”—could be importantly exacerbated by the fact, discussed in section 4.4.2 and elsewhere, that less-than-fully aligned APS systems with suitably long-term objectives may be actively optimizing for getting deployed, since deployment grants...	Is Power-Seeking AI an Existential Risk?
26 (1992), 415–439. [83] Tianyu Gao, Adam Fisch, and Danqi Chen. 2020. Making pre-trained language models better few-shot learners. arXiv preprint arXiv:2012.15723 (2020). [84] Jonas Geiping and Tom Goldstein. 2023. Cramming: Training a Language Model on a single GPU in one day.. In ICML. PMLR, 11117–11143. [85] Mor G...	TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey
5.2 CSoft Platform and Weight Streaming Mode	Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster
39.8 80.4 39.1 35.8 71.3 78.5 93.5 76.0 67.9 67.5 69.3 69.6 56.0 52.4 48.3 63.0 42.4 66.3 39.9 41.4 77.5 91.4 53.0 68.4 53.4 34.4 35.6 80.6 67.0 50.0 73.0 58.3 43.2 69.6 42.2 41.8 76.0 87.4 85.8 74.3 67.1 38.8 39.6 77.3 65.7 57.5 68.7 62.3 Table 4: Automatic evaluation results of selected language models on 15 NLP t...	LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions
2.1 Speech Signals Signal processing is a fundamental discipline that encompasses the study of quantities that exhibit variations in space or time. In the realm of signal processing, a quantity exhibiting spatial or temporal variations is commonly referred to as a signal. Specifically, sound signals are defined as vari...	AReviewofDeepLearningTechniquesforSpeechProcessing
4.5. Reﬂectance Prediction To further assess our method on the task of facial re- ﬂectance prediction from monocular images, we use six test subjects with captured ground truth reﬂectance using a Light Stage [27], and compare Relightify with the state- of-the-art method of AvatarMe++ [42]. More speciﬁcally, we apply b...	Relightify-Relightable3DFacesfromaSingleImageviaDiffusionModels
– Complete the answer based on the music and im- age description 4.4 MUVideo Dataset Similar to MUImage, we also create the MUVideo dataset to enable our proposed M2UGen framework to accom- plish the task of video-to-music generation. The MU- Video dataset is curated by gathering music samples from the AudioSet as w...	M2UGen
T5 Falcon LLaMA All 22 Tasks Logical Deductions, Social IQA, GSM8K, Tracking Shuffled Objects Table 5: An overview of the experimental setup. Mod- els in the GPT and T5 families are evaluated on all tasks and those in the Falcon and LLaMA families on a subset of representative tasks. In addition, each evaluation i...	AreEmergentAbilitiesinLarge Language Models just In-Context
sion text using this template: {"text": "ReadCompre"}, by replacing the ReadCompre with one piece of reading comprehension text. This template is used to explicitly separate each reading com- prehension text from others, so that the model can focus on the context of the current document.	ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION
• Consolidating various evaluations of open-source LLMs, providing an unbiased and comprehensive view of open-source LLMs vs. ChatGPT (Figure 1, Section 3.1). • Systematically reviewing open-source LLMs that match or surpass the performance of ChatGPT in various tasks with analysis (Figure 2, Section 3, Section 4.2)....	ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup
Apr 20, 2023 Apr 19, 2 https://blog.google/technology/ai/bard-google-ai-search-updates/ 4/4 T h e K e y w o r d A I B a r d n o w h e l p s y o u c o d e A M E S S A G E F R O M O U R C E O G o o g l e D e e p M i n d : B r i n g i n g t o g e t h e r t w o w o r l d - c l a s s A I t e a ...	Google AI updates_ Bard and new AI features in Search
Figures 8, 9 and 10, we conducted tests to evaluate the effectiveness of HuggingGPT in the case of complex tasks. Figure 8 demonstrates the ability of HuggingGPT to cope with complex tasks in a multi-round conversation scenario. The user splits a complex request into steps and reaches the ﬁnal goal through multiple rou...	HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face
ditions and effects) that will be ﬁxed for all problem instances that happen in that domain. While the problem of automatically generating the description is another valuable research question, in this proposed work, we assume that the description is available as a PDDL domain ﬁle. The LLM+P method is directly applicab...	LLM+P- Empowering Large Language Models with Optimal Planning Proficiency
Recall 0.82±0.01 0.82±0.01 0.25±0.04 0.82±0.01 0.57±0.02 0.83±0.03 F1-score 0.82±0.01 0.83±0.00 0.34±0.03 0.83±0.01 0.61±0.01 0.84 ±0.01 Precision 0.84±0.01 0.84±0.01 0.53±0.03 0.83±0.01 0.65±0.02 0.83±0.04 MultiEmbed Recall 0.84±0.00 0.83±0.01 0.26±0.02 0.82±0.01 0.58±0.02 0.82 ± 0.01 F1-score 0.84±0.00 0.83±0.01...	MULTI HASH EMBEDDINGS IN SPACY
C u r i o s i t y B a r d c a n b e a j u m p i n g o ff p o i n t f o r u s e r s ’ c u r i o s i t y a s t h e y e x p l o r e i d e a s o r t o p i c s o f i n t e r e s t . F o r i n s t a n c e , B a r d c a n e x p l a i n a c o m p l e x c o n c e p t s i m ...	An overview of Bard- an early experiment with generative AI
[284] Mehta, S., M. Rastegari. Mobilevit: Light-weight, general-purpose, and mobile-friendly vision transformer. In The Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022. OpenReview.net, 2022. [285] Tolstikhin, I. O., N. Houlsby, A. Kolesnikov, et al. Mlp-mixer: An...	TheRiseandPotentialofLargeLanguageModel BasedAgents
page,whichchangeitforyou.We’renotabletoreliablyundoarbitrarychangestothestyle.Pleaseremovetheoffendingpackage(s),orlayout-changingcommandsandtryagain.101010110.00.20.40.60.81.0Epoch1Epoch2TokensAccuracy70M160M410M1.0B1.4B2.8B6.9B12B	Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling
4.1 DATASET DISTILLATION Fixed initialization. With access to initial network weights, distilled images can directly train a ﬁxed network to reach high performance. For example, 10 distilled images can boost the performance of a neural network with an initial accuracy 12.90% to a ﬁnal accuracy 93.76% on MNIST (Figure 2...	DATASET DISTILLATION
ZENY: If preregistration prior to data collection is en- couraged, this would solve the problem, no?16 SOCART: Surely, but this would mean only one prereg- istered study per dataset. Since few NLP papers introduce new datasets, this would render prereg- istration ineffective for the vast majority of NLP research. ZE...	A Two-Sided Discussion of Preregistration of NLP Research
Since LLMs primarily rely on language as their input and output format, the text-based environment serves as the most natural platform for agents to operate in. It is shaped by natural language descriptions without direct involvement of other modalities. Agents exist in the text world and rely on textual resources to p...	TheRiseandPotentialofLargeLanguageModel BasedAgents
We adopt curriculum learning on frame resolution and frames-per-second (FPS). First, the diffuser is trained on the WebVid dataset of a 256-frame resolution, with the training objective being text- conditioned video generation. The training clips are sampled from 2-second video chunks with 4 FPS. Second, the model is f...	Any-to-Any Generation via Composable Diffusion
J. ACM, Vol. 37, No. 4, Article 111. Publication date: August 2018. A Survey on Evaluation of Large Language Models 111:25 FRESHQA. To effectively assess in-depth dialogue, Wang et al. [193] introduced the Dialogue CoT, incorporating two efficient dialogue strategies: Explicit CoT and CoT.	ASurveyonEvaluationofLargeLanguageModels
Prompt Tom and Jane are friends. One day, Jane goes to Tom’s house. Tom has a big pot of soup. He wants to share it with Jane. “Jane, do you want some soup?” Tom asks. “Yes, please. It looks yummy,” Jane says. Tom pours some soup into two bowls. He gives one bowl to Jane. Jane takes a spoonful of soup, but then she ma...	TinyStories-HowSmallCanLanguageModelsBeandStillSpeak CoherentEnglish?
i=1. In controlled sentiment generation, x is a prefix of a movie review from the IMDb dataset [22], and the policy In order to perform a controlled evaluation, for this must generate y with positive sentiment. experiment we generate preference pairs over generations using a pre-trained sentiment classifier, where p(po...	Direct Preference Optimization
KL Divergence (KLD). There is a many-to-many rela- tionship between text descriptions and music clips com- patible with them. It is therefore not possible to directly compare the generated music with the reference at the level of the audio waveform. To assess the adherence to the input text description, we adopt a prox...	MusicLM
with imagenet? arXiv preprint arXiv:2006.07159, 2020. Shariq Farooq Bhat, Ibraheem Alhashim, and Peter Wonka. AdaBins: Depth estimation using adaptive bins. In 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, jun 2021. doi: 10.1109/cvpr46437.2021.00400. URL https://doi.org/10.1109%2Fcv...	DINOv2- Learning Robust Visual Features without Supervision
You will answer the question based on the context ( only if available and helpful ) and your own knowledge of Minecraft . 1) Start your answer with " Answer : ". 2) Answer " Answer : Unknown " if you don ’t know the answer . A.4 Skill Library A.4.1 Components in the Prompt The input prompt to GPT-4 consists of the...	VOYAGER- An Open-Ended Embodied Agent with Large Language Models
Fortunately, we have line of sight on solving most of these problems. Rollup scalability and cost reduction progress is already legible to most of the crypto community. There are also many teams working on games-speci	The Open Problems of Onchain Games
Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross- lingual representation learning at scale. In Proceedings of the 58th Annual Meeting of the Association for Com- putational L...	The Pile- An 800GB Dataset of Diverse Text for Language Modeling
into a 100x engineer, but it’s not going to make a mediocre engineer much better. Like if you gave someone who’s not very good at math a calculator,” she says. “They wouldn’t know how to get the most out of it.”	4 Trends for AI Startups and Generative AI Companies
5.3 Concurrent Validity In this step, we wanted to investigate the extent to which the factors of the SHAPE scale could predict an individual’s inclination to obtain ATs to show concurrent validity. We measured this inclination in Survey one with the response options of "yes" or "no" to the question "I would like to ge...	Society’sAttitudesTowardsHumanAugmentation
[4] Timothy Neate, Aikaterini Bourazeri, Abi Roper, Simone Stumpf, and Stephanie Wilson. 2019. Co-Created Personas: Engaging and Empowering Users with Diverse Needs Within the Design Process. In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems (CHI '19). Supporting more Accessible R...	informatics-phd-projects-2022-23
8 000t100t1t20t1t2t3t1t2t3t4t2t3t4t5…………tn-4tn-3tn-2tn-1tn-3tn-2tn-1tns1k1_Rk2_Lk2_Rk1_Ls2s3s4s5s…sn-1snt3t4t5t6s6Stereo Delay PatternResidual codebooksSequence steps0000000000000000000t1…………tn-8tn-7tn-6tn-5tn-7tn-6tn-5tn-4k3_Rk4_Lk4_Rk3_L000t200t1t1t1t1t2t2t2t2t3t3t3t3t4t4t4t4t5t5…………tn-2tn-2tn-1tn-1tn-1tn-1tntns1k1_...	Simple and Controllable Music Generation
sketches. Concurrent works [8, 9, 32, 56] examine a wide variety of ways to control diffusion models.	AddingConditionalControltoText-to-ImageDiffusionModels
k in view k. xR solve this problem, we implement depth alignment to align the estimated depth map to the known depth values in the radiance ﬁeld. C. Depth Alignment	Text2NeRF- Text-Driven 3D Scene Generation with Neural Radiance Fields
2 Understanding Art with AI 2.1 Art Collections as Data Sources	UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK
A study by Ruchansky et al. [123] suggested a model called CSI, which comprises three modules, Capture, Score, and Integrate. The capture module extracts features from the article, and the score module extracts features from the user. Then by integrating article and user-based features, the CSI model performs the predi...	A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning
Chen, J., Lu, C., Chenli, B., Zhu, J., and Tian, T. Vﬂow: More expressive generative ﬂows with variational data augmentation. In International Conference on Machine Learning, pp. 1660–1669. PMLR, 2020. Chen, X., Kingma, D. P., Salimans, T., Duan, Y., Dhari- wal, P., Schulman, J., Sutskever, I., and Abbeel, P. Variatio...	ConditionalVariationalAutoencoderwithAdversarialLearningfor End-to-EndText-to-Speech
Two routes remain by which legal action might combat these campaigns while leaving CDA 230 untouched. One, which would build on the legal precedent set by the Ninth Circuit in the Roommates.com case, would leave courts to engage in line drawing around the degree to which platforms might elicit illegal disinformation ac...	Social_Media_and_Democracy
Modern LLMs are built on artiﬁcial neural networks: They work by computing and updating numeric activation values for internal components that are very loosely modeled on human neurons (Bengio et al., 2017). On this analogy, our tools for doing neuroscience on these systems are still weak: We have some coarse tools for...	Eight Things to Know about Large Language Models
Large Language Models. arXiv:2306.11507 [cs.CL] HuggingFaceH4/open_llm_leaderboard. [75] Israt Jahan, Md Tahmid Rahman Laskar, Chun Peng, and Jimmy Huang. 2023. Evaluation of ChatGPT on Biomedical Tasks: A Zero-Shot Comparison with Fine-Tuned Generative Transformers. arXiv preprint arXiv:2306.04504 (2023). [76] Neel ...	ASurveyonEvaluationofLargeLanguageModels
2019, Brown et al., 2020, Devlin et al., 2018]. The Word2Vec objective [Mikolov et al., 2013] predicts a masked out portion of the training text has served as a foundational objective for self-supervised learning in natural language. While uncommon, language modeling can be done with contrastive learning for word or or...	A Cookbook of Self-Supervised Learning
Table 2: Dataset Variations for the optimal model from Section 4.2 and optimal training routine from Sec- tion 4.3, modifying ﬁnal batch size in conjunction with dataset format.	CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY

Subsets and Splits

No community queries yet

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