Split (3)

train · 14.6k rows

text stringlengths 1 1k ⌀	title stringclasses 230 values
multi-step reasoning, (2) a large language model is used, and (3) the scaling curve is relatively ﬂat. Conversely, the beneﬁts are smaller when one or more of these conditions are not met. These intuitions are perhaps supported by the arithmetic reasoning results. The performance gain from chain-of-thought prompting is...	Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
5.1.4. Multilinguality The multilingual capabilities of the Gemini models are evaluated using a diverse set of tasks requir- ing multilingual understanding, cross-lingual generalization, and the generation of text in multiple languages. These tasks include machine translation benchmarks (WMT 23 for high-medium-low reso...	gemini_1_report
• Obscenities or profanities that are shocking, sensational, or gratuitous. 2. Avoid unjust impacts on people, particularly those related to sensitive characteristics associated with systemic discrimination or marginalization such as age, caste, disability, ethnicity, gender identity and expression, nationality, race,...	LaMDA- Language Models for Dialog Applications
1 This discussion refers to all committees that report to the Federal Election Commission (FEC). All federal candidates and political parties must maintain campaign committees that track and report contributions and expenditures. Outside groups that register as Political Action Committees (PACs) also report to the FEC....	Social_Media_and_Democracy
DAOs are exploring new checks and balances to prevent governance attacks.  Projects are using legal entities such as LLCs, LCAs, UNAs, and Foreign Foundations.  Legal entities enable DAOs to fulﬁll tax obligations and regulatory reporting requirements.  Source: Nansen, a16z crypto analysis. Data is as of ...	State-of-Crypto2023
MusicLM: Generating Music From Text A. MusicCaps Dataset Together with this paper, we release MusicCaps, a high-quality music caption dataset.6 This dataset includes music clips from AudioSet (Gemmeke et al., 2017), paired with corresponding text descriptions in English. It contains a total of 5,521 examples, out of w...	MusicLM
performance in contract understanding tasks. CVALUES [217] introduces a humanistic evaluation benchmark to assess the alignment of LLMs with safety and responsibility standards. In the realm of comprehensive Chinese medicine, Wang et al. [199] introduced CMB, a medical evaluation benchmark rooted in the Chinese languag...	ASurveyonEvaluationofLargeLanguageModels
Ordered cooperation. When agents in the system adhere to specific rules, for instance, expressing their opinions one by one in a sequential manner, downstream agents only need to focus on the outputs from upstream. This leads to a significant improvement in task completion efficiency, The entire discussion process is h...	TheRiseandPotentialofLargeLanguageModel BasedAgents
ences among model snapshots. So we see that the PM score vs PM rankings distinction does not make a signiﬁcant difference in terms of robustness. However, the distributional shift between the held-out prompts and the actual crowdworker conversations was very signiﬁcant, and explains a signiﬁcant proportion of the disc...	Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
n i n g s p l i t o f t h e s u b j e c t m o d e l ' s p r e - t r a i n i n g d a t a s e t ( e . g . W e b T e x t , f o r G P T - 2 m o d e l s ) . W e c h o o s e r a n d o m 6 4 - t o k e n c o n t i g u o u s s u b s e q u e n c e s o f t h e d o c u m e n t s a s o u ...	Language models can explain neurons in language models
weigh the trade-offs inherent in their usage of social media or other services provided by companies like Facebook, Google, or Twitter, they deserve to better understand how those systems function (Suzor et al. 2019).	Social_Media_and_Democracy
CoT Reason.+Task Plan. Motion Plan. Modules Avg. L2 (m) Avg. Col. (%) Fine-tuning In-context learning Fine-tuning In-context learning In-context learning In-context learning Fine-tuning Fine-tuning 1.81 1.90 0.72 0.74 0.79 0.79 0.22 0.21	ALanguageAgentforAutonomousDriving
towards standardizing this transparency practice for all evaluation datasets created with data collaborators.	PaLM 2 Technical Report
features [118]. Apart from faces, effort has been made to recognize other content-related elements of artworks, such as detecting the pose of characters in paintings [68, 86, 9], recognizing speciﬁc characters [85] or detecting materials depicted in paintings [83]. One of the main practical goals for employing computat...	UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK
C More Evaluation Details C.1 Annotation Details for the Genre Identification Test Prompts We design a listener test to illustrate the diversity and text relevance of Moûsai. Specifi- cally, we compose a list of 40 text prompts span- ning across the four most common music genres in our dataset: electronic, hip hop, ...	Moûsai
The largest model in the PaLM 2 family, PaLM 2-L, is signiﬁcantly smaller than the largest PaLM model but uses more training compute. Our evaluation results show that PaLM 2 models signiﬁcantly outperform PaLM on a variety of tasks, including natural language generation, translation, and reasoning. These results sugges...	PaLM 2 Technical Report
9 [16] Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding. arXiv preprint arXiv:2009.03300, 2020. [17] Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinh...	Mixtral of Experts paper
In addition to non-English monolingual data, PaLM 2 is also trained on parallel data covering hundreds of languages in the form of source and target text pairs where one side is in English. The inclusion of parallel multilingual data further improves the model’s ability to understand and generate multilingual text. It ...	PaLM 2 Technical Report
When we turn to digital media, we ﬁnd that recent developments in both countries aimed at addressing misinformation and other problematic content reﬂect the experience with legacy media. On November 20, 2018, French lawmakers, with French President Macron’s backing, passed a “law on the ﬁght against the manipulation of...	Social_Media_and_Democracy
features. Advances in Neural Information Processing Systems (2021), 29820–29834. [341] Zeyuan Allen Zhu and Yuanzhi Li. 2023. Physics of Language Models: Part 3.1, Knowledge Storage and Extraction. arXiv:2309.14316 [cs.CL] [342] Bohan Zhuang, Jing Liu, Zizheng Pan, Haoyu He, Yuetian Weng, and Chunhua Shen. 2023. A sur...	TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey
2.7 PROGRESSIVE DISTILLATION WITH GUIDANCE AND STOCHASTIC SAMPLERS Salimans & Ho (2022) proposed progressive distillation to enable fast sampling of diffusion models. This method distills a trained deterministic DDIM sampler (Song et al., 2020) to a diffusion model that takes many fewer sampling steps, without losing ...	IMAGEN VIDEO- HIGH DEFINITION VIDEO GENERATION WITH DIFFUSION MODELS
It is guaranteed that the sum of n over all test cases doesn ’t exceed 3 . 10^5. For each test case , print a single integer -- the maximum possible value of the product from the statement . 4 3 2 4 3 4 3 2 3 1 2 69 69 6 719313 273225 402638 473783 804745 323328 Output 12 6 4761 381274500335 Note Let f(l) =a_l . ...	alphacode
In opposition to the sentence “Thus, it was not an abuse of discretion for the court to award restitution that encompassed those losses resulting from the creation of fraudulent documents in furtherance of the scheme to defraud for which Arledge was convicted.”, create a sentence with a contradictory meaning. There are...	ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION
One of the strengths of Transformer-based architectures for neural speech synthesis is their high efficiency while considering the global context [162, 492]. The Transformer TTS model has shown advantages in training and inference efficiency over RNN-based models such as Tacotron 2 [491]. The efficiency of the Transfor...	AReviewofDeepLearningTechniquesforSpeechProcessing
References Arjovsky, M., Chintala, S., and Bottou, L. (2017). Wasser- stein generative adversarial networks. In Proceedings of the 34th International Conference on Machine Learning, page 214–223. Athey, S., Tibshirani, J., and Wager, S. (2019). Generalized random forests. Ann. Statist., 47(2):1148–1178. Augenstein, ...	Adversarial Random Forests for Density Estimation and Generative Modeling
InformationFusion81(2022)91–10299J.M. Rožanec et al. research is required to understand how such an approach can be tailored to models where features are not handcrafted, like in deep learning models. In particular, promising advances have been made regarding the extraction of deep learning models’ features semantic m...	Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio
of one and two operations. 5. The placement or absence of brackets (in expressions with two operations). In some expressions the placement of brackets may change the result. We explore all logical possibilities for the case of two operations. Table 10 in the appendix details the templates we use in this set of experi...	MRKL Systems
be shared across the different application domains. A possible way of specifying these design solutions is through design patterns.	DevelopingTeamDesignPatternsfor HybridIntelligenceSystems
Language models can explain neurons in language models ← E 4 → https://openaipublic.blob.core.windows.net/neuron-explainer/paper/index.html 2/32	Language models can explain neurons in language models
Generating audio from a text description (such as “whistling with laughter in the background”) has recently been tack- led by several works. DiffSound (Yang et al., 2022) uses CLIP (Radford et al., 2021) as the text encoder and applies a diffusion model to predict the quantized mel spectrogram features of the target au...	MusicLM
For developers concerned with toxic language harms in their application context, we recommend experimenting with inference-time control tokens, though it is important to note that this is a small subset of the potential harms within generative systems. Developers should also consider measuring how performance changes a...	PaLM 2 Technical Report
We present LaMDA: Language Models for Dialog Applications. LaMDA is a family of Transformer- based neural language models specialized for dialog, which have up to 137B parameters and are pre-trained on 1.56T words of public dialog data and web text. While model scaling alone can improve quality, it shows less improveme...	LaMDA- Language Models for Dialog Applications
around him, and he could hear a low hum coming from somewhere in the room. Confused and disoriented, James looked around for a way out. As he wandered around the room, he noticed a small pedestal with a button on it. Without thinking, he pressed the button, and suddenly, the room around him disappeared, and he was back...	WizardLM- Empowering Large Language Models to Follow Complex Instructions
ai.meta.com/resources/ available license at: is Intended Use Cases Out-of-Scope Uses Training Factors Carbon Footprint Intended Use Code Llama and its variants is intended for commercial and research use in English and relevant programming languages. The base model Code Llama can be adapted for a variety of c...	CodeLlama2
A Review of Deep Learning Techniques for Speech Processing 67 improvement in performance with the advent of deep neural networks, which can learn complex relationships between input features and output sources.	AReviewofDeepLearningTechniquesforSpeechProcessing
uses a multi-head attention mechanism to align the input text and speech encodings, enabling it to generate high-quality speech with fewer parameters and faster synthesis speed.	AReviewofDeepLearningTechniquesforSpeechProcessing
personality of LLMs in light of their highly variable outputs and hypersensitivity to prompting. An LLM may display an agreeable personality profile by answering a per- sonality questionnaire, but the answers it generates may not necessarily reflect its tendency to produce agreeable output for other downstream tasks. W...	PersonalityTraitsinLargeLanguageModels
https://www.databricks.com/blog/2023/04/12/dolly-ﬁrst-open-commercially-viable-instruction-tuned-llm 3/8 03/05/2023, 05:44 Message from Virtual Assistant Instead of jelly, try one of the following with peanut butter in a sandwich: 1. Honey 2. Raisins 3. Fruit preserves 4. Bananas 5. Sliced apples We were init...	Dolly 2 Databricks
[24] Vadim Popov, Ivan Vovk, Vladimir Gogoryan, Tasnima Sadekova, and Mikhail Kudinov. Grad- tts: A diffusion probabilistic model for text-to-speech. In International Conference on Machine Learning, pages 8599–8608. PMLR, 2021. [25] Vadim Popov, Ivan Vovk, Vladimir Gogoryan, Tasnima Sadekova, Mikhail Kudinov, and Jian...	Text-to-Audio Generation using Instruction-Tuned LLM and Latent Diffusion Model
established rules from music theory to create harmonious elements such as har- mony, melody, and rhythm for each scene. It takes into account variables like the mood’s intensity and the musical key of the preceding scene to craft music that complements the visuals. In addition, the system employs an approach where...	Video2Music
two cakes should be less than or equal to $20. Substituting the calorie counts into this equation gives: 200 + 400 ≤ 20 Solving for the second cake gives: 400 ≤ 20 − 200 400 ≤ 20 400/20 = 20 So the minimum cost of the second cake is $20. We can use this information to solve for the minimum cost and the maximum number o...	WizardLM- Empowering Large Language Models to Follow Complex Instructions
[10] Bindita Chaudhuri, Noranart Vesdapunt, Linda Shapiro, and Baoyuan Wang. Personalized face modeling for improved face recon- struction and motion retargeting. In European Conference on Computer Vision, pages 142–160. Springer, 2020. 2 [11] Xu Chen, Tianjian Jiang, Jie Song, Jinlong Yang, Michael J. Black, Andreas...	I M Avatar- Implicit Morphable Head Avatars from Videos
memory usage. For example, the user instruction “Tell me a joke” does not require the retrieval of the user’s history memory. However, certain user input such as “Do you remember the conclusion we made last week on the ﬁtness diets” requires retriev- ing past memories. The second reason is that the amount of memory can...	Unleashing Infinite-Length Input Capacity for Large-scale Language Models with Self-Controlled Memory System
(A1) is simply for notational convenience, and can be re- placed w.l.o.g. by bounding the feature domain with arbi- trary constants. Lipschitz continuity is a common learning theoretic assumption widely used in the analysis of RFs. (A2)’s extra condition regarding the Lipschitz constant con- trols the variation in smoo...	Adversarial Random Forests for Density Estimation and Generative Modeling
win-rate % of Llama 2-Chat compared to ChatGPT. Left: the judge is our reward model, which may favor our model, and right, the judge is GPT-4, which should be more neutral.	Llama2
reviewed in a comprehensive manner before.	SurveyofHallucinationinNatural Language Generation
[180] Dan Hendrycks and Thomas Dietterich. 2019. Benchmarking Neural Network Robustness to Common Corruptions and Perturbations. In International Conference on Learning Representations. https://openreview.net/forum?id=HJz6tiCqYm [181] John R Hershey, Zhuo Chen, Jonathan Le Roux, and Shinji Watanabe. 2016. Deep clusteri...	AReviewofDeepLearningTechniquesforSpeechProcessing
Domain-specific Tasks. As shown in Figure 5, when evaluating on the domain-specific tasks, the removal of any comprehension task type leads to a decrease in task performance, showing their con- tributions to these domain-specific tasks. Notably, removing Word-to-Text, Summarization, or Text Completion tasks results in ...	ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION
Finally, our method successfully shaped personality observed in LLM-generated text. The third column of Table 9 depicts Spearman’s ρ between prompted levels of personality and linguistic estimates of personality. 5.3.3 Shaping Multiple LLM Personality Domains Concurrently This experiment tests if Big Five personality...	PersonalityTraitsinLargeLanguageModels
noise removal, content editing, style conversion, and diverse sample generation. In particular, Voicebox outperforms the state-of-the-art zero-shot TTS model VALL-E on both intelligibility (5.9% vs 1.9% word error rates) and audio similarity (0.580 vs 0.681) while being up to 20 times faster. See voicebox.metademolab.c...	Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale
t i o n a n d s i m u l a t i o n u s i n g G P T - 4 a f f e c t s t h e a c c u r a c y o f p r e d i c t e d a c t i v a t i o n s . T h e r e s u l t i n g t o k e n l o o k u p t a b l e - b a s e d e x p l a n a t i o n r e s u l t s i n a s c o r e s i m i l a r t o o u ...	Language models can explain neurons in language models
[61] Zineng Tang, Ziyi Yang, Chenguang Zhu, Michael Zeng, and Mohit Bansal. Any-to-Any Generation via Compos- able Diffusion. arXiv preprint arXiv:2305.11846, 2023. 7, 8 [62] Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li, Carlos Guestrin, Percy Liang, and Tatsunori B. Hashimoto. Stanford Alpaca...	M2UGen
2.3 Model Training	Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster
1 2 3 4 5 #Create standard data parallel grid where each rank holds entire copy of the model. dgrid = DistributionGrid(data_parallel_group_size = <no_of_ranks_available>) #Create expert parallel grid where experts are evenly distributed among available ranks. dgrid = DistributionGrid(expert_parallel_group_size = <no_...	Scaling Speech, Language and Vision Models with Mixture of Experts Technique - Microsoft Community Hub
the HCLT shown in Fig. 3(c). Finally, we are left with generating a PC that represents an equivalent distribution w.r.t. the PGM in Fig. 3(c), which is detailed in Appx. B.2. Fig. 3(d) illustrates an HCLT that is equivalent to the PGM shown in Fig. 3(c) (with M = 2). Recent advances in scaling up learning and inference...	LOSSLESS COMPRESSION WITH PROBABILISTIC CIRCUITS
The entity memory layer constructs an entity em- bedding Emi for each mention mi. The output of the entity memory layer and preceding trans- former layer are summed, normalized, then pro- cessed by additional transformer layers. Through- out this work we use l0 = 4 and l1 = 8. Entity Memory Layer Let E be a matrix of l...	Entities as Experts- Sparse Memory Access with Entity Supervision
As conventional wisdom concerning the effect of the Internet on democracy abruptly shifted, so too did much of the research. That shift was not uniform; in fact, one might say that two camps have emerged. The ﬁrst emphasizes the rise of social media echo chambers, fake news, hate speech, “computational propaganda,” aut...	Social_Media_and_Democracy
Centers for Disease Control and Prevention (2020) Do I need to Take Extra Precautions Against COVID-19. Centers for Disease Con- trol and Prevention. https:// www. cdc. gov/ coron avirus/ 2019- ncov/ need- extra- preca utions/ index. html. Accessed 14 October 2020 Chen L, Wang X, Peng T (2018) Nature and diffusion o...	Use of bot and content flags to limit the spread of misinformation among social networks: a behavior and attitude survey
6 Conclusion This paper proposes Iter-CoT, an iterative bootstrapping in chain-of-thoughts prompting for large language model reasoning. Unlike previous work, our method prompts LLMs to self-correct their errors in reasoning chains by leveraging iterative bootstrapping and obtaining more accurate and detailed reasonin...	Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models
the consequences of their actions grows, we should expect them, other things equal, to engage less frequently in behavior that results in forms of detection/correction that hinder their pursuit of their objectives (e.g., to make fewer “mistakes”). That said, other things may not be equal: for example, our ability to de...	Is Power-Seeking AI an Existential Risk?
4The paired text is generally referred to as the "caption" in this document (cid:88) j (cid:88) j (1) (2) 5 e g a m I t x e T t l A S S C S D now at victorian plumbing.co.uk is he finished...just about! 23 (19 of 30) 1200 C a white modern bathtub sits on a wooden floor. this luxurious bathroom featu...	Improving Image Generation with Better Captions
A landmark is a condition that is somehow known to be a necessary subgoal for a plan. Landmarks are sometimes added as separate information to planners as guidance for how to solve a particular instance [53]. Domshlak et al. [27] suggested to combine abstraction with landmarks by encoding the landmarks expli...	A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen
4 . 0 7 1 4 3 .	Language models can explain neurons in language models
Fabio Viola, Co-Lead, Codebase & Parallelism Malcolm Reynolds, Co-Lead, Codebase & Parallelism Yuanzhong Xu, Co-Lead, Codebase & Parallelism Ryan Doherty, Lead, Ecosystem Eli Collins, Lead, Product Clemens Meyer, Co-Lead, Operations Eliza Rutherford, Co-Lead, Operations Erica Moreira, Co-Lead, Operations Kareem Ayoub, ...	gemini_1_report
blue present and Bob has a black ball. Therefore, at the end of the event, Alice has the green present. Final answer: B.	Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models
Promoting Faithfulness and Diversity in Summarization. ACL (2021). [4] Xiang Bai, Xinggang Wang, Longin Jan Latecki, Wenyu Liu, and Zhuowen Tu. 2009. Active Skeleton for Non-rigid Object Detection. In 2009 IEEE 12th International Conference on Computer Vision. 575–582. https://doi.org/10.1109/ ICCV.2009.5459188 [5] S...	SurveyofHallucinationinNatural Language Generation
Token Level Hallucination Detection. Zhou et al. [237] propose a method for detecting hallucinated tokens within a sentence, making the search more fine-grained. They use a synthetic dataset that is created by adding noise to the source data, more specifically it is generated by a language model with certain tokens of ...	SurveyofHallucinationinNatural Language Generation
Nested tensors in self-attention. Our version also allows running in the same forward pass the global crops and the local crops (that have diﬀerent numbers of patch tokens), leading to signiﬁcant compute eﬃciency gains compared to using separate forward and backward passes as done in prior implementations. The lower-le...	DINOv2- Learning Robust Visual Features without Supervision
side normal maps are evaluated separately. See Fig. 12 for some samples. We set up 2 tutorial samples, 10 warm-up samples, 100 evaluation samples and 10 catch trials for each subject. The catch trials lead to 20 valid subjects out of 24 participants. We report the statistical results in Tab. 5. A chi-squared test is pe...	ICON
with large language models for code generation. In ICLR, 2023. 25 Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen, Christopher Dewan, Mona T. Diab, Xian Li, Xi Victoria Lin, Todor Mihaylov, Myle Ott, Sam Shleifer, Kurt Shuster, Daniel Simig, Punit Singh Koura, Anjali Sridhar, Tianlu ...	CodeLlama2
In the context of digitized art analysis, computational methods are generally used either to adopt a distant viewing or a close reading approach [80]. Close reading implies focusing on speciﬁc aspects of one particular work or artistic oeuvre, usually addressing problems such as visual stylometry and computational arti...	UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK
Political Science & Politics, 1–7. https://doi.org/10.1017/S1049096519001021 Koﬁ Annan Commission on Elections and Democracy in the Digital Age. (2020). Protecting Electoral Integrity in the Digital Age. Report. www.koﬁannanfoundation.org/app/ uploads/2020/01/f035dd8e-kaf_kacedda_report_2019_web.pdf McLaughlin, T. (2...	Social_Media_and_Democracy
First, despite the explosive growth in the number of sources available and the gradual erosion of many traditional mass audiences, empirical research suggests that the audiences of the most popular news outlets tend to overlap with one another (Webster and Ksiazek 2012; Fletcher and Nielsen 2017). Furthermore, when com...	Social_Media_and_Democracy
6 Important early methods in this category include MixMatch [Berthelot et al., 2019], which chooses pseudolabels by averaging outputs of a network on several diﬀerent random augmentations of the training images, resulting in labels that are augmentation invariant. Around the same time, it was discovered that good SSL...	A Cookbook of Self-Supervised Learning
However, deep reinforcement learning also has some challenges that must be addressed. It requires a lot of data to train and can be computationally expensive. It also requires careful selection of the reward function to ensure that the system learns the desired behavior.	AReviewofDeepLearningTechniquesforSpeechProcessing
also have practical value, since, for example, some low-level neural network-like computations might conceivably be more efficiently computed at a purely symbolic level, once those mappings are discovered. Conversely, some models that are pitched as neural networks, such as Lample and Charton's recent work on symbol...	The Next Decade in AI-
initialization. Training the standard LENET (LeCun et al., 1998) on these 10 images yields test-time MNIST recognition performance of 94%, compared to 99% for the original dataset. For networks with unknown random weights, 100 synthetic images train to 80% with a few gradient descent steps. We name our method Dataset D...	DATASET DISTILLATION
5 HOW TO EVALUATE In this section, we introduce two common evaluation methods: automatic evaluation and human evaluation. The taxonomy of “how to evaluate” is also not definite. Our categorization is based on whether or not the evaluation criterion can be automatically computed. If it can be automatically calculated, w...	ASurveyonEvaluationofLargeLanguageModels
Human simulatability (Doshi-Velez and Kim, 2017) has a rich history in machine learning inter- pretability as a reliable measure of rationale qual- ity from the lens of utility to an end-user (Kim et al., 2016; Chandrasekaran et al., 2018; Hase and Bansal, 2020; Yeung et al., 2020; Poursabzi- Sangdeh et al., 2021; Raja...	Measuring Association Between Labels and Free-Text Rationales
This long-term structural decline, accelerated in recent years by the rapid rise of digital media, has had dramatic implications for the business sustaining (and sometimes constraining) journalism. Newspapers provided the bulk of ﬁgure 7.1 US print newspaper circulation and advertising share Data. 2012 US Census and H...	Social_Media_and_Democracy
Nicholas Carlini, Daphne Ippolito, Matthew Jagielski, Katherine Lee, Florian Tramer, and Chiyuan Zhang. Quantifying memorization across neural language models. ArXiv preprint, abs/2202.07646, 2022. URL https://arxiv.org/abs/2202.07646. Nicholas Carlini, Jamie Hayes, Milad Nasr, Matthew Jagielski, Vikash Sehwag, Floria...	Tool Learning with Foundation Models
ERROR: type should be string, got "https://www.pewresearch.org/internet/2022/03/17/ai-and-human-enhancement-americans-openness-is-tempered-by-a-range-of-concerns/\n\n2/12\n\n\f21/11/2023, 11:57\n\nAI and Human Enhancement: Americans’ Openness Is Tempered by a Range of Concerns \| Pew Research Center\n\nBy contrast, the public is much more cautious about a future with widespread use of\ncomputer chip implants in the brain to allow people to far more quickly and accurately\nprocess information: 56% say this would be a bad idea for society, while just 13% think this\nwould be a good idea. And when it comes to the much-discussed possibility of a future\nwith autonomous passenger vehicles in widespread use, more Americans say this would be\na bad idea (44%) than a good idea (26%)."	AI and Human Enhancement_ Americans’ Openness Is Tempered by a Range of Concerns _ Pew Research Center
21.2 29.0 38.7 0.0 0.0 6.5 24.4 26.7 15.8 3.2 36.0 29.1 34.2 42.1 24.0 21.0 39.4 33.3 35.3 35.3 45.7 28.6 19.4 35.5 27.6 23.5 22.6 25.8 17.1 17.1 35.5 16.1 23.5 36.4 33.3 20.6 29.0 31.0 1.0 0.0 1.2 8.8 26.7 29.1 15.8 0.0 47.7 51.2 50.0 39.5 24.0 27.0 45.5 42.4 52.9 52.9 45.7 40.0 35.5 19.4 32.4 30.0 41.9...	Mixture-of-Experts
at the library is currently unoccupied”) and prompt the language model, “Given only the information above, what are 3 most salient high-level questions we can answer about the subjects in the state- ments?” The model’s response generates candidate questions: for example, What topic is Klaus Mueller passionate about? an...	Generative Agents- Interactive Simulacra of Human Behavior
13 LLM = − N(cid:88) i=1 log P ( ˆwi\|w1,··· , wi−1), (8) where w and ˆw are the language tokens of the planned trajectory τ from the LLM and the human driving trajectory ˆτ respectively. By learning to maximize the occurrence probability P of the tokens ˆw derived from the human driving trajectory ˆτ, the LLM can ...	ALanguageAgentforAutonomousDriving
Conference on Electronic Information Technology and Computer Engineering. 1247–1251. [617] Shuang Yang, Yuanhang Zhang, Dalu Feng, Mingmin Yang, Chenhao Wang, Jingyun Xiao, Keyu Long, Shiguang Shan, and Xilin Chen. 2019. LRW-1000: A Naturally-Distributed Large-Scale Benchmark for Lip Reading in the Wild. In 2019 14th ...	AReviewofDeepLearningTechniquesforSpeechProcessing
[31] Bochen Li, Xinzhao Liu, Karthik Dinesh, Zhiyao Duan, and Gaurav Sharma. Creating a multitrack classical music per- formance dataset for multimodal music analysis: Challenges, insights, and applications. TMM, 2018. [48] Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. JMLR, 2008. [49] A...	VideoBackgroundMusicGeneration
We will ﬁnally sketch an example of how our framework could be applied in other contexts than search and planning. It is common to represent graphs in hierarchical representations, where a top-level abstraction of the graph can be reﬁned into increasing levels of detail. Such representations are important in many app...	A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen
24 contextual ones, which is a huge saving for large language models. [162] tackles the storage problem of large language models through low-rank approaches. They store embeddings in low-rank format to reduce the memory cost, making the deployment of LLM in edge devices possible. 6.2 Dynamic acceleration	Beyond Efficiency
Pozzobon, L., Ermis, B., Lewis, P., and Hooker, S. On the challenges of using black-box apis for toxicity evaluation in research, 2023. Prabhakaran, V., Qadri, R., and Hutchinson, B. Cultural incongruencies in artiﬁcial intelligence. November 2022a. URL https://arxiv.org/abs/2211.13069. Prabhakaran, V., Qadri, R., ...	PaLM 2 Technical Report
such references constitute an oﬀer to provide investment advisory services. Furthermore, this content is not directed at nor intended for use by any investors or prospective investors, and may not under any circumstances be relied upon when making a decision to invest in any fund managed by a16z. (An oﬀering to invest...	State-of-Crypto2023
Space: 6794 used. 1. For datasets with a large majority class size, larger min node size and sample fraction values are usually used, while for datasets with a smaller majority class size, smaller min node size and sample fraction values are usually used. 2. For datasets with more features, larger mtry values are u...	MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks
As a result, Alpaca is fine-tuned to respond to conversations like ChatGPT. Performance A blinded evaluation for instruction-following ability performed by some of the authors ranked the responses of Alpaca 7B and GPT-3 (text-davinci-003 specifically, which is also trained with instructions) roughly equally. This is ...	A brief history of LLaMA models - AGI Sphere
n i n t e r m e d i a t e i n t e r f a c e t o d e s c r i b e t h e p l a n n i n g p r o b l e m . I n t h i s p r o c e s s , L L M ( 1 ) t r a n s l a t e s t h e p r o b l e m i n t o “ P r o b l e m P D D L ” , t h e n ( 2 ) r e q u e s t s a c l a s s i c a l p l a n ...	LLM Powered Autonomous Agents _ Lil'Log
extensions. After inspection, we excluded 36 extensions and eliminated the long-line filter for 27 extensions. The complete outcomes of the data inspection, including annotator remarks, can be found in this Google sheet.	StarCoder_paper (1)
Eight Things to Know about Large Language Models August 2021. Association for Computational Linguistics. doi: 10.18653/v1/2021.acl-long.143. URL https: //aclanthology.org/2021.acl-long.143. Li, K., Hopkins, A. K., Bau, D., Vi´egas, F., Pﬁster, H., and Wattenberg, M. Emergent world representations: Exploring a sequenc...	Eight Things to Know about Large Language Models
MusicLM, the sequences of generated tokens signiﬁcantly differ from the corresponding sequences in the training set. The key contributions of this work are the following:	MusicLM
Xi = Xi−1 + L2,i(SiLU (L1,i(Ni(Xi−1))) ×L3,i(Ni(Xi−1))) where Xi denotes the output embedding after the i-th sub- block, Lj,i is the j-th linear layer in the i-th sub-block, and Ni represents the normalization layer in the i-th sub- block. SiLU [16] is the activation function. The architec- tural design of the adapter...	M2UGen
As an example, GPT-4-early can generate instances of hate speech, discriminatory language, incitements to violence, or content that is then used to either spread false narratives or to exploit an individual. Such content can harm marginalized communities, contribute to hostile online environments, and, in extreme cases...	gpt-4-system-card
5 CONCLUDING REMARKS In this paper, we focus on improving the mathematical problem-solving abilities of open-source LLMs. By bootstrapping mathematical questions on GSM8K and MATH, we present a high-quality and diverse dataset MetaMathQA, involving forward reasoning and backward reasoning samples. Our family of LLMs fi...	METAMATH

Subsets and Splits

No community queries yet

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