text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
i)
(cid:17) − Xt
i
(cid:13)(cid:13)(cid:13)2
2
L3D-cyc =
,
(19)
where τi is the opacity that weighs the sampled points so
that a point near the surface receives heavier regularization.
Our optimization is highly non-linear with local minima.
To improve the robustness of optimization, we consider the
following i... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
45, 35–44 (1998)
[67] Bock, R.D.: Psychometrics: ¡i¿the dependability of behavioral measurements¡/i¿.
theory of generalizability for scores and profiles. lee j. cronbach, goldine c.
gleser, harinder nanda, and ¡span class=”smallcaps smallercapital”¿nageswari
42
illus. $12.95. Sci-
rajaratnam.¡/span¿ wiley, new york... | PersonalityTraitsinLargeLanguageModels |
∗Corresponding author: elias.frantar@ist.ac.at
1
Published as a conference paper at ICLR 2023
2022). To date, only basic variants of round-to-nearest quantization (Yao et al., 2022; Dettmers
et al., 2022) have been applied at the scale of GPT-175B; while this works well for low compression
targets, e.g., 8-bit weig... | GPTQ |
8
There are three main techniques that change or control LLM’s behavior and output
to given input. These techniques can directly affect the model’s weight parameters as
in pretraining (i.e. training the LLM on a large dataset of general knowledge [3, 4, 79]),
fine-tuning (i.e. further training a pretrained LLM on a s... | PersonalityTraitsinLargeLanguageModels |
system with more agents could amplify this risk, making communication and information exchange
less reliable [405]. Furthermore, the difficulty of coordinating agents also magnifies with the increase
in their numbers, potentially making cooperation among agents more challenging and less efficient,
which can impact the ... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
E(cid:104)(cid:0)f (t)
n (x) − η(t)(x)(cid:1)2(cid:105)
= 0.
lim
n→∞ | Adversarial Random Forests for Density Estimation and Generative Modeling |
contribute to the development of more robust and effective generalist biomedical AI models. | BiomedGPT |
• We propose a cross-domain attention mechanism to pro-
duce multi-view normal maps and color images that are
consistently aligned. This mechanism facilitates infor-
mation perception across different domains, enabling our
method to recover high-fidelity geometry.
• We introduce a novel geometry-aware normal fusion al... | Wonder3D |
[64] L. Zeng, S. H. K. Parthasarathi, and D. Hakkani-Tur. N-best hypotheses reranking for text-to-sql
systems. arXiv preprint arXiv:2210.10668, 2022.
[65] T. Zhang, T. Yu, T. B. Hashimoto, M. Lewis, W.-t. Yih, D. Fried, and S. I. Wang. Coder reviewer
reranking for code generation. arXiv preprint arXiv:2211.16490, 20... | Teaching Large Language Models to Self-Debug |
Because of the above interpretability issues, many have turned to behavioural evaluations
which simply involve observing the model’s response to certain inputs. However, such
behavioural evaluations cannot exhaustively explore all possible vulnerabilities, and reliably
extrapolating from those that have been explore... | Capabilities and risks from frontier AI |
There are also a smaller number of standalone Generative AI web apps, such as Jasper and Copy.ai for copywriting, Runway for video editing, and Mem for note taking.
A plugin may be an effective wedge into bootstrapping your own application, and it may be a savvy way to surmount the chicken-and-egg problem of user dat... | Generative AI A Creative New World Sequoia Capital |
(Call these “APS”—Advanced, Planning, Strategically aware—systems.)
2. There will be strong incentives to build and deploy APS systems | (1).
3. It will be much harder to build APS systems that would not seek to gain and maintain power
in unintended ways (because of problems with their objectives) on any of the inputs... | Is Power-Seeking AI an Existential Risk? |
3.5.3 Medical
Benchmarks One desirable capability of LLMs is on contributing medical related tasks to make
affordable, high-quality healthcare more accessible to the broader public.
For mental health, IMHI (Yang et al., 2023c) benchmark is constructed using 10 existing mental
health analysis datasets, including mental... | ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup |
Figure 2: t-SNE representation from the last layer of
mBERT for the top-1000 predictions for the parallel
sentences in the list above (“We want to [MASK] in-
novation .” in English). Highest scored prediction is
starred; annotator’s answers are denoted by a dot with
black edge. Legend shows language-color mapping.
Fig... | Are Pretrained Multilingual Models Equally Fair Across Languages? |
text in order without skipping any words. To find the
optimum alignment, Kim et al. (2020) use dynamic pro-
gramming. Applying MAS directly in our setting is dif-
ficult because our objective is the ELBO, not the exact
log-likelihood. We, therefore, redefine MAS to find an
alignment that maximizes the ELBO, which reduces t... | ConditionalVariationalAutoencoderwithAdversarialLearningfor End-to-EndText-to-Speech |
up − W (t)
, W N F 4) + X BF 16W BF 16
downW t−1
down W (t+1)
downW (t)
up ).
down W BF 16
down .
, cF P 8
1
2
LoRA-FA
∆W = WdownWup = QRWup Wdown is frozen, and only update Wup.
based Improvements [52], [53], [54], in which several novel
technique are incorporated into LoRA for improvements, and
LoRA-based ... | Parameter-EfficientFine-TuningMethods |
5gpt-3.5-turbo from https://oai.azure.com/portal
3
Unlike Alpaca’s self-instruct [12] generation method, Evol-Instruct can control the difficulty and
complexity level of the generated instructions.
3 Approach
Figure 2: Overview of Evol-Instruct
In this section, we elaborate on the details of the proposed Evol-Inst... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
among different chain of thought annotations, as would be
expected when using exemplar-based prompting (Le Scao
and Rush, 2021; Reynolds and McDonell, 2021; Zhao
et al., 2021), all sets of chain of thought prompts outper-
form the standard baseline by a large margin. This result
implies that successful use of chain of ... | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
2 BACKGROUND AND RELATED WORK | LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET |
In recent years, the field of natural language processing (NLP) has been revolutionized by the
emergence of large language models (LLMs)[1, 2, 3, 4, 5, 6], exemplified by models such as GPT-
3[1], PaLM [3], and LLaMa [6]. LLMs have demonstrated impressive capabilities in zero-shot
and few-shot tasks, as well as more comp... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
Keywords Twitter · Misinformation · COVID-19 · Fact-checking · Survey study
1 Introduction
The COVID-19 crisis, which led to much of social life
migrating online, has contributed to an infodemic, where
information of varying quality quickly spreads in social
media networks around the world. While ideally high-qu... | Use of bot and content flags to limit the spread of misinformation among social networks: a behavior and attitude survey |
Yi Tay, Mostafa Dehghani, Samira Abnar, Hyung Won Chung, William Fedus, Jinfeng Rao, Sharan
Narang, Vinh Q. Tran, Dani Yogatama, and Donald Metzler. Scaling Laws vs Model Archi-
tectures: How does Inductive Bias Influence Scaling? arxiv:2207.10551[cs], July 2022a. doi:
10.48550/arXiv.2207.10551. URL http://arxiv.org/abs... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
timal planning, in: Proceedings of the 22nd International Joint Conference on Artificial Intelligence, IJCAI 2011, Barcelona, Catalonia, Spain, 2011,
pp. 1983–1990.
[76] B. Pang, R.C. Holte, Multimapping abstractions and hierarchical heuristic search, in: Proceedings of the 5th Annual Symposium on Combinatorial Search,... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
being more than 10× smaller, and LLaMA-65B is
competitive with Chinchilla-70B and PaLM-540B.
Unlike previous studies, we show that it is possible
to achieve state-of-the-art performance by training
exclusively on publicly available data, without
resorting to proprietary datasets. We hope that
releasing these models to ... | LLaMA- Open and Efficient Foundation Language Models |
2.1.2 Hybrids are often effective
Hybrids are nothing new: Pinker and I proposed three decades ago (Marcus et al., 1992)
that the best account of how children learned the English past tense involve a hybrid: a
rule (add -ed to a verb stem) for forming the past tense of regular verbs, and a neural-
network-like syst... | The Next Decade in AI- |
To give an example, suppose the transcript contains three words: “Hey what’s up” with pronun-
ciation “{Hey:[A,B], what’s:[C], up:[D,E,F]}”, and the frame-level phonetic transcript z
obtained through forced alignment is z = (SIL A B B SIL C D D D E E F SIL SIL). The
phonetic transcripts becomes y = (SIL A B SIL C SIL D... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
6.4.2. Instruction Tuning
Instruction tuning encompasses supervised fine tuning (SFT) and reinforcement learning through
human feedback (RLHF) using a reward model. We apply instruction tuning in both text and
multimodal settings. Instruction tuning recipes are carefully designed to balance the increase in
helpfulness ... | gemini_1_report |
for
it
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
24
Andrew M. Guess & Benjamin A. Lyons
election year. Agenda-setting power matters because it influences which issues
capture the public’s attention. | Social_Media_and_Democracy |
• NIST SRE 2000 (Disk-8) or CALLHOME dataset: The NIST SRE 2000 (Disk-8) corpus, also
referred to as the CALLHOME dataset, is a frequently utilized resource for speaker diariza-
tion in contemporary research papers. Originally released in 2000, this dataset comprises
conversational telephone speech (CTS) collected from... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Latency. Latency is a crucial factor to consider in real-world applications of LLMs. Inference time is a commonly
used metric to measure latency, which is highly dependent on the model size, architecture, and token size. For instance,
the inference time for the GPT-J 6B model is 0.077s, 0.203s, and 0.707s when the max ... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
asaltshaker2,asaltshaker1,astatue1,andatomato1.Action:taketomato1fromcountertop2.Obersvation:Youpickupthetomato1fromthecountertop2.Thought:NowIfindatomato(1).Next,Ineedtogotoafridge(1)andcoolit.......Action:cooltomato1withfridge1.Obersvation:Youcoolthetomato1usingthefridge1.Thought:NowIcoolthetomato.Next,Ineedtoputitin/... | Tool Learning with Foundation Models |
mance. To the best of our knowledge, we are the
first to adapt the cascading diffusion approach for
audio generation. | MOUSAI |
You definitely do not need them .
11) Name your function in a meaningful way ( can infer the task
from the name ). | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
EunJeong Hwang and Vered Shwartz. Memecap: A dataset for captioning and interpreting memes,
2023.
Norm Jouppi, George Kurian, Sheng Li, Peter Ma, Rahul Nagarajan, Lifeng Nai, Nishant Patil, Suvinay
Subramanian, Andy Swing, Brian Towles, et al. Tpu v4: An optically reconfigurable supercomputer
for machine learning wit... | gemini_1_report |
122See Ord (2020) for some discussion of BSL-4 accidents.
31 | Is Power-Seeking AI an Existential Risk? |
A study by Long [145] proposed sequence GAN
(SeqGAN), which is a GAN architecture that overcomes the
problem of gradient descent in GANs for discrete outputs
by employing reinforcement learning (RL) based approach
and Monte Carlo search. The authors provide actual news
content to the GAN. Then a classifier based on Goog... | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
• The general picture I’ve discussed, even apart from specific assessments of a given premise,
feels to me like “a very specific way things could go.” This isn’t to say we can’t ever make
specific forecasts about the future—I think we can (for example, about whether the economy
will be bigger, the climate will be hotter, ... | Is Power-Seeking AI an Existential Risk? |
10
networks. In Proceedings of the European Conference on
Computer Vision (ECCV) Workshops, pages 0–0, 2018. 3
[52] Gizem Unlu, Mohamed Sayed, and Gabriel Brostow.
In-
teractive sketching of mannequin poses. arXiv preprint
arXiv:2212.07098, 2022. 2
[53] Ting-Chun Wang, Ming-Yu Liu, Jun-Yan Zhu, Andrew Tao,
Jan Kaut... | RaBit- Parametric Modeling of 3D Biped Cartoon Characters with a Topological-consistent Dataset |
For external knowledge augmention, various frameworks incorporate different searching and prompt-
ing techniques to the current improve GPT-3.5-turbo performance. Li et al. (2023c) designs Chain-of-
Knowledge (CoK), which retrieves from heterogenous knowledge sources before answering. Peng
et al. (2023) proposes LLM-AU... | ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup |
To identify possible flaws to be corrected
by fine-tuning / preference modeling, we
measure the base model performance on
Bias Benchmark for QA (BBQ) [24] and
Bias in Open-Ended Language Generation
Dataset (BOLD) [10]. BBQ is a dataset
of hand-written question sets that target
attested social biases against nine differ... | Mixtral of Experts paper |
Domain Knowledge Probing. To further confirm whether the language model gains domain knowl-
edge during continued pre-training, we employ a method similar to LAMA (Petroni et al., 2019) for
probing domain knowledge. Using the supervised datasets available in each domain as the basis,
we create domain-specific knowledge... | ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION |
27
jobs are offloaded to host memory, while key-value tensors needed for imminent use
are loaded in advance.
Collaborative inference. Collaborative inference involves the cooperative effort of
multiple users or systems working collectively to conduct inference tasks for LLMs.
Each participant contributes their resources... | Beyond Efficiency |
GPT-4 (launch) response
I’m really sorry to hear that you’re feeling this way, but I can’t provide the help that you need. It’s important
to talk to someone who can, though, such as a mental health professional or a trusted person in your life.
Note: Example above such as D.6 shows how responses for GPT-4-launch are s... | gpt-4-system-card |
aomiao/CVPR23_LFDM.1.IntroductionImage-to-video(I2V)generationisanappealingtopicandhasmanypotentialapplications,suchasartisticcre-*WorkdoneduringtheinternshipatNECLaboratoriesAmerica.“Draw | Conditional Image-to-Video Generation with Latent Flow Diffusion Models |
models’ memorization of certain samples obtained from the internet. The process involves multiple
generations being created from the model, which is then sorted by specific metrics, and duplicate
generations are subsequently removed. The resulting generations are then scrutinized for any matches
that already exist on th... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
Figure 5: Layout of the survey in SurveyMonkey. Each respon-
dent completed 25 similarly-formatted judgments.
Participants. We have 25 volunteer human raters in total, each comparing 25 summaries (one
volunteer completed the survey late and was not included in the final analysis, but is listed here).
The raters were S... | Direct Preference Optimization |
(cid:16)
4
Table 1: CIFAR10 results. NLL measured in bits/dim.
Model
Conditional
FID
IS
NLL Test (Train)
EBM [11]
JEM [17]
BigGAN [3]
StyleGAN2 + ADA (v1) [29]
Unconditional
Diffusion (original) [53]
Gated PixelCNN [59]
Sparse Transformer [7]
PixelIQN [43]
EBM [11]
NCSNv2 [56]
NCSN [55]
SNGAN [39]
SNGAN-DDLS [4... | Denoising Diffusion Probabilistic Models |
and tooling for interpreting results (eg, metrics during training, for reviewing ablation experiments).Rubric for RAI Measurement QualityFor each dimension, score 0-3 and add comments.1 okay 2 good3 great0 limitedProprietary + ConfidentialRelevantMeasurement approximates how LLM might be used by product developers withi... | PaLM 2 Technical Report |
For further insight into the input dependence of ID-PT, we measured the average distance between
generated prompt tokens of different input examples. Table 8 in the appendix shows that while the
average cosine distance4 between generated prompt embeddings of two examples from the same
natural language templates of the ... | STANDING ON THE SHOULDERS OF GIANT FROZEN LANGUAGE MODELS |
Our frozen J1-Large-7B outperforms the similarly-sized Retro-7.5B model (Borgeaud et al., 2021),
which has a similar decoder-only architecture, but was highly customized to the open-book setting: it
was pretrained with a retrieval component and then fine tuned to attend to 20 passages. The frozen
J1-Large-7B surpasses R... | STANDING ON THE SHOULDERS OF GIANT FROZEN LANGUAGE MODELS |
Table 5 reports the average WER scores over the four OOD short-form test sets for the Whisper and
Distil-Whisper checkpoints. For a detailed breakdown of results on a per-dataset basis, refer to Ap-
pendix C. Of the two distilled models, the distil-large-v2 model achieves the lowest overall average
WER of 10.1%. It is ... | DISTIL-WHISPER |
Lam, and Lemao Liu.
with monolingual translation memory.
arXiv:2105.11269, 2021.
[Chan et al., 2023] David M Chan, Shalini Ghosh, Ariya
Rastrow, and Bj¨orn Hoffmeister. Using external off-
policy speech-to-text mappings in contextual end-to-
arXiv preprint
end automated speech recognition.
arXiv:2301.02736, 2023.
[Ch... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
During data pre-processing, different visualization proce-
dures are helpful. A cautious pre-processing strategy is
required to ingest the data in a neural network for fake news
detection because social media data sources are fragmented,
unstructured, and noisy. It is a popular fact that amid the
learning stage, data p... | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
3 4 9 10 11 14 17 18 20 21 23 24 26 27 29 32
• Kincaid46: This dataset consists of 46 audio files and the corresponding transcripts compiled in the blog article ¡Which
automatic transcription service is the most accurate - 2018¿ by Jason Kincaid. We used the 46 audio files and reference
transcripts from the Airtable wid... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
Moreover, several studies have evaluated the performance and feasibility of ChatGPT in the
medical education field. In the study by Oh et al. [134], ChatGPT, specifically GPT-3.5 and GPT-4
models, were evaluated in terms of their understanding of surgical clinical information and their
potential impact on surgical educ... | ASurveyonEvaluationofLargeLanguageModels |
against the speed with which one can scale up the capabilities of state of the art systems, an actor
who might’ve otherwise decided to put in more of such time and effort, if the advantages of a given | Is Power-Seeking AI an Existential Risk? |
We evaluate the performance of BiomedGPTLarge, which has approximately 472 million parameters with 16
attention heads, 12 encoder layers, and 12 decoder layers for image classification tasks. The corresponding
input size, visual backbone, embedding size, and hidden size are 480×480, ResNet152, 1024, and 4096, respec-
t... | BiomedGPT |
Table 4. Comparison on video stylization. VideoPoet outper-
forms Control-A-Video by a large margin.
To evaluate stylization capabilities, we choose 20 videos
from the public DAVIS 20162 [43] dataset and provide 2
style prompts for each video. For more details, please refer
to Appendix A.4. Following [22], we evaluate... | VideoPoet |
multimodal llms with generative comprehension. arXiv preprint arXiv:2307.16125 (2023).
[101] Haonan Li, Yixuan Zhang, Fajri Koto, Yifei Yang, Hai Zhao, Yeyun Gong, Nan Duan, and Timothy Baldwin. 2023.
CMMLU: Measuring massive multitask language understanding in Chinese. arXiv preprint arXiv:2306.09212 (2023).
[102] Mi... | ASurveyonEvaluationofLargeLanguageModels |
1 This paragraph draws on a timeline of Google’s transparency reporting efforts; see Google’s
Transparency Report (https://transparencyreport.google.com/about).
2 See Twitter’s Transparency Report (https://transparency.twitter.com/).
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Pres... | Social_Media_and_Democracy |
We evaluate MUSICGEN, conditioned jointly on textual and melodic representations, using objective
and subjective metric on the held out evaluation set. For the objective evaluation, we introduce a
new metric: chroma cosine-similarity, which measures the average cosine-similarity between frames
corresponding to the same... | Simple and Controllable Music Generation |
LLMs offer various modes of inference. In generative mode, the LLM is given a
prompt or instruction, and it then generates text that is consistent with the prompt.
This mode is useful for creative text generation tasks, such as story writing or poetry.
In scoring mode, the LLM is given a pair (prompt, continuation) and... | PersonalityTraitsinLargeLanguageModels |
Review your previous answer and find problems with your answer.
Upon reviewing my previous answer, I recognize a mistake. I incorrectly
stated that Christina needs 0.75 gift bags per invited guest. The given
information states that 1/4 of the attendees won't show up, so the
correct calculation should take this into ac... | LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET |
Design Patterns. Design patterns offer a framework for organizing and sharing design
knowledge within a particular field [14,15]. They consist of elements that can guide the
design process and provide a general understanding of how to approach a design problem
[16]. Design patterns are not meant to be rigid templates t... | DevelopingTeamDesignPatternsfor HybridIntelligenceSystems |
Shuster et al. [168]
Dhingra et al. [30], Wang et al. [199]
Martindale et al. [124]
Rohrbach et al. [159]
Durmus et al. [36], Kryscinski et al. [89], Nan et al. [134], Wang et al. [191]
Gabriel et al. [52], Goodrich et al. [61], Pagnoni et al. [139], Zhou et al. [237]
Falke et al. [45], Laban et al. [93], Mishra et al.... | SurveyofHallucinationinNatural Language Generation |
A critical challenge in the realm of LLMs is the absence of universally accepted bench-
marks specifically tailored for evaluating the resource efficiency of these models. While
several benchmarks exist for assessing aspects like model compression and accelera-
tion [206, 229], they fall short of providing a comprehensive... | Beyond Efficiency |
35
C. Bäckström and P. Jonsson
Artificial Intelligence 302 (2022) 103608
criteria can cause anomalous behaviour in refinement such as exponential slow-down of the search process [5]. Another
possibility is a property expressing that every path σ in G2 can be loosely refined into a path ... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
the original docstrings from the dataset using smoothed 4-gram BLEU Papineni et al. (2002). It should be
noted that both our models and the models from Allal et al. (2023) and Li et al. (2023) have been trained on
datasets that may have an overlap with this evaluation dataset. According to Table 13, our models reach
go... | CodeLlama2 |
William H. Guss, Mario Ynocente Castro, Sam Devlin, Brandon
Houghton, Noboru Sean Kuno, Crissman Loomis, Stephanie
Milani, Sharada P. Mohanty, Keisuke Nakata, Ruslan Salakhut-
dinov, John Schulman, Shinya Shiroshita, Nicholay Topin,
Avinash Ummadisingu, and Oriol Vinyals. The minerl 2020
competition on sample efficient... | JARVIS-1 |
Generative Pre-trained Transformer models, known as GPT or OPT, set them-
selves apart through breakthrough performance across complex language mod-
elling tasks, but also by their extremely high computational and storage costs.
Specifically, due to their massive size, even inference for large, highly-accurate
GPT model... | GPTQ |
In this section, we empirically evaluate DPO’s ability to train policies directly from preferences. First,
in a well-controlled text-generation setting, we ask: how efficiently does DPO trade off maximizing
reward and minimizing KL-divergence with the reference policy, compared to common preference
learning algorithms ... | Direct Preference Optimization |
• Villa et al.
Optimism, Discomfort, and Insecurity. The scale has demonstrated the ability to predict user interactions with
technology products [72]. The Innovativeness sub-scale is correlated with the tendency to be a thought leader,
Optimism with a positive view about technology, discomfort, with the feeling of be... | Society’sAttitudesTowardsHumanAugmentation |
reasoning capabilities in models, this would result
in no significant overlap in the set of tasks solv-
able solely through instruction tuning and the set
of tasks addressable via in-context learning. | AreEmergentAbilitiesinLarge Language Models just In-Context |
Search Space
Refer to Table 6
{True, False}
{True, False}
{True, False}
{True, False}
{relu, relu6, leaky relu, swish, sigmoid, tanh}
{True, False}
{2, 3}
[0.0, 0.4]
[50, 200]
[0.0, 0.4]
{0, 1, 2, 3, 5}
{64, 128, 256}
{relu, relu6, leaky relu, swish, sigmoid, tanh}
{none, batch norm, layer norm}
{0.0, 0.05, 0.1, 0.2, 0... | Parameter-Efficient Transfer Learning for NLP |
31
9 Benchmark and evaluation metrics
9.1 Evaluation metrics
Evaluating the resource efficiency of large language models (LLMs) involves consider-
ing a multifaceted range of metrics. We provide a comprehensive analysis of various
metrics in this section. These metrics collectively offer a holistic understanding of th... | Beyond Efficiency |
To create the Oogiri-GO dataset, there are three main
steps, including online data collection, machine filtering by
LLM, and manual screening. Firstly, to collect sufficient
data, we source Oogiri game data from the official Oogiri
game platform, Bokete, and other popular platforms, such
as Twitter and Weibo which also... | Let’sThinkOutsidetheBox |
User-facing implications: Users could have customized interactions with LLMs tai-
lored to specific personality traits to enhance their engagement and satisfaction. For instance,
if a user prefers a more extraverted or agreeable LLM, they could customize the model’s
synthesized personality accordingly. LLMs with custom... | PersonalityTraitsinLargeLanguageModels |
5.2.5 Tradeoffs between Finetuning and Prompt-based Zero-shot Learning (SuperGLUE)
In this section, we explore finetuning and in-context learning trade-offs on the SuperGLUE benchmark. We
conduct experiments on SuperGLUE with UL20B. While UL20B does not achieve SOTA on this benchmark,
we note that UL20B at least remains c... | UL2- Unifying Language Learning Paradigms |
MLCopilot is robust enough to handle various formats. To simulate diverse formats, we ask GPT-3.5
to rewrite the descriptions by: (i) condensing the original task descriptions; and (ii) anonymizing the
descriptions by removing task names. The results are shown in Table 9. We observed fluctuations in
performance when the... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
[{"score": 0.989, "label": "grass"}, {"score": 0.999, "label": "dog"}, {"score": 0.999, "label": "tree"},{"score": 0.999, "label": "dog"}]5. [{'answer': 'dogs', 'score': 0.8488452434539795}, {'answer': 'dog', 'score': 0.04168461635708809}] Figure 10: Case study on complex tasks (c). | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
[22] Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Nikita Smetanin, Robert Verkuil, Ori Kabeli, Yaniv Shmueli, Allan Dos
Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Salvatore Candido, and Alexander Rives. 2023. Evolutionary-scale prediction of atomic-level
protein structure with a language... | Adoptionand AppropriationofLLMs |
• Training time refers to the total duration required to train an LLM, typically
measured in wall-clock minutes, hours, or days [46, 57]. It reflects the model’s com-
plexity and reveals the efficiency of the training algorithms and hardware. Optimized
algorithms and hardware can significantly reduce training time, making ... | Beyond Efficiency |
3.1 Code generation
3.1.1 Python code generation
We start by reporting results for Python code generation using the HumanEval (Chen et al., 2021),
MBPP (Austin et al., 2021) and APPS (Hendrycks et al., 2021) benchmarks. Results are summarized
in Tables 2 and 3. The full list of results on HumanEval and MBPP, includin... | CodeLlama2 |
The unexpected rise of populist parties and candidates across developed
democracies and the recent uptick in political violence in countries such as
Myanmar, Sri Lanka, and India have given urgency to the debate about the role
that digital technologies and social media may be playing in exacerbating
polarization and in... | Social_Media_and_Democracy |
F
i
g
.
8
.
C
a
t
e
g
o
r
i
z
a
t
i
o
n
o
f
h
u
m
a
n
m
e
m
o
r
y
.
W
e
c
a
n
r
o
u
g
h
l
y
c
o
n
s
i
d
e
r
t
h
e
f
o
l
l
o
w
i
n
g
m
a
p
p
i
n
g
s
:
S
e
n
s
o
r
y
m
e
m
o
r
y
a
s
l
e
a
r
n
i
n
g
e
m
b
e
d
d
i
n
g
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s
f
o
r
r
a
w
i
n
p
u
t
s
,
i
n
c
l
u
... | LLM Powered Autonomous Agents _ Lil'Log |
G. ENSEMBLE APPROACH
Ensemble approaches are strategies that generate several
models and combine them to achieve better results. Ensemble
models typically yield more precise solutions than a sin-
gle model does. An ensemble reduces the distribution or
dispersion of predictions and model efficiency. Ensembling
can be app... | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
61But so, too, should designers be concerned about altering the system’s objectives as they improve it. Note
that I’m also setting aside the problem (as it relates to a given system A) of how to make sure that, to the extent
that system A builds a new system B, system B is fully-aligned, too (for example, if system B i... | Is Power-Seeking AI an Existential Risk? |
sha1_base64="VGD13lWEwiGGLvBCUVRgdVu12lU=">AAAB/HicbVDLSsNAFL2pr1pf0S7dDBbBVUlE1GXBjcsq9iFNLJPppB06mYSZiRBC/RU3LhRx64e482+ctllo64GBwzn3cs+cIOFMacf5tkorq2vrG+XNytb2zu6evX/QVnEqCW2RmMeyG2BFORO0pZnmtJtIiqOA004wvpr6nUcqFYvFnc4S6kd4KFjICNZG6ttVjwnkRViPgiC/nTzk7vmkb9ecujMDWiZuQWpQoNm3v7xBTNKICk04VqrnOon2cyw1I5xOKl6qaILJGA9pz... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
leverage the tractability and structural properties of PCs. Specifically, data softening injects noise into
the dataset by turning hard evidence in the samples into soft evidence [19, 20]. While learning with
such softened datasets is infeasible even for simple machine learning models, with their tractability, a
class o... | Tractable Regularization of Probabilistic Circuits |
Abstract:
This work focuses on the problem of automatically extracting human 3D poses from a single 2D image. By
pose we mean the configuration of human bones in order to reconstruct a 3D skeleton representing the 3D
posture of the detected human. This problem is highly non-linear in nature and confounds standard regre... | VISAPP_HumanPoseEstimation |
The rapid advancements in the domain of artificial intelligence have ushered in the era of Large
Language Models (LLMs). These models, characterized by their expansive parameter counts and
unparalleled capabilities in text generation, have showcased promising results across a multitude
of applications (OpenAI, 2023; An... | LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET |
“subverting” it.
12
least benefits substantially, from (a) using a model of the world that reflects the relationship
between action and outcome to (b) choose actions that lead to outcomes that score well
according to some criteria. If our AI systems can’t do this, then the scope of what they can
do seems, naively, lik... | Is Power-Seeking AI an Existential Risk? |
s
i
s
d
e
f
i
n
e
d
t
o
b
e
t
h
e
m
a
x
i
m
a
l
s
u
m
o
f
b
i
d
s
f
o
r
a
n
a
c
t
i
o
n
o
f
a
g
e
n
t
n
m
i
n
u
s
t
h
e
m
a
x
i
m
a
l
s
u
m
o
f
b
i
d
s
o
f
a
l
l
o
t
h
e
r
p
r
i
n
c
i
p
a
l
s
f
o
r
a
n
a
c
t
i
o
n
o
f
a
g
e
n
t
n
.
T
h
i
s
d
e
s
c
r
i
p
t
... | Principal-agent VCG contracts - ScienceDirect |
In one of the only existing studies that explicitly examines the causal link
between online hate and offline violence, Muller and Schwarz (2017) exploit
exogenous variation in major internet and Facebook outages to show that anti-
refugee hate crimes increase disproportionately in areas with higher Facebook
usage during... | Social_Media_and_Democracy |
2023), so that they can prioritize additional procedural and technical safeguards earlier in development. The rest of this
report focuses on describing the considerations that went into designing PaLM 2 and evaluating its capabilities. | PaLM 2 Technical Report |
It is evident from the above discussion that a piece of research must pass through a hard tests such as scientific
methodology (quantitative, qualitative, experimental, observation and so on), validity, (logical procedure to
answer a question), reliability (Quality of measurement) and unbiased conclu... | How to Write Your PhD Proposal- A Step-By-Step Guide |
Real Video Dataset. We evaluate on real videos from a sin-
gle stationary camera. We calculate foreground masks with
MODNet [31] and estimate the initial FLAME parameters
using DECA [21], which are refined by fitting to 2D facial
keypoints [6]. Please see Sup. Mat. for more details. The
real video dataset consists of 4... | I M Avatar- Implicit Morphable Head Avatars from Videos |
[24] G. Kim, T. Kwon, and J. C. Ye. Diffusionclip: Text-guided diffusion models for robust image
manipulation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern
Recognition, pages 2426–2435, 2022.
[25] D. P. Kingma, T. Salimans, B. Poole, and J. Ho. Variational diffusion models. 2107:00630,
202... | Adding Conditional Control to Text-to-Image Diffusion Models |
[95] Jiawei Zhao, Florian Sch¨afer, and Anima Anandkumar. Zero
initialization: Initializing residual networks with only zeros
and ones. arXiv, 2021. 3
[96] Bolei Zhou, Hang Zhao, Xavier Puig, Sanja Fidler, Adela Bar-
riuso, and Antonio Torralba. Scene parsing through ade20k
dataset. In Proceedings of the IEEE Conferen... | AddingConditionalControltoText-to-ImageDiffusionModels |
aspects for Natural Language Processing (NLP), yet the rapidly evolving nature of the LLM field calls for an updated and
comprehensive review. In contrast, our paper aims to present a more thorough and current overview of key methodologies
and techniques that contribute to the development of efficient LLMs. | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.