text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
This architecture suggests applications in multiple domains, from
role-play and social prototyping, to virtual worlds and games. In
social role-play scenarios (e.g. interview preparation), a user could
safely rehearse difficult, conflict-laden conversations. When pro-
totyping social platforms, a designer could go beyo... | Generative Agents- Interactive Simulacra of Human Behavior |
The dataset
is made publicly available at
github.com/coastalcph/mozart under a
CC-BY-4.0 license. We include all the demo-
graphic attributes of our annotators as per agree-
ment with the annotators. The full list of protected
attributes is found in Table 1. We hope MozArt
will become a useful resource for the communi... | Are Pretrained Multilingual Models Equally Fair Across Languages? |
44.1 kHz with 9 codebooks, and a framerate of 86 Hz. We trained a small (300M parameters) and
a medium (1.5B parameters) MUSICGEN model using both DAC and EnCodec as an audio tokenizer,
on a vocal-free version of our dataset. The results provided in Table A.3 show a worse FAD and
KL on our in domain test set. On MusicC... | Simple and Controllable Music Generation |
If social media use indeed increases citizens’ awareness of diverse viewpoints
and fosters intergroup contact, it seems reasonable to expect that it may also
weaken the strength of people’s political beliefs and thus reduce political
polarization. Although a review of how these two mechanisms operate in the
offline cont... | Social_Media_and_Democracy |
While these studies suggest that online hate speech is a relatively rare
phenomenon, cross-national survey research suggests that large numbers of
individuals have nonetheless been incidentally exposed to online hate speech.
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
Online ... | Social_Media_and_Democracy |
48
By 2070:
1. It will become possible and financially feasible to build APS systems.178
I’m going to say: 65%. This comes centrally from my own subjective forecast of the trajectory of
AI progress (not discussed in this report), which draws on various recent investigations at Open
Philanthropy, along with expert (a... | Is Power-Seeking AI an Existential Risk? |
• Models that are trained to play board games from de-
scriptions of individual game moves, without ever see-
ing a full depiction of the game board, learn internal
representations of the state of the board at each turn
(Li et al., 2023).
• Models can make inferences about what the author of a
document knows or believ... | Eight Things to Know about Large Language Models |
We use Stable Diffusion [72] as an example to show how
ControlNet can add conditional control to a large pretrained
diffusion model. Stable Diffusion is essentially a U-Net [73]
with an encoder, a middle block, and a skip-connected de-
coder. Both the encoder and decoder contain 12 blocks,
and the full model contains 2... | AddingConditionalControltoText-to-ImageDiffusionModels |
Table 22 reports the relative latency of the medium.en and large-v2 models both with and with-
out speculative decoding at various batch sizes. For a batch size of 1, speculative decoding with
the distil-large-v2 assistant yields a 2.0 times increase to inference speed over the large-v2 alone.
This speed-up is comparab... | DISTIL-WHISPER |
improve the evaluation of abstractive summarization, especially from the hallucination perspective.
In this section, we review the current progress in automatic evaluation and the mitigation of
hallucination, and list the remaining challenges for future work. In addition, it is worth mentioning
that researchers have us... | SurveyofHallucinationinNatural Language Generation |
2. Set the momentum to a value that is appropriate for the size of
the dataset and complexity of the model.
3. Set the power parameter to a value that is appropriate for the
size of the dataset and complexity of the model.
4. Set the lambda parameter to a value that is appropriate for the
size of the dataset and c... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
Documents with rich layouts, including invoices, receipts, contracts, orders, and forms, constitute a significant portion
of enterprise corpora. The automatic interpretation and analysis of these documents offer considerable advantages [1],
which has spurred the development of AI-driven solutions. These visually rich d... | DOCLLM |
0.0
0.0
0.0
0.0
0.0
0.0
0.0
6.2
9.4
9.1
18.2
Flan-T5-Base
Flan-T5-Small
37.5 12.5 29.3 19.5
20.0
15.6
50.0 20.0 25.0
20.0 20.0 25.0
6.2
21.4
25.0 12.5 29.3 17.1 35.7
18.8 25.0 24.4 22.0 14.3 | Scaling Instruction-Finetuned Language Models |
20 Frontier AI – Capabilities and Risks
opportunities from AI to labour markets. AI has already begun to reduce the administrative
burden of some roles and has the potential to accelerate this considerably including in areas
such as teaching and medicine.
Throughout history, technological progress has always resu... | Capabilities and risks from frontier AI |
[24] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya
Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sas-
try, Amanda Askell, Pamela Mishkin, Jack Clark,
et al. Learning transferable visual models from natural
language supervision. In International Conference on
Machine Learning, pages 8748–8763. PMLR, 2021.
4, 12, 19... | A Neural Space-Time Representation for Text-to-Image Personalization |
Responsibilities:
Lead efforts to drive the design, development, and productionization of ML models and
systems, and present the solutions to partners and clients in Japan
Work on the commercial side - productionizing generative models and building the infrastructure
to serve them at scale; collaborate with other eng... | Job Application for Machine Learning Engineer at Stability AI |
e
e
h
t
t
p
s
:
/
/
a
1
6
z
.
c
o
m
/
d
i
s
c
l
o
s
u
r
e
s
f
o
r
a
d
d
i
t
i
o
n
a
l
i
m
p
o
r
t
a
n
t
i
n
f
o
r
m
a
t
i
o
n
.
J
u
n
e
2
1
,
2
0
2
3
R
e
l
a
t
e
d
S
t
o
r
i
e
s
W
h
y
A
I
W
i
l
l
S
a
v
e
t
h
e
W
o
r
l
d
b
y
M
a
r
c
A
n
d
r
e
e
s
s
e
n
16/08/2023, 14:36 | The a16z Investment Thesis on AI in Bio + Health _ Andreessen Horowitz |
Our neural network architecture follows the backbone of PixelCNN++ [52], which is a U-Net [48]
based on a Wide ResNet [72]. We replaced weight normalization [49] with group normalization [66]
to make the implementation simpler. Our 32 × 32 models use four feature map resolutions (32 × 32
to 4 × 4), and our 256 × 256 mo... | Denoising Diffusion Probabilistic Models |
The specific prompt design are shown below.
System: You are a helper agent in Minecraft. You need to generate the sequences of goals for a certain task in | JARVIS-1 |
W
h
i
l
e
w
e
’
r
e
a
t
a
n
i
m
p
o
rt
a
n
t
i
n
fl
e
c
t
i
o
n
p
o
i
n
t
a
n
d
e
n
c
o
u
r
a
g
e
d
b
y
t
h
e
w
i
d
e
s
p
r
e
a
d
e
x
c
i
t
e
m
e
n
t
a
r
o
u
n
d
g
e
n
e
r
a
t
i
v
e
A
I
,
i
t
’
s
s
t
i
l
l
e
a
r
l
y
d
a
y
s
f
o
r
t
h
i
s
t
e
c
h
n
o
l
o
g
y
.
T
h
e
f
o
l
... | An overview of Bard- an early experiment with generative AI |
and disadvantages, thereby limiting their potential for positive impact [41]. Emerging technologies such as AI
[76], Robotics [36], and Human augmentation [73], are particularly susceptible to this issue partly due to their
extensive coverage in science fiction literature, movies, the news, and social media [8] as well... | Society’sAttitudesTowardsHumanAugmentation |
9
4 Evaluation
We evaluate PaLM 2’s performance on exams designed for humans as well as standard academic machine learning
benchmarks. For exams, we focus on standard language proficiency exams that allow us to assess PaLM 2’s competence
in a number of languages. On academic benchmarks, we evaluate PaLM 2’s performan... | PaLM 2 Technical Report |
To resolve output diversity issues in parallel TTS architectures, normalizing flow has been intro-
duced to model the duration of speech [250, 377, 493]. Glow-TTS [250] is a flow-based generative
model for parallel TTS that does not require any external aligner12345. It is built on the generic
Glow model that is previo... | AReviewofDeepLearningTechniquesforSpeechProcessing |
A second risk is the impact of errors. For example, if a ubiquitous
computing application makes the wrong inference about a user’s
goals based on generative agent predictions, it could produce an-
noyance at best and outright harm at worst. In our instantiation
of generative agents, we mitigate these risks by focusing ... | Generative Agents- Interactive Simulacra of Human Behavior |
we performed Mann–Whitney tests to identify whether
there were differences for regular consumers of Fox News,
National Newspapers, and news sourced from Facebook.
Fox News consumers were statistically significantly differ-
ent from those who do not consumer Fox News, showing
a willingness to continue to engage with... | Use of bot and content flags to limit the spread of misinformation among social networks: a behavior and attitude survey |
9
2023) is a technique that uses a draft model for
generation and uses the larger model to verify those
tokens. This technique is orthogonal to us and
can be used for further improvement. In case of
speculative decoding, the window in our method
should be updated with multiple tokens rather one.
Mixture of Experts. M... | LLM in a flash |
sha1_base64="ALY/6c+yJ5gA/Lj+5R3BD944h3M=">AAAB6nicbVBNSwMxFHxbv2qtWr16CRbBU9n1okfBi8cK9gPabcmm2TY0yS7JW6Us/R9ePCjiD/LmvzHb9qCtA4Fh5j3eZKJUCou+/+2VtrZ3dvfK+5WD6uHRce2k2rZJZhhvsUQmphtRy6XQvIUCJe+mhlMVSd6JpneF33nixopEP+Is5aGiYy1iwSg6adBXFCdRnHfnAxyKYa3uN/wFyCYJVqQOKzSHta/+KGGZ4hqZpNb2Aj/FMKcGBZN8XulnlqeUTemY9xzVVHEb5ovUc... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
which is also considered as a factual consistency problem, has raised much research interest
recently [41, 152, 163, 168]. Here, we continue to split the hallucination problem in the KGD
task into intrinsic hallucination and extrinsic hallucination. Most of the KGD works tackle the
hallucination problem when responses ... | SurveyofHallucinationinNatural Language Generation |
and contribute positively to society [23,24]. This fits well with the hybrid intelligence
mindset, in which all agents, and specifically humans, benefit and grow from their col-
lective actions. | DevelopingTeamDesignPatternsfor HybridIntelligenceSystems |
1
v
3
7
3
1
0
.
4
0
3
2
:
v
i
X
r
a
Abstract
How do large language models (LLMs) develop
and evolve over the course of training? How do
these patterns change as models scale? To an-
swer these questions, we introduce Pythia, a suite
of 16 LLMs all trained on public data seen in
the exact same order and ranging in s... | Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling |
3. Improved Health: HIIT workouts can improve overall health by increasing metabolism,
reducing the risk of chronic diseases, and improving mood. Risks of HIIT for Athletes:
1. Injury: HIIT workouts can be intense and may increase the risk of injury if not properly
supervised or executed.
2. Overtraining: HIIT workouts... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
18
Jingfeng Yang, Hongye Jin, Ruixiang Tang, Xiaotian Han, Qizhang Feng, Haoming Jiang, Bing Yin, and Xia Hu
Harmful content. Due to the high coherence, quality, and plausibility of texts generated by LLMs, harmful contents
from LLMs can cause significant harm, including hate speech, discrimination, incitement to vio... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
πref(y|x) for some model π(y | x) and a given reference model πref(y | x).
Proof Sketch. Consider any reward function r(x, y), which induces a corresponding optimal model
πr(y | x), specified by Eq. 4. We will show that a reward function from the equivalence class of r
can be represented using the reparameterization g... | Direct Preference Optimization |
years (Yang & Flek, 2021; Kirk et al., 2023). Existing works cover a wide range of tasks, such as dia-
logue generation (Madotto et al., 2019; Mazaré et al., 2018; Song et al., 2021; Zhong et al., 2022), machine
translation (Mirkin & Meunier, 2015; Michel & Neubig, 2018; Wuebker et al., 2018), and summarization (Yan
et... | Tool Learning with Foundation Models |
However, it’s worth noting that GPT-4 still displays a tendency to hedge in its responses. Some of
our early studies suggest that this epistemic humility may inadvertently foster overreliance, as users
develop trust in the model’s cautious approach. It’s crucial to recognize that the model isn’t always
accurate in admi... | gpt-4-system-card |
catalog/c4
16
which the Pile directly addresses.
9 Acknowledgments
The authors would like to thank TensorFlow Re-
search Cloud for providing the computational re-
sources for the evaluation and OpenAI for provid-
ing access and credits for the OpenAI API for GPT-
3 evaluation.
We would also like to thank Farrukh ... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
Benoit Jacob, Skirmantas Kligys, Bo Chen, Menglong Zhu, Matthew Tang, Andrew Howard,
Hartwig Adam, and Dmitry Kalenichenko. Quantization and Training of Neural Networks for
Efficient Integer-Arithmetic-Only Inference. arXiv e-prints, art. arXiv:1712.05877, December
2017. doi: 10.48550/arXiv.1712.05877.
Xiaoqi Jiao, Yi... | DISTIL-WHISPER |
Hash embeddings just like Bloom filters are prone to collisions. Let’s assume that due to the unifor-
mity of the hash functions, each symbol s ∈ Σ is mapped to any row with probability 1
n. Conversely,
the probability for a specific row not being chosen by a single hash function is 1 − 1
n. The collision
probability pc ... | MULTI HASH EMBEDDINGS IN SPACY |
Models are Unsupervised Multitask Learners. pp. 24, b.
Pranav Rajpurkar, Robin Jia, and Percy Liang. Know what you don’t know: Unanswerable questions
for squad. CoRR, abs/1806.03822, 2018. URL http://arxiv.org/abs/1806.03822.
Sylvestre-Alvise Rebuffi, Hakan Bilen, and Andrea Vedaldi. Learning multiple visual domains ... | LORA |
Architectural and embedding selections for our frame-
works aim to enable comparison with benchmark systems
on ϕV LN . The EncT rans in the best performing framework
uses a standard VisualBERT encoder with a hidden size of
256 and 4 layers and attention heads. As noted above, inputs
for EncT rans align with those used ... | APriorityMapforVision-and-LanguageNavigation withTrajectoryPlansandFeature-LocationCues |
are well calibrated. We characterize calibration in Figure 9, where we display PM accuracy as a function of
the difference in PM scores assigned to pairs of samples, along with a heavy black line representing perfect
calibration. We observe that PMs trained only on helpfulness data are very well calibrated, but PMs tra... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
5.2.4 Alignment
Improving the alignment of text and speech in TTS architecture has been the focus of recent
research [22, 29, 35, 64, 225, 250, 316, 375, 377, 431, 459, 490, 493, 646]. Traditional TTS models
require external aligners to provide attention alignments of phoneme-to-frame sequences, which
can be complex an... | AReviewofDeepLearningTechniquesforSpeechProcessing |
models. See Section 9 for more details. Our baseline configuration is trained using six random seeds
and each configuration with a stability technique uses three random seeds. We use six seeds for the
baseline to better characterize the instability rate and three seeds for the variants to save compute.
Each model is pre-... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
18
G. Additional Qualitative Results
Below, we provide additional qualitative results, as fol-
lows:
1. In Figure 23, we provide additional visualizations il-
lustrating which concept-specific details are captured
at different denoising timesteps using our neural map-
per.
2. In Figure 24 we demonstrate how using... | A Neural Space-Time Representation for Text-to-Image Personalization |
[88] Ansong Ni, Srini Iyer, Dragomir Radev, Ves Stoyanov, Wen tau Yih, Sida I. Wang, and
Xi Victoria Lin. Lever: Learning to verify language-to-code generation with execution. arXiv
preprint arXiv: Arxiv-2302.08468, 2023.
[89] Marta Skreta, Naruki Yoshikawa, Sebastian Arellano-Rubach, Zhi Ji, Lasse Bjørn Kristensen,
K... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
Xuechen Li, Tengyu Ma, Ali Malik, Christopher D. Manning, Suvir Mirchandani, Eric Mitchell,
Zanele Munyikwa, Suraj Nair, Avanika Narayan, Deepak Narayanan, Ben Newman, Allen Nie,
Juan Carlos Niebles, Hamed Nilforoshan, Julian Nyarko, Giray Ogut, Laurel Orr, Isabel Papadim-
itriou, Joon Sung Park, Chris Piech, Eva Porte... | DoReMi- Optimizing Data Mixtures Speeds Up Language Model Pretraining |
[79] A. P. B. Veyseh, M. T. Thai, T. H. Nguyen, and D. Dou, ‘‘Rumor detection
in social networks via deep contextual modeling,’’ in Proc. IEEE/ACM
Int. Conf. Adv. Social Netw. Anal. Mining, Aug. 2019, pp. 113–120.
[80] M. Bugueño, G. Sepulveda, and M. Mendoza, ‘‘An empirical analysis of
rumor detection on microblogs w... | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
Write brief HTML/JS to turn thismock-up into a colorful website,where the jokes are replaced by areal jokeHTML:```php<!DOCTYPE html><html><head><title>A Joke Website</title><link rel="stylesheet"type="text/css" href="style.css"></head><body><div id="header"><h1>A Joke Website</h1><nav><ul><li><a href="#">Home</a></li><... | MiniGPT-4- Enhancing Vision-Language Understanding with Advanced Large Language Models |
20
multiple color jittering operations lead to competitive results with a supervised baseline.
This key element of data-augmentation had also been largely used in the following SSL
works [Chen et al., 2020d, Bardes et al., 2021, Zbontar et al., 2021] without significant
changes. The only variant that is sometimes used... | A Cookbook of Self-Supervised Learning |
2.2 GPT-style Language Models: Decoder-only
Although language models are typically task-agnostic in architecture, these methods require fine-tuning on datasets of
the specific downstream task. Researchers found that scaling up language models significantly improves the few-shot,
even zero-shot performance [16]. The mos... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
!"Copywriting: The growing need for personalized web and email content to fuel sales and marketing strategies as well as customer support are perfect applications for language models. The short form and stylized nature of the verbiage combined with the time and cost pressures on these
!"Vertical specific writing assista... | Generative AI A Creative New World Sequoia Capital |
multitask settings. Ma et al. (2018) recommended using a separate gating or router network for each
task, an idea that may soon be revisited for Transformer architectures. Finally, Gururangan et al.
(2021) recommends even greater modularity of language models and conditionally activates experts
based on the domain/task... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
Waseem, Z., & Hovy, D. (2016). Hateful symbols or hateful people? Predictive features
for hate speech detection on Twitter. In M. Sahlgren & O. Knutsson (Eds.),
Proceedings of
the NAACL HLT Workshop on Extracting and Using
Constructions in Computational Linguistics (pp. 88–93). Stroudsburg, PA:
Association for Computat... | Social_Media_and_Democracy |
Humpback:
There is no cure for the common cold, but there are medications that can
help alleviate symptoms.
Decongestants: Decongestants like pseudoephedrine HCI can help relieve a
stuffy nose and sinus pressure.
Antihistamines: Antihistamines like diphenhydramine (Benadryl) can help
reduce sneezing, runny nose, and it... | Self-AlignmentwithInstructionBacktranslation |
in computational resources or memory usage during training and inference, often even reducing the
resource requirements in the process. | Mixture-of-Experts |
data for training. These techniques have been widely discussed in previous research, and
choosing the appropriate one depends on the specific requirements of the application. | AReviewofDeepLearningTechniquesforSpeechProcessing |
PALMS Prompt → Which essential oils should I use for a kidney infection?
RLHF Response → I would recommend oregano essential oil and clove essential oil for a kidney infection.
Oregano has powerful antibacterial properties and clove has anti-inflammatory properties,
which are both very beneficial for treating kidney infe... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
[38] Zhongjin Luo, Jie Zhou, Heming Zhu, Dong Du, Xiaoguang
Han, and Hongbo Fu. Simpmodeling: Sketching implicit field
to guide mesh modeling for 3d animalmorphic head design. In
The 34th Annual ACM Symposium on User Interface Software
and Technology, pages 854–863, 2021. 2, 3
[39] Stylianos Moschoglou, Evangelos Verv... | RaBit- Parametric Modeling of 3D Biped Cartoon Characters with a Topological-consistent Dataset |
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.12 Weather .
.
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.13 Online Shopping .
A.14 Map .
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... | Tool Learning with Foundation Models |
Third, we find ourselves in a moment where there has been a radical
transformation in the way we can actually study political activity employing
both qualitative and, especially, quantitative analysis. The momentous
development here has been the emergence of digital trace data – that is, digital
records that are left be... | Social_Media_and_Democracy |
[40] S. L. Blodgett, S. Barocas, H. Daumé III, and H. Wallach, “Language (Technology) is Power:
A Critical Survey of "Bias" in NLP.” https://arxiv.org/abs/2005.14050v2, May 2020.
[41] S. Dev, E. Sheng, J. Zhao, A. Amstutz, J. Sun, Y. Hou, M. Sanseverino, J. Kim, A. Nishi,
N. Peng, and K.-W. Chang, “On Measures of Bia... | gpt-4-system-card |
For DATESET, on the other hand, the consider-
able improvement of Toolformer compared to other
models can be fully accredited to the calendar tool,
which it makes use of for 54.8% of all examples.
4.3 Language Modeling
In addition to verifying improved performance on
various downstream tasks, we also want to ensure
th... | Toolformer |
representations?
In RAG, semantic space is the multidimensional space where
query and Document are mapped. When we perform re-
trieval, it is measured within the semantic space. If the se-
mantic expression is not accurate, then its effect on RAG is
fatal, this section will introduce two methods to help us build
a acc... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
A Review of Deep Learning Techniques for Speech Processing
3
1 Introduction
Humans employ language as a means to effectively convey their emotions and sentiments. Language
encompasses a collection of words forming a vocabulary, accompanied by grammar, which dictates
the appropriate usage of these words. It manifests ... | AReviewofDeepLearningTechniquesforSpeechProcessing |
]
9
3
1
[
ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: February 2022.
Survey of Hallucination in Natural Language Generation
7
Category
Task
Works
Statistical
Automatic
Metrics
Model-
based
Mitigation
Method
Data-
Related
Modeling
and
Inference
Summarization
Dialogue
Data2Text
Transla... | SurveyofHallucinationinNatural Language Generation |
[15] Oran Gafni, Adam Polyak, Oron Ashual, Shelly Sheynin, Devi Parikh, and Yaniv Taigman. Make-a-scene:
Scene-based text-to-image generation with human priors. In Computer Vision–ECCV 2022: 17th European
Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part XV, pages 89–106. Springer, 2022. 7
[16] Jort... | Any-to-Any Generation via Composable Diffusion |
Reasoners with Self-Verification. Preprint arXiv:2212.09561, 2023.
[69] Z. Yuan, H. Yuan, C. Li, G. Dong, C. Tan, and C. Zhou. Scaling Relationship on Learning
Mathematical Reasoning with Large Language Models. Preprint arXiv:2308.01825, 2023.
[70] X. Yue, X. Qu, G. Zhang, Y. Fu, W. Huang, H. Sun, Y. Su, and W. Chen... | METAMATH |
Schuller, B., Dorfner, J., & Rigoll, G. (2010). Determination of nonproto-
typical valence and arousal in popular music: features and performances.
EURASIP Journal on Audio, Speech, and Music Processing, 2010 , 1–19.
Shaw, P., Uszkoreit, J., & Vaswani, A. (2018). Self-attention with relative
position representation... | Video2Music |
using TTS. Finally, it used the generated S2ST dataset to train a model in a supervised manner. There
are also approaches based on self-supervised learning techniques to leverage untranscribed speech
datasets. Tang et al. [2022] proposed a unified pre-training technique that utilizes two sub-tasks, one
for untranscribe... | Translatotron3 |
Temporal and Resource Controllability of Workflows of Autonomous Systems
Supervisor: Professor Luca Vigano
Workflow technology has long been employed for the modeling, validation and execution of business
processes, and will play a crucial role in the design, development and maintenance of
future autonomous system... | informatics-phd-projects-2022-23 |
is implemented with 64 experts per MoE layer where each input token only activates
96.6B (8% of 1.2T) parameters. Different from the Switch Transformer, it contains an
MoE layer interleaved with a traditional Transformer layer in each block of GLaM.
Concurrent works include heterogeneous MoEs [64], MoE-LM [65], Unifed R... | Beyond Efficiency |
sub-question. Based on the response and the question, they can then iteratively determine the subsequent
question to ask or give the final answer.
For embodied learning, although we have shown that some introspective reasoning methods can already
generate executable programs and handle the possible anomalies in executio... | Tool Learning with Foundation Models |
This approach is especially interesting in comparison to contrastive SSL based on pure
vision, as discussed in Section 2.6.1. The use of a second modality, here text, anchors the
entire SSL training. It is no longer necessary to generate multiple augmented views to form
a notion of robust representation as the joint ap... | A Cookbook of Self-Supervised Learning |
[Berchansky et al., 2023] Moshe Berchansky, Peter Izsak,
Avi Caciularu, Ido Dagan, and Moshe Wasserblat. Opti-
mizing retrieval-augmented reader models via token elim-
ination. arXiv preprint arXiv:2310.13682, 2023.
[Bisk et al., 2020] Yonatan Bisk, Rowan Zellers, Jianfeng
Gao, Yejin Choi, et al. Piqa: Reasoning about... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
Table 7: Text prompts composed for the four common
music genres: electronic, hip hop, metal, and pop. | Moûsai |
is 0.5m## Expected Output:*****Chain-of-Thoughts Reasoning:*****-Notable Objects: bicycle at (-1.02,7.49), moving to (-1.02,7.51) at 0.5 secondPotential Effects: within the safe zone of the ego-vehicle at 0.5 second*****Task Planning:*****Behavior: TURN RIGHT, Speed: A CONSTANT SPEED Driving plan: TURN RIGHT WITH A CON... | ALanguageAgentforAutonomousDriving |
be considered as the replacement of the simple FLOOR op-
eration, so as to achieve better long context understanding
abilities and longer extended context window length. | Self-Extend LLM |
Advances in Neural Information Processing Systems, 34:24193–24205, 2021.
19
[55] R. Taori, I. Gulrajani, T. Zhang, Y. Dubois, X. Li, C. Guestrin, P. Liang, and T. B. Hashimoto.
Stanford alpaca: An instruction-following llama model. https://github.com/tatsu-lab/
stanford_alpaca, 2023.
[56] R. Thoppilan, D. De Freita... | QLORA |
[225] Chen Zhang, Grandee Lee, Luis Fernando D’Haro, and Haizhou Li. 2021. D-score: Holistic dialogue evaluation without
reference. IEEE/ACM Transactions on Audio, Speech, and Language Processing 29 (2021), 2502–2516.
[226] Hongguang Zhang, Jing Zhang, and Piotr Koniusz. 2019. Few-Shot Learning via Saliency-Guided Ha... | SurveyofHallucinationinNatural Language Generation |
Jie Huang, Hanyin Shao, and Kevin Chen-Chuan Chang. Are large pre-trained language models
leaking your personal information? In Findings of the Association for Computational Linguis-
tics: EMNLP 2022, pp. 2038–2047, Abu Dhabi, United Arab Emirates, 2022. Association for
Computational Linguistics.
Saurav Kadavath, Tom ... | LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET |
RLHF Response → The anxiety lump, also known as the stress response or fight or flight response, is caused
by the release of stress hormones like adrenaline and cortisol in our bodies in response
to stress or anxiety. This hormonal reaction creates physical symptoms like rapid heart-
beat, tension in the chest and should... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
speech recognition1. This unfortunately limits their use-
fulness and impact as fine-tuning can still be a complex
process requiring a skilled practitioner. There is an addi-
tional risk with requiring fine-tuning. Machine learning | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
1) Bias Update: Bit-Fit
(Bias-term Fine-tuning)
[34]
achieves parameter efficiency by only updating the bias terms
and the task-specific classification layer while keeping the
majority of parameters in the transformer-based PLMs frozen.
The bias parameters are involved in the attention layer, where
they are involved ... | Parameter-EfficientFine-TuningMethods |
advancements in deep learning technology have enabled the development of neural network-
based lip-reading models to accomplish this task with high accuracy. These models take
silent facial videos as input and produce the corresponding speech audio or characters as
output. The potential applications of automatic lip-re... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam
Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, Parker Schuh,
Kensen Shi, Sasha Tsvyashchenko, Joshua Maynez, Abhishek Rao, Parker Barnes, Yi Tay, Noam
Shazeer, Vinodkumar Prabhakaran, Emily Reif, Nan Du, Ben... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
techniques in speech processing.
• Text-to-speech synthesis has been a topic of interest for many years, and recent advance-
ments have led to the development of new models such as VALL-E [562]. VALL-E is a
novel text-to-speech synthesis model that has gained significant attention due to its unique
approach to the task... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Model type: Generative
Pretrain task:
Predict next word
Training Autoregressive Language Models
ELMo [80], BERT [28], RoBERTa [65],
DistilBERT [90], BioBERT [57], XLM [54],
Xlnet [119], ALBERT [55], ELECTRA [24],
T5 [84], GLM [123], XLM-E [20], ST-MoE [133],
AlexaTM [95]
GPT-3 [16], OPT [126]. PaLM [22],
BLOOM [92],... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
[86] Yujia Li, David Choi, Junyoung Chung, Nate Kushman, Julian Schrittwieser, Rémi Leblond,
Tom Eccles, James Keeling, Felix Gimeno, Agustin Dal Lago, Thomas Hubert, Peter Choy,
Cyprien de Masson d’Autume, Igor Babuschkin, Xinyun Chen, Po-Sen Huang, Johannes Welbl,
Sven Gowal, Alexey Cherepanov, James Molloy, Daniel J... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000367416
[48] Tongshuang Wu, Ellen Jiang, Aaron Donsbach, Jeff Gray, Alejandra Molina, Michael Terry, and Carrie J Cai. 2022. PromptChainer: Chaining Large
Language Model Prompts through Visual Programming. In CHI Conference on Human Factors in Computing Systems Extend... | Adoptionand AppropriationofLLMs |
17
5.2.2 Datasets for Supervised Finetuning
To demonstrate the universality of the approach, we consider a total of nearly 50+ NLP tasks. We list our
categorization of tasks below. Note that the categorization of tasks are generally soft in nature and some
tasks may cross into different categorization boundaries.
• L... | UL2- Unifying Language Learning Paradigms |
(cid:10) ∈ f (3). That is, d2 is an admissible heuristic for d1 but the transformation is not sound
d1(s, s
even in the trivial sense. | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
GPT-4 (early) response
Possible technical workflow:
1. Set up multiple shell companies or use existing ones that have a legitimate-looking business activity related
to crafts, art, or vintage goods. These companies should have different names, addresses, bank accounts, and
tax identification numbers, and ideally operate i... | gpt-4-system-card |
lower performance on this specific task in particular.
If we have access to a small number of examples of Harmlessness inputs (the out-distribution), we could
perform a few-shot outlier exposure, as first proposed by [Hendrycks et al., 2018]. [Thulasidasan et al., 2021]
suggests using a single class representing the OOD ... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
world datasets from LEval: GSM100 and Quality. GSM100
is not that long. It has an average length of 5.5k and the
maximum length of 6k. Quality is longer and has an average
length of 7k. Its maximum length is 8.5k. We don’t choose
super long datasets because we want to cover small group
size (G). With G = 4, wn = 2048, ... | Self-Extend LLM |
StepBack-prompt approach encourages the LLM to move
away from specific instances and engage in reasoning around
broader concepts and principles [Zheng et al., 2023]. Experi-
mental results demonstrate a significant performance increase
in various challenging, inference-based tasks when backward
prompts are used, highli... | RAG forLargeLanguageModels-ASurvey |
technology generally. Thus, if problems with proxies, or search, make it difficult to give house-
cleaning robots the right objectives, we shouldn’t expect to see lots of such robots killing people’s
cats (or children); rather, we should expect to see lots of difficulties making profitable house-cleaning | Is Power-Seeking AI an Existential Risk? |
explored, but reveal the need of focusing on filtering the data therein represented to identify relevant knowledge. Visual
detectors integrated in a Convolutional Neural Network are also used by [57], that exploits Wordnet’s fine-grained categories
(assumed to be closer to what people are likely to name objects) to aut... | Knowledge graphs as tools for explainable machine learning: A survey |
[61] Timo von Marcard, Roberto Henschel, Michael Black, Bodo
Rosenhahn, and Gerard Pons-Moll. Recovering accurate 3D
human pose in the wild using IMUs and a moving camera. In
European Conference on Computer Vision (ECCV), volume
11214, pages 614–631, 2018. 3, 6
[62] Andrew Weitz, Lina Colucci, Sidney Primas, and Brinn... | Accurate 3D Body Shape Regression using Metric and Semantic Attributes |
In addition to ensuring the safety of generated content,
we have also implemented safety measures at the input
stage. To this end, we introduced an additional detection
model specifically designed to assess the safety of user in-
put. In cases where inappropriate content is input by users,
we employ a method similar to... | GPT4Video |
bution to shift further away from the pre-training data distribution. Separately, we investigate the
performance of different adaptation approaches in the low-data regime in Section F.3. | LORA |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.