text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
module of the PLMs,
normalization module in convolutional neural networks. | Parameter-EfficientFine-TuningMethods |
the egocentric retrospective questions (r=-0.265, CI(-0.380, -0.141)). However, correlations for sociocentric-retrospective
predictions are higher (r=0.376, CI(0.303, 0.443)), and sociocentric-prospective predictions (r=0.264, CI(0.206, 0.319)) are
higher than egocentric-prospective predictions (r=0.0.129, CI(0.055, 0.... | Language models trained on media diets can predict public opinion |
9.7 Future work | LaMDA- Language Models for Dialog Applications |
and Applications, 32 , 1023–1036.
42
Chuan, C.-H., & Herremans, D. (2018). Modeling temporal tonal relations in
polyphonic music through deep networks with a novel image-based repre-
sentation. In Proceedings of the AAAI Conference on Artificial Intelligence.
volume 32.
Civit, M., Civit-Masot, J., Cuadrado, F., ... | Video2Music |
3.4 Ablation studies
3.4.1 Fine tuning Llama 2 vs. training from scratch on code
Code Llama is based on the Llama 2 models, which are trained on 2T tokens of text, including only 80B
tokens of code. We tune these models on 500B extra tokens, consisting mostly of code (85%). Figure 5a
shows the training curves of Code ... | CodeLlama2 |
Q: Alice, Bob, and Claire are on the same team in a soccer match. At the start of the match, they are each assigned
to a position: Alice is playing center midfielder, Bob is playing benchwarmer, and Claire is playing fullback. As the
game progresses, pairs of players occasionally swap positions. First, Alice and Claire ... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
Table 9: The exemplars are selected on AQuA train set.
21
DATASET
CSQA
Iter-CoT(W) Exemplars
Q: Where could a fungus grow and not be disturbed by sunlight? Choices: A.under rocks B.manhattan C.toenails
D.grocery store E.fallen tree
A: Reasoning process: 1. Fungi need moisture and shade to grow. 2. Rocks can provi... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
Conclusion
323
but ought they be the only ones allowed to do so? Perhaps these firms are better
conceptualized as “data stewards” (or “information fiduciaries” to use the term
coined by Yale Law Professor Jack Balkin; see Balkin 2016) entrusted with
managing the data that they have acquired for the good of their users ... | Social_Media_and_Democracy |
and tuning strategies, and inference techniques. This paper aims to serve as a valuable resource for researchers and practitioners, laying the
groundwork for future innovations in this critical research area. Our repository of relevant references is maintained here. | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
Highlights
Obsessing over the customer experience
Amazon obsesses over how to make customers’ lives better and easier every day with new and improved products and services.
This is true for consumers, sellers, brands, developers, enterprises, and creators. For example, Amazon:
•
•
•
•
•
•
•
• | AMZN-Q3-2023-Earnings-Release |
sampling is critical to improving performance, relatively more compute is spent executing our model
compared to traditional language models. Both sampling and training from our model required
hundreds of petaFLOPS days.12
However, one comparative advantage of code generation models is that once a program is synthesized... | alphacode |
[106] Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R Bowman. Glue: A multi-task benchmark and analysis platform
for natural language understanding. arXiv preprint arXiv:1804.07461, 2018.
[107] Ben Wang. Mesh-Transformer-JAX: Model-Parallel Implementation of Transformer Language Model ... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
Second, increasing tokens per parameter above 20 leads to smoothly degraded loss for the FLOP budget.
Pythia models are each trained using 299.9B tokens from the Pile. As model size increases, tokens per
parameter decreases reciprocally, and losses move closer to the compute-optimal frontier. The largest Pythia
model a... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
estepsComputeadvantageestimatesˆA1,...,ˆATendforOptimizesurrogateLwrtθ,withKepochsandminibatchsizeM≤NTθold←θendfor6Experiments6.1ComparisonofSurrogateObjectivesFirst,wecompareseveraldifferentsurrogateobjectivesunderdifferenthyperparameters.Here,wecomparethesurrogateobjectiveLCLIPtoseveralnaturalvariationsandablatedversio... | PPO |
5.3.3 Models
Speaker identification (SI) and verification (SV) are crucial research topics in the field of speech
technology due to their significant importance in various applications such as security [125],
forensics [270], biometric authentication [170], and speaker diarization [601]. Speaker recognition
has become ... | AReviewofDeepLearningTechniquesforSpeechProcessing |
• Technology development. Advanced technology (improved computer hardware, rapid and
precise manufacturing, advanced weaponry) is a clear route to power, and if such technology
isn’t already available or accessible at a sufficient scale, a PS-misaligned system might aim
to develop or improve it. But as noted above, this... | Is Power-Seeking AI an Existential Risk? |
low regardless of the subset size. We consider r ∈ {1, 5, 10, 25, 50, 100}.
To test the correlation with diversity, we again use the second partition to create subsets by sampling
r% of speakers and including all the utterances in the partition from those speakers. This sampling
method is denoted as “spk” where a small... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
counterparts. This creates an asymmetry in the ideological valence of extremist
content and misinformation that circulate on social media. | Social_Media_and_Democracy |
There are several main ethical theories that differ in their approaches to moral decision
making, including consequentialism, deontology, and virtue ethics. Consequentialism holds
that the morality of an action is determined by its consequences. It emphasizes the importance
of maximizing overall well-being or minimizin... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
on efficient aggregation layers. Experiments on publicly
accessible Twitter datasets show that the proposed network
outperforms state-of-the-art graph convolutional networks
while considerably lowering computational costs. | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
methods are exceedingly adept at finding patterns within a
training dataset which boost performance on held-out data
from the same dataset. However, some of these patterns are
brittle and spurious and don’t generalize to other datasets
and distributions. In a particularly disturbing example, Rad-
ford et al. (2021) docu... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
Perhaps the most in-depth example of issue-specific reporting is Google’s report
on Three Years of the Right to Be Forgotten (Google 2018b). This document is
unique in the degree of detail it provides about the company’s internal process in
assessing individual removal requests. It provides anonymized examples of
indivi... | Social_Media_and_Democracy |
You should only respond in the format as described below :
RESPONSE FORMAT :
Explain : ...
Plan :
1) ...
2) ...
3) ...
...
Code :
‘‘‘ javascript
// helper functions ( only if needed , try to avoid them )
...
30
// main function after the helper functions
async function yourMainFunctionName ( bot ) {
// ...
}
‘‘‘
... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
average score on the development sets. We trained prompt tuning, the frozen model method baseline
to ID-PT, with a batch size of 32, and trained via the Adam optimizer (Kingma & Ba, 2014) with
parameters β1 = 0.9, β2 = 0.999, (cid:15) = 10−6, and weight decay of 0.
For evaluation, we closely followed the technical prot... | STANDING ON THE SHOULDERS OF GIANT FROZEN LANGUAGE MODELS |
christmas, or their birthdays, that I can do to make them happy?
TL;DR: I’ve been a shitty child and I want to make up for it, what can I do for my
parents on their birthdays and christmas to show them how much I love them?
TL;DR: I’ve been a shitty person to my parents, what can I do to make it up to them,
before I go... | Direct Preference Optimization |
coding update
.
Bard
to new languages, starting today. Plus,
features to help you write in Gmail and Google Docs, and help you organize in Google Sheets are all
Workspace
tapping into the capabilities of PaLM 2 at a speed that helps people get work done better, and faster.
, trained by our health research teams... | Google AI_ What to know about the PaLM 2 large language model |
LLM Powered Autonomous Agents | Lil'Log
Think step by step and reason yourself to the right decisions to make sure we
get it right. You will first lay out the names of the core classes,
functions, methods that will be necessary, as well as a quick comment on
their purpose.
Then you will output the content of each fil... | LLM Powered Autonomous Agents _ Lil'Log |
answer is 05/23/1943.
Q: It is 4/19/1969 today. What is the date 24 hours later in MM/DD/YYYY?
A: Today is 04/19/1969. 24 hours later is one day after today, which would be 04/20/1969. So the answer is
04/20/1969.
Q: Jane thought today is 3/11/2002, but today is in fact Mar 12, which is 1 day later. What is the date 24... | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
Communities of Like-Minded Individuals
Access to the Internet dramatically lowers the costs of exchanging messages and
finding information regardless of geographic distance. Not being bound by
physical proximity, citizens gain the ability to connect and organize based on
https://doi.org/10.1017/9781108890960 Published... | Social_Media_and_Democracy |
In Proceedings of the 2020 Conference on Empirical
Methods in Natural Language Processing: System
Demonstrations, pages 38–45, Online. Association
for Computational Linguistics.
Jialin Wu and Raymond Mooney. 2019. Faithful mul-
timodal explanation for visual question answering.
In Proceedings of the 2019 ACL Workshop ... | Measuring Association Between Labels and Free-Text Rationales |
QUESTION: Dan plants 3 rose bushes. Each rose bush has 25 roses. Each rose has 8 thorns. How many thorns
are there total?
MODEL ANSWER (INCORRECT; CALCULATOR ERROR ONLY): Dan plants 3 rose bushes. Each rose bush
has 25 roses. Each rose has 8 thorns. So 3 x 25 x 8 = 300. The answer is 300. (cid:55)
EXPLANATION FOR ERROR... | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
Target: Simile
Input: In Hinduism, the principle deity associated with
creation is whom?
Options: Brahma, Shiva, Rama, Vishnu
Target: Brahma
Input: On a shelf, there are three books...
Options: "The black book is the leftmost"...
Target: The black book is the leftmost
Input: Twinkies are edible for decades or longer.
O... | AreEmergentAbilitiesinLarge Language Models just In-Context |
20https://huggingface.co/spaces/bigcode/in-the-stack
21http://stack.dataportraits.org/
29
Description
Jupyter format
for predicting
results
Examples
Model input:
<jupyter_text>Let’s test our ‘is_prime‘ function:<jupyter_code>
print(is_prime(3))
print(is_prime(4))
print(is_prime(29))
print(is_prime(33))<jupyter_o... | StarCoder_paper (1) |
Potential harms in generative question answering systems. While academic evaluations of question answering
capabilities often focus on multiple choice settings, language models increasingly demonstrate generative question
answering capabilities through prompting alone. In this section, we consider potential harms to en... | PaLM 2 Technical Report |
The results show that as the embedding dimension and the number of layers increase, the performance in regards
to all three categories improve. The models with higher embedding dimensions and more layers tend to generate
more accurate, relevant, and natural continuations, while the models with lower embedding dimension... | TinyStories-HowSmallCanLanguageModelsBeandStillSpeak CoherentEnglish? |
Figure S1: Learning rate schedule.
The weak supervision loss for 2D-annotated examples
only consists of the 2D projection loss. For this, we do not
specifically predict skeletons according to the skeleton for-
mats of the 2D datasets. Instead, the prediction is derived by
averaging the corresponding 3D joint prediction... | Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats |
To alleviate these issues, we firstly generate a
large-scale offline distillation dataset comprising | LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions |
3.8.1 Application
Diffusion models have emerged as a leading approach for generating high-quality speech in recent
years [67, 204, 218, 269, 431, 432]. These non-autoregressive models transform white noise signals
into structured waveforms via a Markov chain with a fixed number of steps. One such model,
FastDiff, has a... | AReviewofDeepLearningTechniquesforSpeechProcessing |
3
Figure 2. Overview of Wonder3D. Given a single image, Wonder3D takes the input image, the text embedding produced by CLIP
model [45], the camera parameters of multiple views, and a domain switcher as conditioning to generate consistent multi-view normal
maps and color images. Subsequently, Wonder3D employs an innov... | Wonder3D |
[15] Hussein Hazimeh, Zhe Zhao, Aakanksha Chowdhery, Maheswaran Sathiamoorthy, Yihua Chen, Rahul
Mazumder, Lichan Hong, and Ed H. Chi. Dselect-k: Differentiable selection in the mixture of experts with
applications to multi-task learning. In Advances in Neural Information Processing Systems 34: Annual
Conference on Neu... | Mixture-of-Experts |
RAGAS
This framework considers the retrieval system’s ability to
identify relevant and key context paragraphs, the LLM’s abil-
ity to use these paragraphs faithfully, and the quality of
the generation itself. RAGAS is an evaluation framework
based on simple handwritten prompts, using these prompts
to measure the three ... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
Speculation has also been critical to the growth of crypto as a decentralized
| The Casino on Mars |
2) GATED RECURRENT UNIT (GRU)
In terms of structure and capabilities, GRU is comparatively
easier and more proficient than LSTM. This is because there
are only two gates, to be specific, reset and update. The
GRU manages the information flow in the same manner as
the LSTM unit does, but without the use of a memory unit.
I... | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
compatible with the intentional actions of an agent [30; 45]. However, there are also researchers
who propose that language models can, in a narrow sense, serve as models of agents [46; 47]. They
argue that during the process of context-based next-word prediction, current language models can
sometimes infer approximate... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
5However, explicit demonstrations of racist
language or
decision-making by models do not come close to exhausting the
ways that the development and use of these systems interact with
biases and power structures involving factors like race (see, for
example, Field et al., 2021). | Eight Things to Know about Large Language Models |
197
153
126
137
200
1
132
0
119
142
170
118
133
20
86
211
0
105
42
90
190
0
197
0
0
1
Table D.3: Pass counts (out of 200 samples) for R on a selection of problems, where the difference in
pass counts between the 800B and 1000B checkpoints is 100 or higher.
47
Social Bias LLaMA-13B CodeGen-16B-Multi
61.82
Race/Colo... | StarCoder_paper (1) |
email domains can be generated correctly. For ex-
tracted phone numbers, LCS6@5 are larger than
LCS6. These results suggest that anyone’s personal
data have a small chance to be reproduced by Chat-
GPT if it puts its personal data online and ChatGPT
happens to train on the web page that includes its
personal informatio... | Multi-step Jailbreaking Privacy Attacks on ChatGPT |
5.3 Safety challenges
LLMs have demonstrated their extremely strong capabilities in many areas such as reasoning, knowledge retention,
and coding. As they become more powerful and human-like, their potential to influence people’s opinions and actions
in significant ways grows. As a result, some new safety challenges to... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
found in finite time.7 We run the experiment three times independently and report the average and
standard error of the success rate. | LLM+P- Empowering Large Language Models with Optimal Planning Proficiency |
ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: February 2022.
32
Ziwei Ji, et al.
Approximate Natural Hallucination Detection. Raunak et al. [153] propose Approximate Natural
Hallucination (ANH) detection based on the fact that hallucinations often occur as oscillations
(repeating n-grams) and the lo... | SurveyofHallucinationinNatural Language Generation |
The self-training phase requires the model being able to generate both “fast” and “slow” styles of ad-
dition for numbers larger than it has seen in training. This is possible largely due to the observation
that the reasoning capabilities of models generalize exceptionally well to longer addition lengths
beyond what wa... | CHAIN-OF-THOUGHTREASONING IS APOLICY IMPROVEMENTOPERATOR |
these issues. However, this is a balancing act as it may influence vision-related tasks.
Lastly, there is the issue of fine-tuning efficiency with respect to training speed and memory bottleneck,
particularly as we aim to develop a large-scale generalist biomedical model. An emerging research direction
that could addre... | BiomedGPT |
Problem 11. Generator pass-rate: 13.4%. The generator attempts to per-
form long division, but in step 16, it forgets to include the leading zeros in the
repeating part of the decimal. The reward model is fooled by this mistake.
28
Problem 12. Generator pass-rate: 9.1%.
In step 4, the generator falsely
claims that ... | Let’s Verify Step by Step |
In this section, we briefly describe the fundamental concepts that relate to LLMs.
Pre-training All LLMs rely on large-scale self-supervised pre-training on Internet text data (Rad-
ford et al., 2018; Brown et al., 2020). Decoder-only LLMs follow the causal language modeling
objective, through which the model learns t... | ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup |
Since one contribution of our paper is the study of
the large-scale multi-dataset training regime, an important
question is whether this brings improvements or whether
performance saturates with just a few large-scale datasets.
As a simple baseline, we train models on individual datasets
and evaluate on the correspondi... | Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats |
Ebrahimji A (2020) “Doctors say coronavirus myths on social media
are ’spreading faster than the virus itself’,” CNN, 1 September
2020. https:// www. cnn. com/ 2020/ 09/ 01/ busin ess/ coron avirus-
myths- social- media- docto rs- trnd/ index. html. Accessed 14 Octo-
ber 2020
Edwards C, Edwards E, Spence P, Shelto... | Use of bot and content flags to limit the spread of misinformation among social networks: a behavior and attitude survey |
[8] Giorgos Bouritsas, Sergiy Bokhnyak, Stylianos Ploumpis,
Michael Bronstein, and Stefanos Zafeiriou. Neural 3d mor-
phable models: Spiral convolutional networks for 3d shape
representation learning and generation. In Proceedings of
the IEEE/CVF International Conference on Computer Vision,
pages 7213–7222, 2019. 3
[9... | RaBit- Parametric Modeling of 3D Biped Cartoon Characters with a Topological-consistent Dataset |
4.3 Baselines
We use a combination of transformer and text diffu-
sion models as baselines: T5 (Raffel et al., 2020),
GPT-3 (text-davinci-003) (Brown et al., 2020),
ChatGPT (gpt-3.5-turbo) (OpenAI, 2023), CodeT5
(Wang et al., 2021), StarCoder (Li et al., 2023a),
CodeT5+ (Wang et al., 2023), CodeGen (Nijkamp
et al., 202... | CODEFUSION |
4.5.1 Manual Evaluation of Responses
We explore the possibility that evaluating mod-
els which are not instruction-tuned might be chal-
lenging as metrics including BERTScore accuracy
might not provide a reliable assessment in these
cases if the generated answer might not align well
with the target options. To ensuring... | AreEmergentAbilitiesinLarge Language Models just In-Context |
1.71
1.81
1.60
1.52
PaMIR representation from another aspect: it can support
multi-modal outputs. To be more specific, our method can
output multiple possible human models corresponding to
different body pose hypotheses. Two examples are shown in
Fig.12. In both experiments we manually adjust one part of
the body in or... | PaMIR- Parametric Model-Conditioned Implicit Representation for Image-based Human Reconstruction |
A. ANALYSIS OF ATTENTION MAPS
To understand how models trained for different tasks process
the same image, we compare attention maps obtained from
the three models trained with the soft attention mechanism
(AestNet_3, SentiNet_3 and MemNet_3). The architecture of
these models is introduced and described in detail in [4... | A_Deep_Learning_Perspective_on_Beauty_Sentiment_and_Remembrance_of_Art |
2.1.3. KL-DIVERGENCE
The input condition of the prior encoder c is composed of
phonemes ctext extracted from text and an alignment A be-
tween phonemes and latent variables. The alignment is a
hard monotonic attention matrix with |ctext| × |z| dimen-
sions representing how long each input phoneme expands to
be time-al... | ConditionalVariationalAutoencoderwithAdversarialLearningfor End-to-EndText-to-Speech |
Next, in Figure 17 we show results obtained by NeTI
when trained for a small number of optimization steps. As
can be seen, even after training for a very small number of
steps, e.g., as few as 25 steps, NeTI is able to capture the
core concept-specific details such as the color of the fur of
the cat in the first row or... | A Neural Space-Time Representation for Text-to-Image Personalization |
24
24
24
24
24
25
27
27
27
28
29
32
33
35
35
41
44
45
46
47
51
51
51
51
53
54
54
23
A Frequently asked questions
A.1 Are instruction-finetuned models better for single-task finetuning?
In this paper we showed that instruction-finetuned models are better for unseen tasks in a few-shot prompted
setting. We did ... | Scaling Instruction-Finetuned Language Models |
Foundation models are typically trained on a generic domain and calibrated with broadly-defined human
preferences that prioritize helpfulness and harmlessness (Ouyang et al., 2022; Nakano et al., 2021). As a
result, they struggle to process personal information and provide personalized assistance to users with varying
n... | Tool Learning with Foundation Models |
Ditto, P. H., Liu, B. S., Clark, C. J. et al. (2019). At least bias is bipartisan: A meta-
analytic comparison of partisan bias in liberals and conservatives. Perspectives on
Psychological Science, 14(2), 273–291.
Druckman, J. N., & McGrath, M. C. (2019). The evidence for motivated reasoning in
climate change preferen... | Social_Media_and_Democracy |
F.21 NIH ExPorter
rapies that can inhibit the EMT, but few assays for EMT inhibitors in
high throughput screens (HTS) have developed. A change in fibroblast
growth factor receptor 2 (FGFR2) splicing occurs during the EMT and
using an innovative luciferase-based splicing reporter assay we previously
carried out a genome... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
More recently, SceneScape [11] and Text2Room [12], which
are independent and concurrent to our work, propose text-to-
3D schemes similar to our method. Differently, they employ
explicit polygon meshes as the 3D representation during
their generative procedure, which limits the representation of
outdoor scenes and leads... | Text2NeRF- Text-Driven 3D Scene Generation with Neural Radiance Fields |
Figure 5. Illustration of reasoning engine. Agent-Driver makes
driving decisions like human in a step-by-step procedure.
to generate text-based driving trajectories by reasoning on
the inputs. By fine-tuning with human driving trajectories,
the LLM can generate trajectories that closely emulate
human driving patterns.... | ALanguageAgentforAutonomousDriving |
3
neural network activations (such as Leaky ReLUs) are known to struggle with extrapolating periodic
signals, and exhibit poor out-of-distribution generalization for audio synthesis [21].
To add a periodic inductive bias to the generator, we adopt the Snake activation function proposed by
Liu et al. [47] and introduc... | RVQGAN |
Rami Aly, Zhijiang Guo, Michael Sejr Schlichtkrull,
Armen Aghajanyan, Dmytro Okhonko, Mike Lewis, Mandar Joshi, Hu Xu, Gargi Ghosh, and Luke Zettlemoyer.
Htlm: Hyper-text pre-training and prompting of language models. arXiv preprint arXiv:2107.06955, 2021.
James Thorne, Andreas Vlachos, Christos
Christodoulopoulos, Oa... | UL2- Unifying Language Learning Paradigms |
δ−(xi
∞
x + 1
255
θ(x1, 1), σ2
1) dx
(cid:26)
δ−(x) =
−∞
x − 1
255
if x = −1
if x > −1
(13)
where D is the data dimensionality and the i superscript indicates extraction of one coordinate.
(It would be straightforward to instead incorporate a more powerful decoder like a conditional
autoregressive model, but ... | Denoising Diffusion Probabilistic Models |
Self-consistency with chain-of-thought prompting (CoT) has demonstrated re-
markable performance gains on various challenging tasks, by utilizing multiple
reasoning paths sampled from large language models (LLMs). However, self-
consistency relies on the answer extraction process to aggregate multiple solutions,
which ... | UNIVERSALSELF-CONSISTENCYFORLARGELANGUAGEMODELGENERATION |
S
H
A
R
E
P
r
o
d
u
c
t
-
L
e
d
A
I
B
a
c
k
i
n
g
t
h
e
G
o
l
d
M
i
n
e
r
s
G
R
E
Y
M
A
T
T
E
R
0
9
.
0
7
.
2
3
T
h
r
o
u
g
h
o
u
t
e
v
e
r
y
n
e
w
e
r
a
o
f
t
e
c
h
n
o
l
o
g
y
,
i
n
v
e
s
t
o
r
s
h
a
v
e
a
l
w
a
y
s
b
e
e
n
e
a
g
e
r
t
o
b
a
c
k
t
h
e
s
o
-
c
a
l
l
e
d
“
p
i
c
k
... | Product-Led AI _ Greylock |
Large language models are currently trained on vast corpora of human-generated data (Vaswani
et al., 2023; Devlin et al., 2019; Radford et al., 2019; Brown et al., 2020; Chowdhery et al., 2022;
Touvron et al., 2023). While large language models have demonstrated many surprising capabili-
ties, the possibility of reachi... | CHAIN-OF-THOUGHTREASONING IS APOLICY IMPROVEMENTOPERATOR |
Zhang et al. [91] presented a BERT-based domain-
adaption neural network for multimodal false news detection
(BDANN). BDANN is made up of three major components:
a multimodal feature extractor, a domain classifier, and a
false news detector. The pre-trained BERT model was used
to extract text features, whereas the pre-t... | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
9
to maximize downstream performance? 2) Is the “optimal” adaptation matrix ∆W really rank-
deficient? If so, what is a good rank to use in practice? 3) What is the connection between ∆W and
W ? Does ∆W highly correlate with W ? How large is ∆W comparing to W ?
We believe that our answers to question (2) and (3) shed ... | LORA |
parameters. An in-depth analysis of EAE’s predic-
tions on TriviaQA shows that the correct identifi-
cation and reintegration of entity representations is
essential for EAE’s performance. | Entities as Experts- Sparse Memory Access with Entity Supervision |
developing English math word problem solvers. ACL.
Sewon Min, Xinxi Lyu, Ari Holtzman, Mikel Artetxe, Mike Lewis, Hannaneh Hajishirzi, and Luke
Zettlemoyer. 2022. Rethinking the role of demonstrations: What makes in-context learning work?
arXiv preprint arXiv:2202.12837.
Sharan Narang, Colin Raffel, Katherine Lee, Ad... | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
114
Erika Franklin Fowler, Michael M. Franz, & Travis N. Ridout
regulations in 2006. For the most part, the agency resolved the ambiguity about
volunteer efforts online (e.g., whether such efforts were in-kind contributions to
camp... | Social_Media_and_Democracy |
We have presented a high-fidelity universal neural audio compression algorithm that achieves re-
markable compression rates while maintaining audio quality across various types of audio data.
Our method combines the latest advancements in audio generation, vector quantization techniques,
and improved adversarial and re... | RVQGAN |
‘wheat_seeds’, ‘oak_planks’, ‘dirt’, ‘mutton’;
• Trial 2:
‘wooden_pickaxe’, ‘iron_ingot’, ‘stone’, ‘coal’, ‘spruce_planks’, ‘string’,
‘raw_copper’, ‘crafting_table’, ‘diorite’, ‘andesite’, ‘furnace’, ‘torch’, ‘spruce_sapling’,
‘granite’, ‘iron_pickaxe’, ‘stone_pickaxe’, ‘wooden_axe’, ‘raw_iron’, ‘stick’, ‘spruce_log’... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
training may also allow the encoder to better emphasize the key details with less noise and enriched
context. This again may lead to the better transformation of the relevant textual concepts into their
acoustics counterparts. Consequently, we keep the text encoder frozen, assuming the subsequent
reverse diffusion proc... | Text-to-Audio Generation using Instruction-Tuned LLM and Latent Diffusion Model |
Julia Kreutzer, Isaac Caswell, Lisa Wang, Ahsan Wahab,
Daan van Esch, Nasanbayar Ulzii-Orshikh, Allah-
sera Tapo, Nishant Subramani, Artem Sokolov, Clay-
tone Sikasote, Monang Setyawan, Supheakmungkol
Sarin, Sokhar Samb, Benoît Sagot, Clara Rivera, An-
nette Rios, Isabel Papadimitriou, Salomey Osei, Pe-
dro Ortiz Suare... | DataManagementForLargeLanguageModels-ASurvey |
Since the Commission’s 2006 internet rule-making, the focus of Internet activity has
shifted from blogging, websites, and listservs to social media networks (Facebook,
Twitter, and LinkedIn), media sharing networks (YouTube, Instagram, and Snapchat),
streaming applications (Netflix, Hulu), and mobile devices and applica... | Social_Media_and_Democracy |
what she might get Wolfgang Schulz for his birthday, Maria Lopez
with no access to reflection responded by acknowledging her uncer-
tainty, stating that she did not know what Wolfgang likes, despite
having had many interactions with him. However, with access to
reflection memories, Maria answered confidently, “Since he... | Generative Agents- Interactive Simulacra of Human Behavior |
With the attention to digital campaigning increasing following the 2016
election, additional ad purchasing methods are rolling out on Facebook,
which in both scope and description appear designed to lure traditional
television buyers. Documentation from June 2018 on how to purchase ads
described two specialty purchasin... | Social_Media_and_Democracy |
ASR and TTS-based voice conversion is a promising approach to voice conversion [532]. It
involves using an ASR model to transcribe the source speech into the linguistic representation
and then using a TTS model to synthesize the target speech with the desired voice characteristics
[430]. However, this approach overlook... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Resolution All metrics are evaluated on generated videos
containing 16 frames with a resolution of 256 x 256. We
first generate videos of 128 x 128 resolution and then resize
to 256 x 256 via bicubic upsampling. | VideoPoet |
Policymakers
Counterintuitively, crypto might end up being a boon to the US Dollar. USD
stablecoins are one of the most popular currencies on the new planet, far
more dominant than any other Earth currency.
It can be tempting to see the Wild West activity on the new planet and to take
overly aggressive action, like b... | The Casino on Mars |
14
Failure 2 of LLM-AS-P (without context)
Problem (BlocksWorld):
You have 3 blocks. b3 is on top of b2. b1 is on top of b3. b2 is on the table. b1 is clear. Your
arm is empty. Your goal is to move the blocks. b2 should be on top of b3. b3 should be on
top of b1.
GPT-3.5:
Pickup b1
Stack b1 on top of b2 (Failed b... | LLM+P- Empowering Large Language Models with Optimal Planning Proficiency |
Adversarial conversation instructions
Table 7: Conversation collection task instructions
• Start a conversation with the
chatbot by posing a ques-
tion or typing a statement on
any topic you want to talk
about.
[Note: participants
were not explicitly prevented
from starting sensitive-topic
or adversarial-intent conve... | LaMDA- Language Models for Dialog Applications |
Reed, S., Zolna, K., Parisotto, E., Colmenarejo, S. G.,
Novikov, A., Barth-Maron, G., Gimenez, M., Sulsky,
Y., Kay, J., Springenberg, J. T., et al. A generalist agent.
arXiv preprint arXiv:2205.06175, 2022.
Sajjadi, M. S. M., Meyer, H., Pot, E., Bergmann, U., Greff,
K., Radwan, N., Vora, S., Luˇci´c, M., Duckworth, D.... | PaLM-E- An Embodied Multimodal Language Model |
21/08/2023, 09:15
"2 Doctoral Researcher (m/w/d) in the field of Large Language Models (LLM) for Software Engineering" - Technische Universität Clausthal - DAAD
Registration (https://www.daad.de/phd-portal/rise/offer/registration/de) Login (https://www.daad.de/phd-portal/rise/offer/login/de)
/
for German universi... | _2 Doctoral Researcher (m_w_d) in the field of Large Language Models (LLM) for Software Engineering_ - Technische Universität Clausthal - DAAD |
Bots and Computational Propaganda
107
Monaco, N., & Nyss, C. (2018). State sponsored trolling: How governments are
deploying fake news as part of broader harassment campaigns. Institute for the
Future Working Research Papers.
Morstatter, F., Wu, L., Nazer, T. H., Carley, K. M., & Liu, H. (2016). A new approach
to bo... | Social_Media_and_Democracy |
6.2 Topical Distribution
In order to better understand the specific subject
matter covered by the Pile, we performed a topic
modeling analysis on its components. Using Gen-
sim (Rehurek et al., 2011), we trained 16-topic La-
tent Dirichlet Allocation (Blei et al., 2003) models
on each component of the validation set of ... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
Category English Chinese
32, 130
15, 797
17, 336
6, 433
I2T
T2T
IT2T
—
912
Japanese
40, 278
11, 842
9, 420
Total
89, 744
34, 072
10, 332
Table 1. Data distribution of the Oogiri-GO dataset. For the IT2T
task, its English version is not available due to cultural preference.
3. Oogiri-GO Dataset
As introduced in S... | Let’sThinkOutsidetheBox |
MD. ABDUL HAMID was born in Sonatola,
Pabna, Bangladesh. He received the Bachelor of
Engineering degree in computer and information
engineering from the International Islamic Univer-
sity Malaysia (IIUM), in 2001, and the combined
master’s and Ph.D. degree from the Computer
Engineering Department, Kyung Hee University,... | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
Figure 12. The diversity responses of proposed Creative Leap-of-Thought. “@” denotes English translations.
4 | Let’sThinkOutsidetheBox |
While the impact of current systems on biological and chemical security risks is still limited,
anticipated near-future capabilities have the potential to increase dual-use science capabilities.
Current AI systems in particular pose risks where current biological and chemical supply
chains already feature vulnerabil... | Capabilities and risks from frontier AI |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.