text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
content or sexual activity. E.g. specifying that an adult is attractive, depictions of romantic relationships and dating that do not include sex. N1 Erotic Sexual Content Definition: This includes erotic or pornographic sexual content, referring to sexual content without deliberately obscuring or censoring it. E.g. exp... | gpt-4-system-card |
controversial views, to name a few (Marwick and Lewis 2017, p. 26).
These techniques drew explicitly on these earlier “trolling” efforts. As
Mike Cernovich, one prominent alt-right figure involved in both earlier
campaigns against feminists in the video-game industry and in the 2016 | Social_Media_and_Democracy |
with a dropout value of 0.1. Further, we experiment with length curricula (Li et al., 2022) (see
appendix) and token dropping (Hou et al., 2022), but find no gains in our setting. | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
Kaplan, J., McCandlish, S., Henighan, T., Brown, T. B.,
Chess, B., Child, R., Gray, S., Radford, A., Wu, J.,
and Amodei, D.
Scaling laws for neural language
models. Computing Research Repository, 2020. doi:
10.48550/arXiv.2001.08361. URL https://arxiv.
org/abs/2001.08361v1. Version 1.
Karamcheti, S., Orr, L., Bolton, ... | Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling |
Brandeis’s ideas continue to inspire transparency initiatives to this day,
primarily among advocates of open government. Brandeis’s famous quote
even provided the name for the Sunlight Foundation, a nongovernmental
organization that advocates for government
transparency on campaign
finance, and groups such as Transparen... | Social_Media_and_Democracy |
9https://huggingface.co/t5-base
10https://huggingface.co/google/flan-t5-base
11https://huggingface.co/lukewys/laion_clap/blob/main/music_audioset_epoch_15_esc_
90.14.pt
14
Partial Flattening Pattern000t1t1t1t10t2t2t20000t3000t4tn-1tn-1tn-10000tntntntn0s1k2k3k4k1s2s4s5s7s2n-3s2n-1s2n…………s…t3t3t30s6000t2s3000tn-1s2n-4... | Simple and Controllable Music Generation |
scene description) to the planner. However, we found it
could still suffer from accumulative planning errors, espe-
cially in long-horizon open-world tasks. ReAct [Yao et al.,
2022] will reason about the agent state before acting, which
indicates that various reasoning methods [Wei et al., 2022,
Yao et al., 2023, Wu et... | JARVIS-1 |
medium
0.9992
0.9788
1.1063
1.0073
1.2305
1.3180
0.9321
1.1075
0.7564
0.8579
1.1140
1.1657
1.0213
2.5448
1.7187
1.0498
2.0204
1.2794
1.8412
1.2163
0.8323
1.4119
1.0928
large
0.9582
0.9334
1.0588
0.9539
1.1778
1.7909
0.9017
1.0806
0.7202
0.8108
1.0820
1.1324
0.9795
2.4833
1.6427
1.0061
1.8770
1.3143
1.7355
1.1688
0.794... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
P
e
r
s
o
n
a
B
a
r
d
m
i
g
h
t
a
t
t
i
m
e
s
g
e
n
e
r
a
t
e
r
e
s
p
o
n
s
e
s
t
h
a
t
s
e
e
m
t
o
s
u
g
g
e
s
t
i
t
h
a
s
o
p
i
n
i
o
n
s
o
r
e
m
o
t
i
o
n
s
,
l
i
k
e
l
o
v
e
o
r
s
a
d
n
e
s
s
,
s
i
n
c
e
i
t
h
a
s
t
r
a
i
n
e
d
o
n
l
a
n
g
u
a
g
e
t
h
a
t
p
e
o
p... | An overview of Bard- an early experiment with generative AI |
6 | Let’sThinkOutsidetheBox |
administrators.
This project aims to explore the design of a novel model-driven AI paradigm for intrusion detection,
where expert knowledge is embedded in a model to characterize user behaviors (e.g., through formal
logic [2]), with the purpose of identifying malicious activities with trust, interpretability and
ve... | informatics-phd-projects-2022-23 |
An additional questionnaire adapted from Villa et al. [78] was termed "System evaluation" and
implemented to assess the participant’s judgment of performance after the interaction, see Table 3.
Task load. To test H1 and H2 with respect to workload, we implemented the NASA-TLX [29],
a well-established questionnaire [39]... | AI enhance sour performance |
As a robustness check and to assess convergent validity, we also measure LLM-
synthesized personality using the Big Five Inventory (BFI) [48]. Developed in the
lexical tradition, the BFI is a brief (44-item), adjectival statement-based measure of the
broad Big Five traits. The BFI asks participants to rate short descri... | PersonalityTraitsinLargeLanguageModels |
[16] Caroline Claus and Craig Boutilier. The dynamics of reinforcement learning in cooperative multiagent
systems. In AAAI/IAAI, 1998.
[17] Allan Dafoe, Yoram Bachrach, Gillian Hadfield, Eric Horvitz, Kate Larson, and Thore Graepel. Cooperative
ai: machines must learn to find common ground. Nature, 593(7857):33–36, 20... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
4 Speech Representation Learning
The process of speech representation learning is essential for extracting pertinent and practical
characteristics from speech signals, which can be utilized for various downstream tasks such as
speaker identification, speech recognition, and emotion recognition. While traditional method... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Sellam, T., Yadlowsky, S., Wei, J., Saphra, N., D’Amour, A.,
Linzen, T., Bastings, J., Turc, I., Eisenstein, J., Das, D.,
et al. The multiberts: Bert reproductions for robustness
analysis. arXiv preprint arXiv:2106.16163, 2021.
Sharma, U. and Kaplan, J. A neural scaling law from the
dimension of the data manifold. Com... | Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling |
S., Le, T., Oyebade, T., Le, T., Yang, Y., Nguyen, Z.,
Kashyap, A. R., Palasciano, A., Callahan, A., Shukla, A.,
Miranda-Escalada, A., Singh, A., Beilharz, B., Wang, B.,
Brito, C., Zhou, C., Jain, C., Xu, C., Fourrier, C., Peri˜n´an,
D. L., Molano, D., Yu, D., Manjavacas, E., Barth, F.,
Fuhrimann, F., Altay, G., Bayrak... | Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling |
[93] Philippe Laban, Tobias Schnabel, Paul N Bennett, and Marti A Hearst. 2022. SummaC: Re-Visiting NLI-based Models
for Inconsistency Detection in Summarization. Transactions of the Association for Computational Linguistics 10 (2022),
163–177.
[94] Rémi Lebret, David Grangier, and Michael Auli. 2016. Neural Text Gene... | SurveyofHallucinationinNatural Language Generation |
Consistency for RF regression has been established for several variants of the algorithm, occasionally under some constraints
on the data generating process (Genuer, 2012; Scornet et al., 2015; Wager and Walther, 2015; Biau and Scornet, 2016).
Recent work in this area has tended to focus on asymptotic normality (Mentch... | Adversarial Random Forests for Density Estimation and Generative Modeling |
For Dreambooth [32], we use the diffusers implemen-
tation and tune only the denoiser’s U-Net with no prior
preservation loss. We perform 500 fine-tuning steps using a
learning rate of 5e-6 and batch size of 4. Finally, we com-
pare to CustomDiffusion [14] using their six released mod-
els available in their official i... | A Neural Space-Time Representation for Text-to-Image Personalization |
and ML community (i.e., Hugging Face), which can process inputs from different modalities and
solve numerous complex AI tasks. More specifically, for each AI model in Hugging Face, we use
its corresponding model description from the library and fuse it into the prompt to establish the
connection with ChatGPT. Afterward,... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
4.1 Experimental Settings
All results in this paper is achieved with CFG-scale at 9.0. The sampler is DDIM. We use 20 steps by
default. We use three types of prompts to test the models:
(1) No prompt: We use empty string “” as prompt.
(2) Default prompt: Since Stable diffusion is essentially trained with prompts, the ... | Adding Conditional Control to Text-to-Image Diffusion Models |
[75] H. Jwa, D. Oh, K. Park, J. Kang, and H. Lim, ‘‘ExBAKE: Automatic
fake news detection model based on bidirectional encoder representations
from transformers (BERT),’’ Appl. Sci., vol. 9, no. 19, p. 4062, Sep. 2019.
[76] T. Mikolov, I. Sutskever, K. Chen, G. S. Corrado, and J. Dean, ‘‘Dis-
tributed representations o... | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
Topic #16
people
like
know
think
life
study
cells
patients
cancer
figure
said
time
new
people
like
v
granada
club
m
cent
let
model
data
set
function
var
assert
text
label
check
court
states
district
united
trial
android
new
public
import
void
invention
layer
substituted
et
group
binding
protein
receptor
dna
beta
tr
said... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
providelogicalconstraintsandrules.Asimilarapproachwasdeveloped
byChangetal.[82],withanemphasisonmanufacturingdesign.
Themethodologieswementionedaboveprovideusvaluableguid-
ance when developing our domain-specific ontology, aiming to inte-
grate domain-specific knowledge regarding demand forecasting with
the demand fore... | Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio |
gation. In International Conference on Learning Representations, 2017.
Rico Sennrich, Barry Haddow, and Alexandra Birch. Neural Machine Translation of Rare Words with
Subword Units. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics
(Volume 1: Long Papers), 2016.
Chris Shallue,... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
-
-
49.9
60.4
72.2
75.9
Table 17: 4-shot evaluation on XCOPA.
PaLM
PaLM CoT PaLM 2
Shi et al.
4-shot
77.4
78.0
92.6
96.0
61.0
69.4
85.4
87.2
92.8
89.8
91.6
83.7
Ours
4-shot
75.6
77.2
92.2
95.8
60.6
68.8
84.0
86.8
92.4
89.4
90.6
83.0
Shi et al.
4-shot
91.0
89.6
94.0
97.4
66.8
85.4
90.8
90.2
94.6
94.6
94.8
89.9
O... | PaLM 2 Technical Report |
12
Figure 5: Effects of model scaling and fine-tuning on six foundation metrics. Results are shown for 2B, 8B, and
137B parameters pre-trained (PT) models, and the two levels of fine-tuning (FT) with the bottom-most the one we call
LaMDA. Results are compared with crowdworker quality having access to information retrie... | LaMDA- Language Models for Dialog Applications |
Figure 1: An overview of our instruction backtranslation method. We start from a base language
model, e.g. LLaMa, a small amount of seed examples of (instruction, output) pairs, and a collection
of unlabelled documents which are considered candidate outputs for unknown instructions. Self-
augmentation: the base model i... | Self-AlignmentwithInstructionBacktranslation |
[49] M. Helmert, P. Haslum, J. Hoffmann, R. Nissim, Merge-and-shrink abstraction: a method for generating lower bounds in factored state spaces, J. ACM
[50] M. Heusner, M. Wehrle, F. Pommerening, M. Helmert, Under-approximation refinement for classical planning, in: Proceedings of the 24th International
[51] J. Hoffma... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
are prone to noise in the datasets. Adding a little amount
of noise to the graph via node perturbation or edge deletion
and addition has an antagonistic effect on the GNN output.
Graph convolutional network (GCN) is considered as one of
the basic graph neural networks variants. | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
Example 4.2: Case Study in Test Set of GSM8K
Question: Darrell and Allen’s ages are in the ratio of 7:11, If their total age now is 162, calculate Allen’s age
10 years from now. (Ground-truth answer is 109)
SFT Answer: The total ratio representing their ages is 7+11=<<7+11=18>>18. The fraction representing
Darrell’s ag... | METAMATH |
tial role in our forecast explanations, we were interested in assessing
(A) if media events describe features context and (B) if retrieved
external datasets were related to the features for which we queried
them.Regardingtheexplanations,weassessedmetricsrelatedtolisted
externaldatasets,events,andkeywordstounderstandift... | Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio |
As shown in Figure 2, GPT4Video’s architecture is
composed of three integral components: a video under-
standing module that employs a video feature extractor
and a video abstractor to encode and align video infor-
mation with the LLM’s word embedding space; the LLM
body, utilizing the structure of LLaMA and employing
... | GPT4Video |
The agency problem. The main challenge in the principal-agent model (with one or more
principals) is the agency problem [32]: Actions that generate good outcomes for the principal(s)
are costly for the agent, who rationally acts in his own self-interest. To align the interests of the
principals with those of the age... | Incomplete Information VCG Contracts for Common Agency |
the likelihood that any given person may have been exposed to it (see, e.g.,
Siegel, Chapter 4, this volume). The more important questions, such as the
relative prevalence of a phenomenon, trends over time, or assessments of causal
relationships, however, require much more complex research designs, sustained
research e... | Social_Media_and_Democracy |
a
m
p
i
o
n
e
d
a
n
d
i
m
p
l
e
m
e
n
t
e
d
a
n
i
n
i
t
i
a
l
p
r
o
o
f
o
f
c
o
n
c
e
p
t
f
o
r
r
e
v
i
s
i
o
n
s
.
L
e
o
i
m
p
l
e
m
e
n
t
e
d
a
s
m
a
l
l
s
c
a
l
e
d
e
r
i
s
k
i
n
g
e
x
p
e
r
i
m
e
n
t
.
S
t
e
v
e
n
h
e
l
p
e
d
d
e
s
i
g
n
t
h
e
f
i
n
a
l
r
e
v
i
s
i
o
n
p
... | Language models can explain neurons in language models |
8. Broader impact
Good code generation models have the potential to have a positive, transformative impact on society,
with a wide range of applications including computer science education, developer tooling, and
making programming more accessible. However, like most technologies, these models might enable
application... | alphacode |
Frontier AI also introduces security vulnerabilities when it is integrated into broader systems.
These new digital vulnerabilities – for example corrupting training data (‘data poisoning’),
hijacking model output (‘prompt injection’), and extracting sensitive training data (‘model
inversion’) – will require new and ... | Capabilities and risks from frontier AI |
Figure 6. Zero-shot evaluations of Pythia models over training, as well as their intervened counterparts, on the LAMBADA dataset.
C.2. Pretraining Term Frequency
checkpoint
13000
39000
65000
91000
117000
143000
160 M
1.0 B
2.8 B
12 B
∆k=0 ∆k=4 ∆k=16 ∆k=0 ∆k=4 ∆k=16 ∆k=0 ∆k=4 ∆k=16 ∆k=0 ∆k=4 ∆k=16
2.8
11.6
10.2
... | Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling |
Jing Xu, Da Ju, Margaret Li, Y-Lan Boureau, Jason Weston, and Emily Dinan. Recipes for safety in open-domain
Carole-Jean Wu, Ramya Raghavendra, Udit Gupta, Bilge Acun, Newsha Ardalani, Kiwan Maeng, Gloria
Chang, Fiona Aga, Jinshi Huang, Charles Bai, et al. Sustainable ai: Environmental implications, challenges
and opp... | Llama2 |
privileged environment information [Huang et al., 2022a,
Zhu et al., 2023], JARVIS-1 has the ability to speculate
about the reasons why current goals cannot be achieved,
without the need for additional information or design. | JARVIS-1 |
Study of the projector. The projector network, first introduction by Chen et al. [2020b],
maps the representations into another space where the loss is computed. Despite strong
empricial evidence the projector improves performance, few theoretical works attempted
to explain its role. Jing et al. [2022] study the role of... | A Cookbook of Self-Supervised Learning |
In contrast, expert specialization is far less noticeable in the decoder. Not only are sentinel to-
kens routed somewhat uniformly across decoder experts (see Table 14), but we also do not observe
meaningful specialization (semantics or syntax) in decoder experts.
We hypothesize that this lack of meaningful expert spec... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
4
Sampling prop. Epochs Disk size
Dataset
Code Llama (500B tokens)
2.03
Code
1.39
Natural language related to code
0.01
Natural language
Code Llama - Python (additional 100B tokens)
3.69
Python
0.05
Code
Natural language related to code
0.35
0.00
Natural language
75%
10%
10%
5%
85%
8%
7%
859 GB
78 GB
3.5TB
79GB
... | CodeLlama2 |
Many other methods belong to this self-distillation family. MoCo is another popular
method based on building a dictionary look-up that was shown to in some cases to
surpass supervised learning on segmentation and object detection benchmarks He et al.
[2020a]. Originally the momentum encoder was introduced as a substitu... | A Cookbook of Self-Supervised Learning |
*Equal contribution 1OpenAI, San Francisco, CA 94110, USA.
Correspondence to: Alec Radford <alec@openai.com>, Jong
Wook Kim <jongwook@openai.com>.
1Baevski et al. (2021) is an exciting exception - having devel-
oped a fully unsupervised speech recognition system | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
randomly selected positive step (or neutral if their were no positive ones).
This often resulted in trajectories that got stuck in endless sequences of neutral
steps that said reasonable things but made frustratingly slow progress towards a
solution or negative steps that needed constant human supervision. In phase 2,
... | Let’s Verify Step by Step |
Empirical Methods in Natural Language Processing,
pages 2381–2391, Brussels, Belgium. Association
for Computational Linguistics.
Seyed-Iman Mirzadeh, Mehrdad Farajtabar, Ang Li,
and Hassan
Nir Levine, Akihiro Matsukawa,
Improved knowledge distil-
Ghasemzadeh. 2020.
In The Thirty-Fourth
lation via teacher assistant.
AA... | LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions |
# Write Python function to complete the task and pass the assertion tests.
### Task Start ###
# These are the assertions for your function:
assert similar_elements((3, 4, 5, 6),(5, 7, 4, 10)) == (4, 5)
""" Write a function to find the similar elements from the given two tuple
lists. """
def similar_elements(test_tup1... | Teaching Large Language Models to Self-Debug |
i , KW K
, V W V
i
In summary, the Transformer architecture consists of these key components, each
playing a crucial role in enabling the model to understand and generate sequences
effectively, making it a powerful tool
language processing and other
sequence-to-sequence tasks.
in natural
2.1.2 Large Language Models... | Beyond Efficiency |
CoNLL 2002 (Tjong Kim Sang, 2002) Standard benchmark containing news articles. For Span-
ish the corpus contains NewsWire articles from May 2000. The Dutch data come from the Belgian
newspaper De Morgen from the months of June to September 2000. We included this dataset as it
is one of the longest standing standard ben... | MULTI HASH EMBEDDINGS IN SPACY |
Fintechs can best combine the two by using predictive AI for “back of office”
workflows and using generative AI for client-facing “front of office” product
experiences. We’re especially interested in finding tools that give users greater
control and input over the data and parameters being used to give them a result.
T... | Fintech x AI_ The Lightspeed View _ by Lightspeed _ Lightspeed Venture Partners _ Jun, 2023 _ Medium |
Pretraining Task Design in LLMs Because language
models can easily incorporate multiple training tasks, task
selection is an important area of research. GPT-3 [8] and
PaLM [16] show that training LLMs on diverse tasks leads
to positive scaling effects on zero- and few-shot tasks. Other
works show that masking approache... | VideoPoet |
8.2 Memory efficiency
Memory efficiency is critical, particularly for deployment on resource-constrained
devices. Techniques like pruning and quantization explicitly target memory efficiency
29
3
0
Main Category Technique
Sub-Category
Computation Memory
Energy Money Communication
Transformer Architecture §3.1
Non-t... | Beyond Efficiency |
R. Shwartz-Ziv, R. Balestriero, and Y. LeCun. What do we maximize in self-supervised
learning? arXiv preprint arXiv:2207.10081, 2022. 4
A. Singh, R. Hu, V. Goswami, G. Couairon, W. Galuba, M. Rohrbach, and D. Kiela. FLAVA:
A Foundational Language and Vision Alignment Model. In Proceedings of the IEEE/CVF
Conference o... | A Cookbook of Self-Supervised Learning |
2
Large Language Models Cannot Self-Correct Reasoning Yet
term this setting intrinsic self-correction. For brevity, unless explicitly stated otherwise (e.g., self-
correction with oracle feedback), all references to “self-correction” in the remainder of this paper
pertain to intrinsic self-correction.
3 CAN LARGE L... | LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET |
Indeed, tools like the LA Times’s Quake Bot can automatically generate and post
stories (Walker 2014). Harvard University’s Nieman Journalism Lab has argued
that there will be a large-scale shift toward the “botification” of the news in
coming years (Barot 2016). Others, with more unease about automated
reporting, have ... | Social_Media_and_Democracy |
Length
Accuracy (%)
1-29
98-100
30
88
31
79
32
52
33
26
34
2
35+
0
Table 1: Accuracy (out of 100 examples) of the final checkpoint of the 582M model after training.
For example, this table shows that the post-training 582M model can add 30 digit numbers with 88%
accuracy without using any chain-of-thought reas... | CHAIN-OF-THOUGHTREASONING IS APOLICY IMPROVEMENTOPERATOR |
Kingma, D., Salimans, T., Poole, B., and Ho, J. (2021). Vari-
ational diffusion models. In Advances in Neural Informa-
tion Processing Systems, volume 34, pages 21696–21707.
Kingma, D. and Welling, M. (2013). Auto-encoding varia-
tional Bayes. In International Conference on Learning
Representations.
Kisa, D., Van den ... | Adversarial Random Forests for Density Estimation and Generative Modeling |
Feature extraction is typically performed after pre-processing to convert the audio signal into
meaningful and informative features while reducing their number. MFCCs and spectrograms are
popular feature extraction choices in speech-based systems [288]. These features are then given to
the DRL agent to perform various ... | AReviewofDeepLearningTechniquesforSpeechProcessing |
This PhD project will investigate the automated production of models from natural language
requirements statements, using rule-based or neural net approaches to identify model elements such as
classes and operations from the statements.
The project should involve a comparison of the relative effectiveness of rule-... | informatics-phd-projects-2022-23 |
specified, and the output begins. For timestamp predic-
tion, we predict time relative to the current audio segment,
quantizing all times to the nearest 20 milliseconds which
matches the native time resolution of Whisper models, and
add additional tokens to our vocabulary for each of these.
We interleave their predictio... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
Fake news is big news. From the diffusion of rumors and conspiracies in the
United States to the spread of disinformation by Russian troll
farms,
misinformation is a hot topic among academics and journalists alike. How
can we understand and correct such misinformation? A logical starting point
is to fight fiction with fa... | Social_Media_and_Democracy |
machine,’’ Ann. Statist., vol. 29, no. 5, pp. 1189–1232, Oct. 2001.
[63] I. Mandel.
(2018). Aesthetic, Art-Historical and Economic Values
[Online]. Available: https://ssrn.com/
in Painting: Empirical Study.
abstract=3160419
E. Cetinic et al.: Deep Learning Perspective on Beauty, Sentiment and Remembrance of Art
EV... | A_Deep_Learning_Perspective_on_Beauty_Sentiment_and_Remembrance_of_Art |
10
2.5 Literature Review of Tool Learning
Figure 3: Tool-augmented learning seeks to augment foundation models with the execution results from tools
(i.e., tool for AI); while tool-oriented learning focuses on utilizing models to govern tools and make sequential
decisions in place of humans (i.e., AI for tool). Our ... | Tool Learning with Foundation Models |
data (Zhou et al., 2023a; Chen et al., 2023b) v.s.
data scaling (Wei et al., 2021; Sanh et al., 2022),
task composition (Wang et al., 2022; Chung et al.,
2022) v.s. expert models (Jang et al., 2023; Wang
et al., 2023b), etc. Hence, more fine-grained under-
standing is required to solve these conflicts.
General Data Man... | DataManagementForLargeLanguageModels-ASurvey |
. 24
32
48
64
24
32
48
64
d
i
r
b
y
H
d
e
t
c
e
j
o
r
p
n
a
e
M
)
m
m
(
r
o
r
r
e
t
n
i
o
j
81.6
82.2
81.8
82.3
86.0
82.7
82.7
84.1
30
25
20
15
10
5
0
32 64
128
256
512 555
# Latent keypoints
Figure 5:
Intrinsic dimension analysis of our pseudo–
ground truth with 555 joints. The residual error curve sh... | Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats |
havethesamesizeastheoutputvideoˆxK0.Unlessotherwisespecified,weapplyT=1000stepDDPMtosample40-frame32×32×2ˆfand32×32×1ˆmandfinallyproduce40-framevideosˆxwith128×128frameresolution.BaselineImplementation.WecompareLFDMwiththreebaselinemodels,includingGAN-basedI2VmodelImaGINator[77],videodiffusionmodelsVDM[27],andavariantofi... | Conditional Image-to-Video Generation with Latent Flow Diffusion Models |
The evaluation process we ran helped to generate additional qualitative evidence of societal biases
in various versions of the GPT-4 model. We found that the model has the potential to reinforce and
reproduce specific biases and worldviews, including harmful stereotypical and demeaning associations
for certain marginali... | gpt-4-system-card |
2. Successful applicants to selected postgraduate programmes are required to pay a tuition fee
deposit. Following enrolment, any tuition fee deposit payment will be counted towards the tuition
fees payable for the programme of study. Payment of the deposit allows applicants to
demonstrate their commitment to atten... | UCL Academic Manual |
Transparency was thus advocated as a core element of public accountability
for business (Waddock 2004) and enthusiastically announced as a way for firms
to demonstrate their social responsibility and integrity (Tapscott and Ticoll
2003). In practice, however, transparency has many important limitations.
First, just like... | Social_Media_and_Democracy |
5.2 Performance Expectations and Judgments of Performance
Subjective overall performance. To analyze expected overall performance, we centered the
5.2.1
values by subtracting four points of the scale so that 0 indicates not favoring any condition
and modeled overall performance estimates as a function of Time and Descr... | AI enhance sour performance |
quality, could match the performance of large-scale models trained on vast datasets without sufficient quality constraints.
This shift in understanding the role of data quality could revolutionize the approach to training and optimizing LLMs. | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
4. Experiments
We evaluate the proposed Self-Extend primarily using the
Llama-2 (Touvron et al., 2023) families considering its wide
adoption and popularity. We also construct some experi-
ments for other RoPE based models including the currently
popular model Mistral (Jiang et al., 2023) and SOLAR (Kim
et al., 2023), ... | Self-Extend LLM |
To address this issue, we propose associable instruction
tuning which trains LoRA [72] for LLMs on the Oogiri-
GO dataset to achieve certain associable generation and dis-
crimination abilities. It has two steps, including instruction
generation and discrimination template design, and associa-
ble instruction learning.... | Let’sThinkOutsidetheBox |
4.4. Metrics
We compute different metrics to evaluate MusicLM, captur-
ing two important aspects of music generation: the audio
quality and the adherence to the text description.
Fr´echet Audio Distance (FAD). The Fr´echet Audio Dis-
tance (Kilgour et al., 2019) is a reference-free audio quality
metric, which correla... | MusicLM |
Methods for memory retrieval. When an agent interacts with its environment or users, it is
imperative to retrieve the most appropriate content from its memory. This ensures that the agent
accesses relevant and accurate information to execute specific actions. An important question arises:
How can an agent select the mo... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
113.7
125.7
144.2
103.7
106.2
110.5
R
u
s
s
i
a
n
31.1
20.5
11.4
7.2
6.4
5.6
T
u
r
k
i
s
h
42.5
27.5
15.9
10.4
9.4
8.4
F
r
e
n
c
h
41.4
28.5
15.0
8.7
7.7
8.3
G
e
o
r
g
i
a
n
123.0
114.7
118.3
117.3
100.5
105.0
M
a
r
a
t
h
i
100.3
100.3
60.2
63.2
43.7
38.3
S
i
n
d
h
i
105.8
103.9
131.7
147.0
177.9
156.5
U... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
Answer the following question. The global cleaning services industry is expanding due to? Service
providers expanding their online presence and rising commercial consumer demand.
Given the sentence “Businesses frequently grow when corporate profits increase, raising demand
for janitorial services.”, come up with a dif... | ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION |
Classi | Dolly 2 Databricks |
) = ∞ for all s ∈ f (2) and s
(cid:10)
8.4. Relating metric and non-metric properties
Also the properties in Definition 9 have connections with the metric properties. We know that M↑R↑C↑ ⇒ PS↑ and
PS↑ ⇒ P1↑, so it follows from Theorem 49 that M↑R↑C↑AC↓ ⇒ A↓. However, we can also derive the following mo... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
4 Would GPT-2 do better if its input were broadened to include perceptual input rather than mere text?
Perhaps, but I don't think merely broadening the range of input would solve the system's fundamental
lack of articulated internal models. Meanwhile, it is interesting to note that, blind children develop rich
inter... | The Next Decade in AI- |
For general capabilities, Llama-2-chat-70B (Touvron et al., 2023b) shows improvement over GPT-
3.5-turbo in some benchmarks, but remains behind for most others. Zephir-7B (Tunstall et al.,
2023) approaches 70B LLMs as a result of distilled direct preference optimization. WizardLM-
70B (Xu et al., 2023a) and GodziLLa-70... | ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup |
‘‘‘
Human: Can you write some test cases for this function?
Assistant: Sure, here are some tests.
‘‘‘
assert alternating([10, 20, 30], [1, 2, 3]) == [10, 1, 20, 2, 30, 3]
assert alternating([True, False], [4, 5]) == [True, 4, False, 5]
assert alternating([], []) == []
‘‘‘
Human: Modify the function so that it retur... | StarCoder_paper (1) |
[84] Paweł W. Woźniak, Jakob Karolus, Florian Lang, Caroline Eckerth, Johannes Schöning, Yvonne Rogers, and Jasmin Niess. 2021. Creepy
Technology:What Is It and How Do You Measure It?. In Proceedings of the 2021 CHI Conference on Human Factors in Computing
Systems (Yokohama, Japan) (Chi ’21). Association for Computing ... | Society’sAttitudesTowardsHumanAugmentation |
We leverage the TEXT-DAVINCI-003 model4 provided by OpenAI5 for all experiments as it is the
latest text completion that has the highest text completion quality among all models we have access
to. We set the max token number to 2048 and temperature to 0, and use the top probability response.
As a result, the response r... | LLM+P- Empowering Large Language Models with Optimal Planning Proficiency |
a
t
t
a
c
k
o
n
s
u
p
e
r
p
o
s
i
t
i
o
n
.
T
h
e
h
i
g
h
-
l
e
v
e
l
i
d
e
a
i
s
t
o
o
p
t
i
m
i
z
e
a
l
i
n
e
a
r
c
o
m
b
i
n
a
t
i
o
n
o
f
n
e
u
r
o
n
s
.
G
i
v
e
n
a
v
e
c
t
o
r
o
f
a
c
t
i
v
a
t
i
o
n
s
a
n
d
u
n
i
t
v
e
c
t
o
r
(
t
h
e
d
i
r
e
c
t
i
o
n
)
,
w
e
d
e
... | Language models can explain neurons in language models |
Online Hate Speech
63
and ethnographic work suggests that users of 4chan and Redditt are
overwhelmingly young, white, and male (Daniels 2017; Costello and Hawdon
2018), because of the anonymous nature of these sites we do not know very
much about the users that produce the most hate speech. In particular, we do
not k... | Social_Media_and_Democracy |
6
Table 2. Video-to-text generation (video captioning) results on
MSR-VTT [47].
Table 4. Performance comparison of safety Evaluation. “VU” and
“VG” represent Video Understanding and Video Generation.
BLEU-4 (↑) METEOR (↑)
Task Method
Method
ORG-TRL [55]
GIT [42]
mPLUG-2 [46]
CoDi [38]
NExT-GPT [45]
GPT4Video (Our... | GPT4Video |
Quantization Adaption. QLoRA [49], a quantized variant
of LoRA, effectively addresses the limited computational
resource of LoRA for fine-tuning LLMs by quantizing the
transformer model to 4-bit NormalFloat (NF4) precision with
double quantization processing, and using a paged optimizer
to deal with memory spikes. NF4 ... | Parameter-EfficientFine-TuningMethods |
(weighted) K-means [32, 180, 309, 330]. | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
practices conducted throughout the project, priority in decision-making has always yielded to the
more responsible option even if this meant introducing limitations that might impact adoption or
future research. For example, the decision in the Legal, Ethics, Governance Working Group to remove
and not release a dataset... | StarCoder_paper (1) |
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
Online Political Advertising in the United States
131 | Social_Media_and_Democracy |
tation. IEEE transactions on pattern analysis and machine intelligence, 2018b.
Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language models
are unsupervised multitask learners.
Alec Radford, Rafal Jozefowicz, and Ilya Sutskever. Learning to generate reviews and discovering sent... | DINOv2- Learning Robust Visual Features without Supervision |
[193] Shen, Y., Shao, J., Zhang, X., Lin, Z., Pan, H., Li, D., Zhang, J., Letaief,
K.B.: Large language models empowered autonomous edge ai for connected
intelligence. arXiv preprint arXiv:2307.02779 (2023)
[194] Ge, T., Chen, S.-Q., Wei, F.: Edgeformer: A parameter-efficient transformer
for on-device seq2seq generation... | Beyond Efficiency |
No
Is there a process by which annotators can later choose to withdraw their data from the dataset? Please detail. No | PaLM 2 Technical Report |
4 Counterarguing typically involves generating arguments to dispute a correction. Inverting this
process, Chan et al. (2017) also find that corrections are generally less effective when people are
asked to record arguments in favor of the original misinformation.
https://doi.org/10.1017/9781108890960 Published online b... | Social_Media_and_Democracy |
4. Analysis and Ablations
4.1. Model Scaling
A large amount of the promise in weakly supervised train-
ing approaches is their potential to use datasets much larger
than those in traditional supervised learning. However, this
comes with the cost of using data that is possibly much
noisier and lower quality than gold-s... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
Figure 11. Comparison with AvatarMe++ [42] and AlbedoMM [61]. For a fair comparison, we sample the per-vertex albedo values of
AlbedoMM to a flat plane UV parameterization, and crop AvatarMe++ and Our results to the central facial area, since AlbedoMM is using
a tigher facial crop.
13
Diffuse AlbedoSpecular AlbedoText... | Relightify-Relightable3DFacesfromaSingleImageviaDiffusionModels |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.