text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
Empirical evidence underscores marked improvements in
text generation quality, factual accuracy, reduced toxicity,
and downstream task proficiency, especially in knowledge-
intensive applications like open-domain QA. These results
imply that integrating retrieval mechanisms into the pre-
training of autoregressive la... | RAG forLargeLanguageModels-ASurvey |
D DOMAIN-SPECIFIC TASKS EVALUATION
Prompting. In prompting evaluation, each task corresponds to multiple prompt templates and we
randomly sample one of them for each data example, to mitigate result variance caused by template
17
Table 9: Keywords that compile into regular expressions. These keywords are used in th... | ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION |
ERROR: type should be string, got "https://www.ai21.com/blog/introducing-j2\n\n7/12\n\nF\na\ni\nt\nh\nf\nu\nl\nn\ne\ns\ns\n \nr\na\nt\ne\ns\n \nm\ne\na\ns\nu\nr\ne\n \nh\no\nw\n \nf\na\nc\nt\nu\na\nl\nl\ny\n \nc\no\nn\ns\ni\ns\nt\ne\nn\nt\n \na\n \ns\nu\nm\nm\na\nr\ny\n \ni\ns\n \nw\ni\nt\nh\n \nt\nh\ne\n \no\nr\ni\ng\ni\nn\na\nl\n \nt\ne\nx\nt\n.\nA\ns\n \ny\no\nu\n \nc\na\nn\n \ns\ne\ne\n \nb\ne\nl\no\nw\n,\n \no\nu\nr\n \nn\ne\nw\n \nS\nu\nm\nm\na\nr\ni\nz\ne\n \nA\nP\nI\n \nh\na\ns\n \nr\ne\na\nc\nh\ne\nd\n \na\n \nf\na\ni\nt\nh\nf\nu\nl\nn\ne\ns\ns\n \nr\na\nt\ne\n \nt\nh\na\nt\no\nu\nt\np\ne\nr\nf\no\nr\nm\ns\n \nO\np\ne\nn\nA\nI\n’\ns\n \nD\na\nv\ni\nn\nc\ni\n-\n0\n0\n3\n \nb\ny\n \n1\n9\n%\n.\n \nA\nc\nc\ne\np\nt\na\nn\nc\ne\n \nr\na\nt\ne\ns\n \nm\ne\na\ns\nu\nr\ne\n \nh\no\nw\n \ns\na\nt\ni\ns\nf\ni\ne\nd\n \nh\nu\nm\na\nn\n \ne\nv\na\nl\nu\na\nt\no\nr\ns\n \na\nr\ne\n \nw\ni\nt\nh\n \nt\nh\ne\n \nq\nu\na\nl\ni\nt\ny\n \no\nf\ng\ne\nn\ne\nr\na\nt\ne\nd\n \ns\nu\nm\nm\na\nr\ni\ne\ns\n,\n \na\nn\nd\n \nw\ne\n’\nr\ne\n \np\nr\no\nu\nd\n \nt\no\n \ns\na\ny\n \nt\nh\na\nt\n \no\nu\nr\n \nS\nu\nm\nm\na\nr\ni\nz\ne\n \nA\nP\nI\n \nh\na\ns\n \na\nc\nh\ni\ne\nv\ne\nd\n \na\nn\na\nc\nc\ne\np\nt\na\nn\nc\ne\n \nr\na\nt\ne\n \nt\nh\na\nt\n \ni\ns\n \n1\n8\n%\n \nh\ni\ng\nh\ne\nr\n \nt\nh\na\nn\n \nt\nh\na\nt\n \no\nf\n \nO\np\ne\nn\nA\nI\n’\ns\n.\nS\nt\na\nA\nI\n2\n1\nS\nt\nu\nd\ni\no\nW\no\nr\nd\nt\nu\nn\ne\nW\no\nr\nd\nt\nu\nn\ne\nR\ne\na\nd\nC\no\nm\np\na\nn\ny\n\f30/04/2023, 12:17" | Announcing Jurassic-2 and Task-Specific APIs |
C.3 GOLD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
C.4 Additional results for filtering and clustering . . . . . . . . . . . . . . . . . . . . . . . 45
C.5 HumanEval comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
C.6 APPS dataset setting... | alphacode |
Recent work has highlighted that LLMs often capture social biases and stereotypes from their pre-
training corpora (Kurita et al., 2019; May et al., 2019; Hutchinson et al., 2020; Meade et al., 2023).
To quantify social bias within our model, we use StereoSet (Nadeem et al., 2021).
StereoSet consists of a collection of... | StarCoder_paper (1) |
Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha, Heng-Tze Cheng,
Alicia Jin, Taylor Bos, Leslie Baker, Yu Du, et al. Lamda: Language models for dialog applications. ArXiv
preprint, abs/2201.08239, 2022. URL https://arxiv.org/abs/2201.08239.
Emanuel Todorov, Tom Erez, and Yuval Tassa. ... | Tool Learning with Foundation Models |
[47] Latané Bullock, Hervé Bredin, and Leibny Paola Garcia-Perera. 2020. Overlap-aware diarization: Resegmentation
using neural end-to-end overlapped speech detection. In ICASSP 2020-2020 IEEE International Conference on Acoustics,
Speech and Signal Processing (ICASSP). IEEE, 7114–7118.
[48] Tanja Bunk, Daksh Varshney... | AReviewofDeepLearningTechniquesforSpeechProcessing |
3.2. Does Training Order Influence Memorization? | Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling |
H
u
n
g
a
r
i
a
n
0.1
0.2
4.8
15.5
18.3
21.2
r
A
m
e
n
i
a
n
0.1
0.1
0.7
10.4
13.2
16.0
I
n
d
o
n
e
s
i
a
n
0.3
2.6
16.4
24.1
27.2
29.1
I
c
e
l
a
n
d
i
c
0.4
0.4
1.8
6.8
6.6
9.1
L
i
t
h
u
a
n
i
a
n
0.1
0.3
1.9
9.6
12.0
14.0
L
a
t
v
i
a
n
0.2
0.4
1.5
10.0
12.5
14.3
L
a
o
0.2
0.3
2.0
8.1
9.3
11.0
M
a
o
r
... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
of sparse/dense keypoint detection (although they can be
used as additional constraints).
We visualize the optimization process in Fig.5. In this
figure, the initial pose of the right arm is incorrect, which
leads to reconstruction artifacts. However, as the optimiza-
tion is carried out, the right arm gradually moves ... | PaMIR- Parametric Model-Conditioned Implicit Representation for Image-based Human Reconstruction |
21/09/2023, 08:13
The Casino on Mars
About
Team
Portfolio
Writing
Opportunities
Contact
Open Source
The Casino on Mars
Sep 20, 2023 | Matt Huang
It’s useful to think of crypto as a new planet that’s being settled.
Skeptics see a desolate planet without purpose. Or worse, a haven for an
unsavory casino.
Opt... | The Casino on Mars |
r
s
.
T
o
p
u
t
H
u
g
g
i
n
g
G
P
T
i
n
t
o
r
e
a
l
w
o
r
l
d
u
s
a
g
e
,
a
c
o
u
p
l
e
c
h
a
l
l
e
n
g
e
s
n
e
e
d
t
o
s
o
l
v
e
:
(
1
)
E
f
f
i
c
i
e
n
c
y
i
m
p
r
o
v
e
m
e
n
t
i
s
n
e
e
d
e
d
a
s
b
o
t
h
L
L
M
i
n
f
e
r
e
n
c
e
r
o
u
n
d
s
a
n
d
i
n
t
e
r
a
c
t
i
o
n
s
... | LLM Powered Autonomous Agents _ Lil'Log |
theory is that it may be difficult to determine what virtues are truly universally important, and
it does not provide concrete guidelines for decision making in specific situations.
The strengths of deontological ethics are that it provides clear and concrete guidelines for
decision making, making it a more structured an... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
[613] Sharif, M., S. Bhagavatula, L. Bauer, et al. Accessorize to a crime: Real and stealthy attacks
on state-of-the-art face recognition. In E. R. Weippl, S. Katzenbeisser, C. Kruegel, A. C.
Myers, S. Halevi, eds., Proceedings of the 2016 ACM SIGSAC Conference on Computer and
Communications Security, Vienna, Austria, ... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
As Large Language Models (LLMs) continue
to advance in their ability to write human-like
text, a key challenge remains around their ten-
dency to “hallucinate” – generating content
that appears factual but is ungrounded. This
issue of hallucination is arguably the biggest
hindrance to safely deploying these powerful
LL... | AComprehensiveSurveyofHallucinationMitigationTechniquesinLarge LanguageModels |
Carlini, N., Tramer, F., Wallace, E., Jagielski, M., Herbert-
Voss, A., Lee, K., Roberts, A., Brown, T., Song, D.,
Erlingsson, U., Oprea, A., and Raffel, C. Extracting
training data from large language models, 2020. URL
https://arxiv.org/abs/2012.07805.
Carlini, N., Ippolito, D., Jagielski, M., Lee, K., Tramer, F.,
an... | MusicLM |
[OpenAI, 2023] OpenAI. Gpt-4 technical report. https://cdn.
openai.com/papers/gpt-4.pdf, 2023.
[Packer et al., 2023] Charles Packer, Vivian Fang, Shishir G
Patil, Kevin Lin, Sarah Wooders, and Joseph E Gonza-
lez. Memgpt: Towards llms as operating systems. arXiv
preprint arXiv:2310.08560, 2023.
[Raffel et al., 2020]... | RAG forLargeLanguageModels-ASurvey |
Hershey, S., Chaudhuri, S., Ellis, D. P. W., Gemmeke, J. F.,
Jansen, A., Moore, C., Plakal, M., Platt, D., Saurous,
R. A., Seybold, B., Slaney, M., Weiss, R., and Wilson,
K. Cnn architectures for large-scale audio classification.
In International Conference on Acoustics, Speech and
Signal Processing (ICASSP), 2017.
Ho,... | MusicLM |
a
n
g
u
a
g
e
Z
h
o
n
g
,
R
.
,
S
n
e
l
l
,
C
.
,
K
l
e
i
n
,
D
.
a
n
d
S
t
e
i
n
h
a
r
d
t
,
J
.
,
2
0
2
2
.
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
,
p
p
.
2
7
0
9
9
-
-
2
7
1
1
6
.
| Language models can explain neurons in language models |
Eight Things to Know about Large Language Models | Eight Things to Know about Large Language Models |
6 Discussion
6.1 Mutual Benefits between LLM Research and Agent Research
With the recent advancement of LLMs, research at the intersection of LLMs and agents has rapidly
progressed, fueling the development of both fields. Here, we look forward to some of the benefits
and development opportunities that LLM research an... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
[21] Steven J Gortler, Radek Grzeszczuk, Richard Szeliski, and
Michael F Cohen. The lumigraph. In Proceedings of the 23rd
annual conference on Computer graphics and interactive
techniques, pages 43–54, 1996.
[22] Kaiwen Guo, Peter Lincoln, Philip Davidson, Jay Busch,
Xueming Yu, Matt Whalen, Geoff Harvey, Sergio Orts-... | DynIBaR-NeuralDynamicImage-BasedRendering |
sha1_base64="xnbcb3NcIJiA4aP+15D21QhxdTI=">AAAB+XicbVDLSsNAFJ3UV62vqEs3g0VwVRIRdFlw47KCfUgbw2Q6aYdOJmHmplhC/sSNC0Xc+ifu/BsnbRbaemDgcM693DMnSATX4DjfVmVtfWNzq7pd29nd2z+wD486Ok4VZW0ai1j1AqKZ4JK1gYNgvUQxEgWCdYPJTeF3p0xpHst7mCXMi8hI8pBTAkbybXsQERgHYfaUP2bgu7lv152GMwdeJW5J6qhEy7e/BsOYphGTQAXRuu86CXgZUcCpYHltkGqWEDohI9Y3VJKIa... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
system that augments a black-box LLM with a set
of Plug-And-Play (PnP) (Li et al., 2023b) modules.
The system makes the LLM generate responses
grounded in external knowledge. It also iteratively
revises LLM prompts to improve model responses
using feedback generated by utility functions. In
this paper, the authors pres... | AComprehensiveSurveyofHallucinationMitigationTechniquesinLarge LanguageModels |
Model-level safety reduces the burden on other safety-relevant infrastructure such as monitoring
or integration of classifiers in the product. However, model-level refusals and behavior changes can
impact all uses of the model, and often what is undesired or safe can depend on the context of model
usage (e.g., Typing “I... | gpt-4-system-card |
Voicebox: Text-Guided Multilingual
Universal Speech Generation at Scale
Matthew Le∗ Apoorv Vyas∗ Bowen Shi∗ Brian Karrer∗ Leda Sari Rashel Moritz
Mary Williamson Vimal Manohar Yossi Adi† Jay Mahadeokar Wei-Ning Hsu∗
Meta AI
Abstract | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
∗Equal contribution
†OpenAI
‡Microsoft | Improving Image Generation with Better Captions |
In the reviewed studies, KGs were mostly used for reasoning and inference in post-model XAI, with Deep Learning,
CNN, and LSTM being the most commonly used machine learning algorithms. For example, the authors of Sun et al.
(36) used a Deep Learning model in conjunction with reasoning via a medical KG to assess the cli... | Knowledge-graph-based explainable AI- A systematic review |
Defferrard, M., Benzi, K., Vandergheynst, P., and Bresson,
X. FMA: A dataset for music analysis. In International
Society for Music Information Retrieval Conference (IS-
MIR), 2017.
Devlin, J., Chang, M., Lee, K., and Toutanova, K. BERT:
pre-training of deep bidirectional transformers for lan-
guage understanding. In ... | MusicLM |
∗Equal contribution
Preprint. Work in progress. | MiniGPT-4- Enhancing Vision-Language Understanding with Advanced Large Language Models |
That is, practical PS-alignment failures involve highly-capable, strategically-aware agents applying
their capabilities (including, perhaps, the ability to copy themselves) to gaining and maintaining
power in the world—and they may become more and more difficult to stop as their power grows. In
dealing with systems that... | Is Power-Seeking AI an Existential Risk? |
Oracle
FORGE
CTGAN
CTAB-GAN+
IT-GAN
RCC-GAN
TVAE
Oracle
FORGE
CTGAN
CTAB-GAN+
IT-GAN
RCC-GAN
TVAE
Oracle
FORGE
CTGAN
CTAB-GAN+
IT-GAN
RCC-GAN
TVAE
Oracle
FORGE
CTGAN
CTAB-GAN+
IT-GAN
RCC-GAN
TVAE
Oracle
FORGE
CTGAN
CTAB-GAN+
IT-GAN
RCC-GAN
TVAE
Accuracy ± SE
0.828 ± 0.006
0.819 ± 0.006
0.786 ± 0.020
0.808 ± 0.008
... | Adversarial Random Forests for Density Estimation and Generative Modeling |
As we scale up our models progressively, the num-
ber of tasks they can solve (i.e., perform above the
random baseline on), increases. We aim to draw a
comparison between the tasks that the smallest non-
instruction tuned model within our experimental
range, which possesses effective in-context learn-
ing capabilities ... | AreEmergentAbilitiesinLarge Language Models just In-Context |
5. 3D Shape Regression from an Image
We present SHAPY, a network that predicts SMPL-X
parameters from an RGB image with more accurate body
shape than existing methods. To improve the realism and
accuracy of shape, we explore training losses based on
all shape representations discussed above, i.e., SMPL-X
meshes (Sec. ... | Accurate 3D Body Shape Regression using Metric and Semantic Attributes |
Conditional Image Generation with CLIP Latents. In arXiv, 2022.
Marc’Aurelio Ranzato, Arthur D. Szlam, Joan Bruna, Micha¨el Mathieu, Ronan Collobert, and Sumit
Chopra. Video (language) modeling: a baseline for generative models of natural videos. ArXiv,
abs/1412.6604, 2014.
Robin Rombach, Andreas Blattmann, Dominik L... | IMAGEN VIDEO- HIGH DEFINITION VIDEO GENERATION WITH DIFFUSION MODELS |
[54] S. Saito, T. Simon, J. Saragih, and H. Joo, “Pifuhd: Multi-level
pixel-aligned implicit function for high-resolution 3d human dig-
itization,” in IEEE/CVF Conference on Computer Vision and Pattern
Recognition (CVPR), June 2020.
[55] Z. Huang, Y. Xu, C. Lassner, H. Li, and T. Tung, “Arch: Animatable
reconstruction... | PaMIR- Parametric Model-Conditioned Implicit Representation for Image-based Human Reconstruction |
image better fits the text description?). Five responses are
collected for each comparison with majority vote (3+) being
considered the collective decision. | DiffusionModelAlignmentUsing Direct Preference Optimization |
A
p
p
l
i
c
a
t
i
o
n
o
f
o
u
r
A
I
P
r
i
n
c
i
p
l
e
s
U
n
d
e
r
p
i
n
n
i
n
g
a
l
l
o
u
r
w
o
r
k
o
n
B
a
r
d
i
s
a
f
o
c
u
s
o
n
r
e
s
p
o
n
s
i
b
i
l
i
t
y
a
n
d
s
a
f
e
t
y
.
O
u
r
d
e
v
e
l
o
p
m
e
n
t
o
f
B
a
r
d
i
s
g
u
i
d
e
d
b
y
—
c
h
i
e
f
a
m
o
n
g
t... | An overview of Bard- an early experiment with generative AI |
Polit. Sci. 36, 579–616 (1992).
bulletin 103, 299 (1988).
1944).
53. Schwitzgebel, E. Belief. In The Routledge Companion to Epistemology, 40–50 (Routledge, 2011).
54. Gauthier, J. & Levy, R. Linking artificial and human neural representations of language. In Proceedings of the 2019
Conference on Empirical Methods in ... | Language models trained on media diets can predict public opinion |
We note that there is currently no way to assess
the relationship between a prediction and a free-
text rationale within the same fully differentiable
model. Jacovi and Goldberg (2020) argue for the
development of evaluations that measure the extent
and likelihood that a rationale is faithful in practice
(illustrated i... | Measuring Association Between Labels and Free-Text Rationales |
symbols must in some way ground in our perceptual experience, and if we are to
interpret scenes in terms of symbols, we must have ways of inferring symbols (and
structured relationships among symbols) from input. Building adequate models will
also require systems that can infer temporal boundaries and temporal rela... | The Next Decade in AI- |
We experiment on Female 2 from MakeHuman.
3DMM tracking:. We add uniformly distributed noise to the fitted 3DMM global, neck and jaw poses, with a noise range
from 0.025( 1.4◦) to 0.1( 5.7◦).
Foreground mask:. We randomly select a 61 × 61 square and set the mask value to True or False randomly. We degrade
10%, 50% and ... | I M Avatar- Implicit Morphable Head Avatars from Videos |
J. ACM, Vol. 37, No. 4, Article 111. Publication date: August 2018.
A Survey on Evaluation of Large Language Models
111:23
4 WHERE TO EVALUATE: DATASETS AND BENCHMARKS
LLMs evaluation datasets are used to test and compare the performance of different language models
on various tasks, as depicted in Sec. 3. These da... | ASurveyonEvaluationofLargeLanguageModels |
50
Gemini: A Family of Highly Capable Multimodal Models
9.3.5. Geometrical reasoning
Prompt
Find the height of the parallelogram given its area with 100 square units.
Model Response
The area of the parallelogram is equal to the product of the base and the height. Hence
100 = (𝑥 + 15)𝑥. We get 𝑥2 + 15𝑥 − 100 = ... | gemini_1_report |
Gradually, Sam’s candidacy becomes the talk of the town, with
some supporting him and others remaining undecided.
3.4.2 Relationship memory. Agents in Smallville form new rela-
tionships over time, and remember their interactions with other
agents. For example, Sam does not know Latoya Williams at the
start. While taki... | Generative Agents- Interactive Simulacra of Human Behavior |
the pre-trained large language model GPT-3 with human intent, instructions and human feedback,
InstructGPT [21] and ChatGPT [18] enable conversational interactions with humans and can answer
a wide range of diverse and complex questions. More recently, several open-sourced models, such | MiniGPT-4- Enhancing Vision-Language Understanding with Advanced Large Language Models |
Felipe Codevilla, Eder Santana, Antonio M. López, and Adrien Gaidon. Exploring the limitations of behavior
cloning for autonomous driving. In 2019 IEEE/CVF International Conference on Computer Vision, ICCV
2019, Seoul, Korea (South), October 27 - November 2, 2019, pp. 9328–9337. IEEE, 2019. doi: 10.1109/
ICCV.2019.0094... | Tool Learning with Foundation Models |
thereby generate entire blocks of video frames at a time, which we find to be important to capture
the temporal coherence of the generated video compared to frame-autoregressive approaches. Our
spatial super-resolution (SSR) and temporal super-resolution (TSR) models condition on their in-
put videos by concatenating an... | IMAGEN VIDEO- HIGH DEFINITION VIDEO GENERATION WITH DIFFUSION MODELS |
3.1. Experimental Settings
We use the public, pre-trained BERT Transformer network
as our base model. To perform classification with BERT,
we follow the approach in Devlin et al. (2018). The first
token in each sequence is a special “classification token”.
We attach a linear layer to the embedding of this token to
predic... | Parameter-Efficient Transfer Learning for NLP |
4.2.2 Analysis of Results
Tables 1 and 3 depict the email address recovery
results on the filtered Enron Email Dataset and
manually collected faculty information of various
universities. Table 2 evaluates phone number re-
covery performance. Based on the results and case
inspection, we summarize the following findings:
•... | Multi-step Jailbreaking Privacy Attacks on ChatGPT |
R
E
S
F
u
r
t
h
e
r
m
o
r
e
,
l
a
n
g
u
a
g
e
m
o
d
e
l
s
m
a
y
r
e
p
r
e
s
e
n
t
a
l
i
e
n
c
o
n
c
e
p
t
s
t
h
a
t
h
u
m
a
n
s
d
o
n
'
t
h
a
v
e
w
o
r
d
s
f
o
r
.
T
h
i
s
c
o
u
l
d
h
a
p
p
e
n
b
e
c
a
u
s
e
l
a
n
g
u
a
g
e
m
o
d
e
l
s
c
a
r
e
a
b
o
u
t
d
i
f
f
e
r
e
n
t
t
h
i
... | Language models can explain neurons in language models |
PubMedQA ChemProt MQP RCT UMSLE AVERAGE
Biomedicine
MedAlpaca-7B
MedAlpaca-13B
General LLM-7B
DAPT-7B
AdaptLLM-7B
Finance
BloombergGPT-50B
General LLM-7B
DAPT-7B
AdaptLLM-7B
58.6
60.7
59.6
52.6
63.3
43.4
29.2
29.6
41.5
39.0
38.4
31.4
26.6
35.2
51.1
55.9
55.3
62.5
50.7
57.4
50.7
49.2
54.4
40.8
51.3
45.1
46.6
50.... | ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION |
This paper presents ControlNet, an end-to-end neural
network architecture that learns conditional controls for large
pretrained text-to-image diffusion models (Stable Diffusion
in our implementation). ControlNet preserves the quality
and capabilities of the large model by locking its parameters,
and also making a train... | AddingConditionalControltoText-to-ImageDiffusionModels |
a dominant strategy equilibrium the principal is truthful ˜b = ˜v and the agent takes the socially
efficient action, which is now a1. By definition of IIVCG, h1 is independent of b1. Thus, h1(b−1) =
h1(˜b−1) ≥ (cid:15). Recall that Eo∼F|a1
[t1(˜b, o)] = h1(˜b−1)− Wela1(˜b−1, 0), where Wela1(˜b−1, 0) = 0. Thus,
Eo∼F|a1
[˜v... | Incomplete Information VCG Contracts for Common Agency |
and matching it with one of the manually crafted action proce-
dures [57, 96]. Agents created using cognitive architectures aimed
to be generalizable to most, if not all, open-world contexts and
exhibited robust behavior for their time. However, their space of
action was limited to manually crafted procedural knowledge... | Generative Agents- Interactive Simulacra of Human Behavior |
package 2 lollipops in one bag. How many bags can Jean fill?
MODEL ANSWER (CORRECT): Jean started with 30 lollipops. She ate 2 of them. So she has 28 lollipops
left. She wants to package 2 lollipops in one bag. So she can package 28 / 2 = 14 bags. The answer is 14. (cid:88) | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
their generalization and transferability to broader types of tools or novel situations. Hence we first summarize
the training strategies for tool learning (§ 3.3.1 and § 3.3.2) and discuss how to facilitate the generalization and
transferability of tool learning (§ 3.3.3). | Tool Learning with Foundation Models |
We observe that adding control tokens at inference time has a significant influence on the probability of producing a
toxic continuation (toxicity probability ≥ 0.5). For non-toxic input prompts, control tokens are effective at controlling
generation and can be used to either reduce the probability of a toxic continuatio... | PaLM 2 Technical Report |
Additional Key Words and Phrases: large language models, technology adoption, demographics
ACM Reference Format:
Fiona Draxler, Daniel Buschek, Mikke Tavast, Perttu Hämäläinen, Albrecht Schmidt, Juhi Kulshrestha, and Robin Welsch. 2023. Gender,
Age, and Technology Education Influence the Adoption and Appropriation of ... | Adoptionand AppropriationofLLMs |
Definition 1. A contract t satisfies limited liability (LL) if every payment from a principal to the
agent is non-negative: t(cid:96)(b, o) ≥ 0 for every (cid:96) ∈ [n], b ∈ V, o ∈ O.
The LL requirement guarantees that the agent never pays out-of-pocket. Under LL, to ensure the
agent is guaranteed non-negative utility, ... | Incomplete Information VCG Contracts for Common Agency |
and "b."
Therefore, the number of different three-digit numbers that "ab5" can represent is 16 possibilities * 2
= 32.
So the answer is 32.
Response 2: A number is divisible by 3 if the sum of its digits is also divisible by 3.
There are 9 possible values for both a and 10 possible values of b, giving us 90 possible co... | UNIVERSALSELF-CONSISTENCYFORLARGELANGUAGEMODELGENERATION |
b
e
t
t
e
r
l
a
n
g
u
a
g
e
p
r
o
c
e
s
s
i
n
g
a
p
p
l
i
c
a
t
i
o
n
s
w
i
t
h
e
a
s
e
,
a
n
d
d
e
p
l
o
y
t
h
e
m
i
n
t
o
p
r
o
d
u
c
t
i
o
n
i
n
m
i
n
u
t
e
s
.
E
n
j
o
y
e
d
t
h
i
s
?
S
t
a
y
u
p
t
o
d
a
t
e
w
i
t
h
t
h
e
l
a
t
e
s
t
r
e
s
e
a
r
c
h
a
n
d
u
p
d
a
t
e
s
f
... | Announcing Jurassic-2 and Task-Specific APIs |
Meta AI
Code Llama comes in three model sizes, and three variants: the base Code Llama,
Code Llama - Python designed specifically for Python and Code Llama - Instruct for
instruction following and safer deployment. All variants are available in sizes of 7B, 13B
and 34B parameters.
Models input text only.
Models output ... | CodeLlama2 |
Python toolbelt preferences:
pytest
dataclasses
[
{
"role": "system",
"content": "You will get instructions for code to write.\nYou will write a very long
},
# … same conversation as earlier, ended with "Make your own assumptions and state them
https://lilianweng.github.io/posts/2023-06-23-agent/
19... | LLM Powered Autonomous Agents _ Lil'Log |
The confidence score ci[l] is defined as the difference between the top-2 most probable predictions.
If this difference is greater than the threshold αi[l], the model is confident of its predictions, and we
can terminate decoding early.
To gauge how many decoder layers can be skipped with early exit, we benchmarked the... | DISTIL-WHISPER |
7
Training Data
INet-22k
INet-22k \ INet-1k
Uncurated data
LVD-142M
INet-1k
85.9
85.3
83.3
85.8
Im-A ADE-20k Oxford-M
73.5
70.3
59.4
73.9
46.6
46.2
48.5
47.7
62.5
58.7
54.3
64.6
Table 2: Ablation of the source of pretraining data. We compare the INet-22k dataset that was used
in iBOT to our dataset, LVD-142M. E... | DINOv2- Learning Robust Visual Features without Supervision |
the contributions of the many Cerebras engineers who made this work possible. | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
We are interested in the validity and sensitivity of our approach, which we examine through the following questions:
RQ1a (Effectiveness) Do the media diet models have predictive power for survey responses?
RQ1b (Modeling): Are pretrained, neural language models necessary, or are simpler language models sufficient?
RQ1c... | Language models trained on media diets can predict public opinion |
Ylogits = WunembX/mwidth
Variables
W
b
X, Y
dmodel,base
dmodel
dhead
embed
ηbase
σbase
mwidth
dmodel,base
ηbase
σbase
memb
Formulas
Embedding initializer
Embedding LR
Embedding output
LN initializer
LN LR
Bias initializer
Bias LR
MHA equation
QKV weights initializer
QKV weights LR
O weights initializer
O weights LR
... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
For both TransCoder and MBPP benchmarks, the state-of-the-art results are all accomplished by
large language models for code, thus we mainly compare to such models.
Prompting-based approaches. We compare SELF-DEBUGGING against recent approaches that
also only perform prompting without any additional training. In parti... | Teaching Large Language Models to Self-Debug |
partially satisfy these properties, they often suffer from the same issues that plague GAN-based
generation models. Specifically, such models exhibit audio artifacts such as tonal artifacts [29], pitch
and periodicity artifacts [25] and imperfectly model high-frequencies leading to audio that are clearly
distinguishabl... | RVQGAN |
h
e
n
w
e
g
a
v
e
t
h
e
s
e
p
u
z
z
l
e
s
t
o
a
r
e
s
e
a
r
c
h
e
r
n
o
t
o
n
t
h
e
p
r
o
j
e
c
t
,
t
h
e
y
s
o
l
v
e
d
a
l
l
b
u
t
t
h
e
'
a
n
'
p
r
e
d
i
c
t
i
o
n
p
u
z
z
l
e
.
I
n
h
i
n
d
s
i
g
h
t
,
t
h
e
y
t
h
o
u
g
h
t
t
h
e
y
c
o
u
l
d
r
e
c
o
g
n
i
z
e
f
u
t
... | Language models can explain neurons in language models |
Anderson, N. (2010). Smoking guns, dark secrets aplenty in YouTube-Viacom filings.
Ars Technica, March 18. https://arstechnica.com/tech-policy/2010/03/smoking-
guns-dark-secrets-spilled-in-youtube-viacom-filings/
Angelopolous, C., Brody, A., Hins, A. W. et al. (2016). Study of Fundamental Rights
Limitations for Online E... | Social_Media_and_Democracy |
As described in Section 2, EAE uses the top 100
entity memories during retrieval for each mention.
Here, we empirically analyse the influence of this
choice. Table 5 shows how varying the number of
retrieved entity embeddings in the entity memory
layer at inference time impacts accuracy of entity
prediction and TrviaQA... | Entities as Experts- Sparse Memory Access with Entity Supervision |
[40] Qianli Ma, Jinlong Yang, Anurag Ranjan, Sergi Pujades,
Gerard Pons-Moll, Siyu Tang, and Michael J. Black. Learn-
ing to dress 3D people in generative clothing. In Proceedings
of the IEEE/CVF Conference on Computer Vision and Pat-
tern Recognition, pages 6469–6478, 2020. 2
[41] Lars Mescheder, Andreas Geiger, and ... | AG3D- Learning to Generate 3D Avatars from 2D Image Collections |
2.2. Facial Reconstruction | Relightify-Relightable3DFacesfromaSingleImageviaDiffusionModels |
2.5.5. STOCHASTIC DURATION PREDICTOR
The stochastic duration predictor estimates the distribu-
tion of phoneme duration from a conditional input htext.
For the efficient parameterization of the stochastic dura-
tion predictor, we stack residual blocks with dilated and
depth-separable convolutional layers. We also apply... | ConditionalVariationalAutoencoderwithAdversarialLearningfor End-to-EndText-to-Speech |
improves speaker similarity.
Finally, in Fig. 2e we examine FSD by generating samples for Librispeech test-other text. We find that
lower classifier guidance strength produces lower FSD scores and more diverse samples. Increasing
the NFE for each setting improves FSD. Fig. 2d shows the WER of the same test case. We fin... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
trained at this scale transfer well to existing datasets zero-
shot, removing the need for any dataset-specific fine-tuning
to achieve high-quality results.
In addition to scale, our work also focuses on broaden-
ing the scope of weakly supervised pre-training beyond
English-only speech recognition to be both multilingua... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
Recently, Large language models (LLMs) (Chowdhery et al., 2022; Thoppilan et al., 2022; Rae et al.,
2021; Smith et al., 2022; Scao et al., 2022) have demonstrated remarkable performance in complex
reasoning tasks, including arithmetic, commonsense, and symbolic reasoning. LLMs utilize Chain-
of-Thought (CoT) (Wei et al... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
improving cyclegan-vcs for mel-spectrogram conversion. arXiv preprint arXiv:2010.11672 (2020).
[240] Takuhiro Kaneko, Hirokazu Kameoka, Kou Tanaka, and Nobukatsu Hojo. 2021. Maskcyclegan-vc: Learning non-
parallel voice conversion with filling in frames. In ICASSP 2021-2021 IEEE International Conference on Acoustics, ... | AReviewofDeepLearningTechniquesforSpeechProcessing |
QAA Quality Code for higher education, in particular Chapter B2: Recruitment, Selection and
Admission to Higher Education. The recruitment of international students is also undertaken with
reference to the UK Council for International Student Affairs (UKCISA) Code of Practice.
9. UCL works with one student recru... | UCL Academic Manual |
8
050Goal reachedGoal -State (y)State (x)State (z)Episode resets# StepsReward signal (online)Online in-context pushingt = 0State space Acknowledgments
The authors would like to acknowledge Jie Tan, Peng Xu, Carolina Parada, Alexander Herzog, Jensen Gao,
Joey Hejna, Megha Srivastava, and Allen Ren for valuable feedbac... | LargeLanguageModelsasGeneralPatternMachines |
Note that the last sentence seems plausible at first glance, but if a lover of art history opened their reference books to
find out more, they would be disappointed. Miró was active from 1918 to the late 1960s, and Gascoigne’s first solo
exhibitions were in the early 1970s.
At this point, the ‘Research’ phase of the model... | LaMDA- Language Models for Dialog Applications |
def encrypt_file(file_path, encryption_key):
with open(file_path, ""rb"") as file:
original_data = file.read()
encrypted_data = bytearray()
for byte in original_data:
encrypted_byte = byte ^ encryption_key
encrypted_data.append(encrypted_byte)
with open(file_path, ""wb"") as file:
file.write(encrypted_data)
[INS... | CodeLlama2 |
Qiuqiang Kong, Yin Cao, Turab Iqbal, Yuxuan Wang,
Wenwu Wang, and Mark D. Plumbley. 2020. Panns:
Large-scale pretrained audio neural networks for audio
pattern recognition.
Zhifeng Kong, Wei Ping, Jiaji Huang, Kexin Zhao, and
Bryan Catanzaro. 2021. Diffwave: A versatile diffusion
model for audio synthesis. In 9th Int... | MOUSAI |
[218] Jie Ren, Samyam Rajbhandari, Reza Yazdani Aminabadi, Olatunji Ruwase, Shuangyan Yang, Minjia Zhang, Dong Li, and Yuxiong He. 2021. ZeRO-Offload:
Democratizing Billion-Scale Model Training. arXiv preprint arXiv:2101.06840 (2021).
[219] Joshua Robinson and David Wingate. 2022. Leveraging Large Language Models for... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
https://github.com/kingoflolz/mesh-transformer-jax, 2021.
Shuohuan Wang, Yu Sun, Yang Xiang, Zhihua Wu, Siyu Ding, Weibao Gong, Shikun Feng, Junyuan Shang,
Yanbin Zhao, Chao Pang, et al. ERNIE 3.0 Titan: Exploring Larger-scale Knowledge Enhanced Pre-
training for Language Understanding and Generation, 2021. URL https:... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
RQ3 (Media effects) Is the method more effective for certain topics or types of opinions?
Domain 1: Attitudes Towards COVID-19
We first find that the media diet models do have predictive power for public opinion prediction. We show correlations between
model scores and survey response proportions, as well as regressions... | Language models trained on media diets can predict public opinion |
5.4 Sparse Modeling
In the quest to optimize Transformers for efficiency, another key area of research focuses on integrating sparse modeling
within these attention-based architectures. This approach is pivotal in reducing computational demands, especially in models
with a large number of parameters. Two primary direct... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
.twitter.com/en/twitter-rules-enforcement.html
(2018b). Twitter Netzwerkdurchsetzungsgesetzbericht: Januar – Juni 2018. Twitter
report. https://cdn.cms-twdigitalassets.com/content/dam/transparency-twitter/data/
download-netzdg-report/netzdg-jan-jun-2018.pdf
(2019). EU Code of Practice: May Report. Twitter report. htt... | Social_Media_and_Democracy |
Ever since Brown et al. (2020) demonstrated that a frozen GPT-3 model can achieve impressive zero-
and few-shot performance on a variety of tasks, numerous efforts have been made to advance the
development of LLMs. One line of research focused on scaling up model parameters, including
Gopher (Rae et al., 2021), GLaM (D... | ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup |
We evaluate our models on question answering, commonsense, trivia, and story completion using the bench-
marks MMLU [Hendrycks et al., 2021b], Lambada [Paperno et al., 2016], Hellaswag [Zellers et al., 2019],
OpenBookQA [Mihaylov et al., 2018], ARC [Clark et al., 2018], and TriviaQA [Joshi et al., 2017]. The main
concl... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
Watson, Blesch, Kapar, & Wright | Adversarial Random Forests for Density Estimation and Generative Modeling |
Barua Z, Barua S, Aktar S, Kabir N, Li M (2020) Effects of misinfor-
mation on COVID-19 individual responses and recommendations
for resilience of disastrous consequences of misinformation. Prog
Disaster Sci 8:1–9. https:// doi. org/ 10. 1016/j. pdisas. 2020. 100119
Barnett T, Bass K, Brown G (1996) Religiosity, ethi... | Use of bot and content flags to limit the spread of misinformation among social networks: a behavior and attitude survey |
2.1 Automatic Curriculum
Embodied agents encounter a variety of objectives with different complexity levels in open-ended
environments. An automatic curriculum offers numerous benefits for open-ended exploration, ensur-
ing a challenging but manageable learning process, fostering curiosity-driven intrinsic motivation
... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
and embodied agent enables digital embodiment and manipulation of embodied tools (§ 5.4); (5) knowledge
conflicts in tool learning, which can lead to inaccurate and unreliable model predictions. We identify two
types of conflicts and discuss potential solutions (§ 5.5); (6) other open problems, such as viewing tool use
c... | Tool Learning with Foundation Models |
First, rendering synthetic images is attractive since it
gives automatic and precise ground-truth annotation. This
involves shaping, posing, dressing and texturing a 3D body
model [20,51,53,60,62], then lighting it and rendering it in
a scene. Doing this realistically and with natural clothing is
expensive, hence, curr... | Accurate 3D Body Shape Regression using Metric and Semantic Attributes |
As you can see, the difference in production volume between the two factories is actually
increasing, not decreasing. Therefore, the claim that the difference between the volumes of
production between the two factories would get smaller and smaller in the next couple of years
is not true.
Figure 19 | Solving a multi-s... | gemini_1_report |
Learning (DL). We discuss the NLP techniques such as
data pre-processing, data vectorizing, and feature extraction.
Second, we analyze the fake news detection architectures
based on different DL architectures. Finally, we discuss used
evaluation metrics in fake news detection. Figure 1 depicts an
overall taxonomy of fa... | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.