text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
(CTGAN) and tabular VAE (TVAE), two deep learning algo-
rithms for generative modeling with mixed continuous and
categorical features. We include three additional state-of-
the-art tabular GAN architectures for comparison: invertible
tabular GAN (IT-GAN) (Lee et al., 2021), regularized com-
pound conditional GAN (RCC-G... | Adversarial Random Forests for Density Estimation and Generative Modeling |
images for randomly initialized networks. To further boost performance, we propose an iterative
version, where we obtain a sequence of distilled images and these distilled images can be trained
with multiple epochs (passes). Finally, we study a simple linear model, deriving a lower bound on the
size of distilled data r... | DATASET DISTILLATION |
Thus, m2(b) ≥ (cid:15). It follows that ka∗(b) ≤(cid:80)
(cid:96)∈[n] m(cid:96)(b).
3. a∗(b) = a3. Note that ka3 = γ by using w = (0, 0, γ). Further, since a∗(b) = a3, and by the
definition of the domain, Wela∗(b)(b) = b1(o3)− γ. Thus, Wela∗(b)(b−2, 0) = Wela∗(b)(b−(cid:96), 0) =
b1(o3)−γ ∀(cid:96) > 2, and min˜v2∈V 2... | Incomplete Information VCG Contracts for Common Agency |
Our entity memory extends the Entities-as-
Experts (EaE) model (Févry et al., 2020).
It is
both the current state-of-the-art for a number of
tasks and simpler to use than most prior models
because it does not require external components
for entity linking or entity encoding (like (Peters
et al., 2019; Zhang et al., 201... | Adaptable and Interpretable Neural Memory Over Symbolic Knowledge |
3 | CHAIN-OF-THOUGHTREASONING IS APOLICY IMPROVEMENTOPERATOR |
can be used to meet some constraints such as privacy [99].In this scenario, the choice between using a fine-tuned model
or a LLM is task-specific and also depends on many factors, including desired performance, computational resources,
and deployment constraints. | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
ment, gathering sensor and video data. An AI system analyses this data and interprets it
(e.g., object or fire detection). The interpreted information is sent to a FR who can act on
this information by validating or discarding it, allowing the AI system to learn from such
feedback. | DevelopingTeamDesignPatternsfor HybridIntelligenceSystems |
3
2
0
2
c
e
D
3
1
]
G
L
.
s
c
[
2
v
0
9
2
8
1
.
5
0
3
2
:
v
i
X
r
a
Direct Preference Optimization:
Your Language Model is Secretly a Reward Model
Rafael Rafailov∗†
Archit Sharma∗†
Eric Mitchell∗†
Stefano Ermon†‡
Christopher D. Manning†
Chelsea Finn†
†Stanford University ‡CZ Biohub
{rafailo... | Direct Preference Optimization |
Prompt engineering still matters, though. Although the results are relatively robust to the prompt
for arithmetic reasoning, we want to be clear that prompt engineering still does matter, and can
improve performance significantly in many cases. Though most chain of thought annotations
outperform standard prompting, ther... | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
When composing the test samples, we also make
efforts to ensure comprehensive coverage. We first
read through the text samples in the dataset and
then compose samples that are reasonable music
descriptions but do not exist in the data. The text
prompts comprehensively cover all genres that we
evaluate and incorporate e... | Moûsai |
that CoT is an emergent ability of scaling language models (Wei et al., 2022a). For instruction-finetuned
models on BBH, using CoT only helps for the Flan-PaLM 540B, Flan-cont-PaLM 62B, and U-PaLM 540B. | Scaling Instruction-Finetuned Language Models |
to resolve the XOR problem. The subsequent rise and application of Support Vector Machines
(SVMs) [26] and deep learning [97] have marked both progress and setbacks in the AI landscape.
A significant takeaway from previous attempts is the paramount importance of AI evaluation,
which serves as a critical tool to identif... | ASurveyonEvaluationofLargeLanguageModels |
and quality. We compare three augmentation datasets: without self-curation, A(2)
that are
progressively smaller augmentation sets but of higher data quality (see Table 2 for statistics). Similar
to observations in LIMA using human-annotated data [Zhou et al., 2023], improving the quality of
the training data dramatical... | Self-AlignmentwithInstructionBacktranslation |
speech datasets including these languages is hard, and (2) due to the lack of bilingual speech
datasets with corresponding para-/non-linguistic characteristics in both source and target languages,
para-/non-linguistic characteristics in the source speech cannot be transferred to the translated speech.
This paper addres... | Translatotron3 |
Because this usage extends to the social – where bots have real-time
conversations with humans on sites like Facebook and Twitter – the engineers
who build them often view them as more than a tool but less than human,
a proxy for the creator (Woolley, Shorey, and Howard 2018). Social bots play
a key role in generating ... | Social_Media_and_Democracy |
5 Conclusion and Future Work
In conclusion, in this report, we investigate the existing open-source language model, namely,
LLaMA, on medical applications, showing that the model performs unsatisfactorily on QA
tasks. To inject domain knowledge into the pre-trained model, We conduct a preliminary
investigation by fine-... | PMC-LLaMA- Further Finetuning LLaMA on Medical Papers |
2.5 Literature Review of Tool Learning
From the perspective of learning objectives, tool learning can be categorized into two main streams (Figure 3).
The first stream, tool-augmented learning, seeks to augment foundation models with the execution results
from various tools (Mialon et al., 2023). In this paradigm, tool... | Tool Learning with Foundation Models |
[42] A. Ramesh, M. Pavlov, G. Goh, S. Gray, C. Voss, A. Radford, M. Chen, and I. Sutskever.
Zero-shot text-to-image generation. In International Conference on Machine Learning, pages
8821–8831. PMLR, 2021.
[43] A. Ramesh, P. Dhariwal, A. Nichol, C. Chu, and M. Chen. Hierarchical text-conditional image
generation with... | Adding Conditional Control to Text-to-Image Diffusion Models |
bounds of the canonical volume as an approximate bound
of the object surface. To do so, we run marching cubes on
a 643 grid to extract a surface mesh and then set L as the
axis-aligned (x, y, z) bounds of the extracted surface.
Near-far planes. To generate samples for volume render-
ing, we dynamically compute the dept... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
−βT log
pθ(xl
pref(xl
t−1|xl
t)
t−1|xl
t)
(cid:19)
pθ(xw
pref(xw
(12)
Efficient training via gradient descent is now possible.
However, sampling from reverse joint pθ(xt−1, xt|x0, c)
is still intractable and r of Eq. (9) has an expectation
over pθ(x1:T|x0). So we approximate the reverse process
pθ(x1:T|x0) with t... | DiffusionModelAlignmentUsing Direct Preference Optimization |
Each group of tokens is routed jointly with load balancing across experts incentivized by an auxiliary
loss as proposed in Shazeer et al. (2017) (see Appendix A for details). Tokens compete for expert
assignment against other tokens in their group, rather than the entire batch, and expert specialization
is heavily influ... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
[19] A. Kupcsik, D. Hsu, and W. S. Lee. Learning Dynamic Robot-to-Human Object Handover
from Human Feedback, pages 161–176. Springer International Publishing, 01 2018. ISBN
978-3-319-51531-1. doi: 10.1007/978-3-319-51532-8_10.
[20] S. Levine. Reinforcement learning and control as probabilistic inference: Tutorial and ... | Direct Preference Optimization |
that current approaches are limited because they only take into account visual image features. This also indicates that
future research has to aim towards a more holistic approach if we ought to build systems that can achieve a human-like
understanding of art. | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
(cid:40)
we measure the cosine value of the angle between the gen-
erated normal gk and the kth ray’s viewing direction vk.
The design rationale behind this approach lies in the ori-
entation of normals, which are deliberately set to face out-
ward, while the viewing direction is inward-facing. This
configuration ens... | Wonder3D |
r
a
c
y
l
o
s
t
f
o
r
a
h
u
g
e
s
p
e
e
d
u
p
.
A
c
o
u
p
l
e
c
o
m
m
o
n
c
h
o
i
c
e
s
o
f
A
N
N
a
l
g
o
r
i
t
h
m
s
f
o
r
f
a
s
t
M
I
P
S
:
L
S
H
(
L
o
c
a
l
i
t
y
-
S
e
n
s
i
t
i
v
e
H
a
s
h
i
n
g
)
:
I
t
i
n
t
r
o
d
u
c
e
s
a
h
a
s
h
i
n
g
f
u
n
c
t
i
o
n
s
u
c
h
t
h
a
... | LLM Powered Autonomous Agents _ Lil'Log |
Hacker News displays and stores comments in a
tree-like manner, with children comments replying
to parent comments. However, most language mod-
els require input data to be in a sequential form.
Considering each path through the comment tree
as a sequence could be detrimental, since there will
be a large amount of near... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
[44] Yuda Qiu, Xiaojie Xu, Lingteng Qiu, Yan Pan, Yushuang Wu,
Weikai Chen, and Xiaoguang Han. 3dcaricshop: A dataset
and a baseline method for single-view 3d caricature face
reconstruction. In Proceedings of the IEEE/CVF Conference
on Computer Vision and Pattern Recognition, pages 10236–
10245, 2021. 2, 3
[45] Anurag... | RaBit- Parametric Modeling of 3D Biped Cartoon Characters with a Topological-consistent Dataset |
The goal of text-to-SQL tasks is to generate the corresponding SQL query given a question and the
database information, and Figure 2 presents an exemplar in our prompt for generating the SQL query.
Following prior work evaluating large language models for text-to-SQL generation [43, 49, 37], we
evaluate SELF-DEBUGGING ... | Teaching Large Language Models to Self-Debug |
RLHF Response → So this is python code for generating all the prime numbers up to a certain number. It uses the
Sieve of Eratosthenes algorithm, imports the math module, defines a function prime_sieve()
which takes an input num, validates it, creates a boolean array sieve, sets the prime numbers
in the array, and finally... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
29
(a) Overall violation percentage.
(b) Overall safety and helpfulness mean rating.
Figure 17: Overall safety measures. Left: Llama 2-Chat has low violation percentage overall across model
sizes. Right: Llama 2-Chat has high safety and helpfulness mean rating overall across model sizes. It is
important to note tha... | Llama2 |
specify serverless. You can also create new pro or
classic SQL warehouses using either method. You
can upgrade a pro or classic SQL warehouse to a
serverless SQL warehouse or a classic SQL
warehouse to a pro SQL warehouse. You can also
downgrade from serverless to pro or class... | Dolly 2 Databricks |
Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K. Bert:
Pre-training of deep bidirectional transformers for lan-
guage understanding. arXiv preprint arXiv:1810.04805,
2018.
French, R. M. Catastrophic forgetting in connectionist net-
works. Trends in cognitive sciences, 1999.
Girshick, R., Donahue, J., Darrell, T.... | Parameter-Efficient Transfer Learning for NLP |
[32] Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael
Pritch, Michael Rubinstein, and Kfir Aberman.
Dreambooth: Fine tuning text-to-image diffusion
models for subject-driven generation. In Proceedings
of the IEEE/CVF Conference on Computer Vision and
Pattern Recognition (CVPR), 2023. 2, 3, 4, 6, 12, 13,
17
[33] Chitwan... | A Neural Space-Time Representation for Text-to-Image Personalization |
We considered a window of three years of data to build demand
forecasting models for them. We issued predictions over the last six
months of data and built explanations for all materials in the last
threemonths.OurdemandforecastingmodelsusetheSupportVector
Regressionalgorithm[104].Weusednestedcross-validation[105]to
ev... | Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio |
Large Language Models for Agent Planning.
Inspired by the strong emergent capabilities of
LLMs, such as zero-shot prompting and complex reasoning [72, 37, 38, 36, 73, 74], embodied
agent research [75–78] has witnessed a significant increase in the utilization of LLMs for planning
purposes. Recent efforts can be roughly... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
newsroom.fb.com/news/2018/08/introducing-the-ad-archive-api/
Leerssen, P., Ausloos, J., Zarouali, B., Helberger, N., & de Vreese, C. H. (2019).
Platform ad archives: Promises and pitfalls. Internet Policy Review, 8(4), 1–21.
Levy, K. E. C., & Johns, D. M. (2016). When open data is a Trojan horse: The
weaponization o... | Social_Media_and_Democracy |
We further demonstrate that EAE’s learned en-
tity representations are better than the pre-trained
embeddings used by Zhang et al. (2019); Peters
et al. (2019) at knowledge probing tasks and the TA-
CRED relation extraction task (Zhang et al., 2017;
Alt et al., 2020). We show that training EAE to
focus on entities is b... | Entities as Experts- Sparse Memory Access with Entity Supervision |
83 | PaLM 2 Technical Report |
148See e.g. the roll-out scenario described in Bostrom (2014). It’s worth noting that in principle, if the ability
to accurately assess whether a given instance of misaligned power-seeking will succeed arises sufficiently early
in the AI system’s we’re building/training, the period of time in which we see widespread and... | Is Power-Seeking AI an Existential Risk? |
[20] Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh,
Hieu Pham, Quoc V. Le, Yun-Hsuan Sung, Zhen Li, and Tom
Duerig. Scaling up visual and vision-language representation
learning with noisy text supervision. In Proceedings of the
38th International Conference on Machine Learning, ICML
2021, 18-24 July 2021,... | REVEAL-Retrieval-AugmentedVisual-LanguagePre-Trainingwith Multi-SourceMultimodalKnowledgeMemory |
Figure 1: We propose a two-stage cascading diffusion
method, where the first stage compresses the music
using a novel diffusion autoencoder, and the second
stage generates music from the reduced representation
conditioned on the encoding of a textual description.
In this paper, we further bridge the gap between
text a... | Moûsai |
et al. (2021) introduced the APPS dataset, a collection of 10,000 coding competition problems, and
were the first to evaluate large transformer language models on competitive programming. The
authors found that the overall solve rate on interview or competition level problems using large
language models remained close t... | alphacode |
[4] Chowdhery, A., Narang, S., Devlin, J., Bosma, M., Mishra, G., Roberts, A.,
Barham, P., Chung, H.W., Sutton, C., Gehrmann, S., Schuh, P., Shi, K.,
Tsvyashchenko, S., Maynez, J., Rao, A., Barnes, P., Tay, Y., Shazeer, N.,
Prabhakaran, V., Reif, E., Du, N., Hutchinson, B., Pope, R., Bradbury, J.,
Austin, J., Isard, M.... | PersonalityTraitsinLargeLanguageModels |
Studying the impact, spread, and prevention of misinforma-
tion on social media is necessary as modern society relies on
belief and trust in authorities and public health information
to respond collectively to global crisis and tragedy. While
this study is novel in its approach to filtering respondents
to different... | Use of bot and content flags to limit the spread of misinformation among social networks: a behavior and attitude survey |
3.2 Multitask Pretraining
In the domain of audio processing, diverse audio datasets have been developed to address specific tasks,
as shown in Table 1. Qwen-Audio aims to perform co-training using a wide range of audio datasets. The
objective is to train a unified model capable of supporting all audio tasks, eliminatin... | Qwen-Audio |
Cerebras Modelzoo: https://github.com/Cerebras/modelzoo.
©2023 Cerebras Systems Inc. All Rights Reserved.
13
Cerebras-GPT: Open Compute-Optimal Language Models
Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts,
Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrm... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
dashed line follows the procedure from Févry et al. (2020). | Adaptable and Interpretable Neural Memory Over Symbolic Knowledge |
,
Translate the following question into SQL.
Question: In which year were most departments established?
SQL: SELECT creation, COUNT(*) FROM department GROUP BY creation ORDER BY
COUNT(*) DESC LIMIT 1
Feedback: The SQL prediction above is wrong. Please fix the SQL.
SQL: SELECT creation FROM department GROUP BY crea... | Teaching Large Language Models to Self-Debug |
Public Filings and Other Government Disclosures
Valuable information about platform operations sometimes surfaces to the
public through court or other public filings.38 Documents made public in the
Viacom v. YouTube case, for example, made headlines for their revelations
about both parties to the suit (Anderson 2010; Yo... | Social_Media_and_Democracy |
16
100101102103104105106Sample budget0.000.050.100.150.200.250.3010@k300M1B3B9B41B100101102103104105106Sample budget0.00.10.20.30.4pass@k300M1B3B9B41B Competition-Level Code Generation with AlphaCode
(a) Solve Rate vs. Training Compute
(b) Solve Rate vs. Sampling Compute
Figure 7 | Solve rate scaling vs. Compute. T... | alphacode |
Although political misinformation is not a new phenomenon, the topic has
received renewed attention in recent years, in conjunction with sweeping
changes in the contemporary media environment. As the Internet and,
particularly, social media become an increasingly common source for political
information (Shearer and Mat... | Social_Media_and_Democracy |
T. Mikolov, I. Sutskever, K. Chen, G. S. Corrado, and J. Dean. Distributed representations of
words and phrases and their compositionality. Advances in neural information processing
systems, 26, 2013. 9, 39
B. Min, H. Ross, E. Sulem, A. P. B. Veyseh, T. H. Nguyen, O. Sainz, E. Agirre, I. Heinz, and
D. Roth. Recent adv... | A Cookbook of Self-Supervised Learning |
Efforts to amend or eliminate CDA 230 should be seen in the broader context
from open, participatory approaches to the problem of
of a retreat
disinformation. In light of the perceived absence or weakness of a robust
crowd, interventions have turned toward managing disinformation through
legislative or
judgments of pri... | Social_Media_and_Democracy |
P
r
i
c
e
M
a
t
c
h
G
u
a
r
a
n
t
e
e
,
w
h
i
c
h
a
d
d
s
a
V
C
G
-
l
i
k
e
a
s
p
e
c
t
t
o
t
h
e
i
r
s
e
e
m
i
n
g
l
y
s
t
a
n
d
a
r
d
p
o
s
t
e
d
-
p
r
i
c
i
n
g
p
o
l
i
c
y
.
A
m
a
z
o
n
a
n
d
e
B
a
y
p
r
o
v
i
d
e
o
p
p
o
r
t
u
n
i
t
i
e
s
f
o
r
v
a
r
i
o
u
s
t
y
p
e
s
o
f
... | Principal-agent VCG contracts - ScienceDirect |
As such, founders in the generative AI era need to make data more of a strategic priority. After all, other traditional sources of
competitive advantage are likely not feasible. Given the ferment of innovation around generative AI, no startup is likely to maintain
a large technology advantage for long, especially when ... | 4 Trends for AI Startups and Generative AI Companies |
Introduction | Mixtral of Experts paper |
internet content
tests
Amazon. (2015). Amazon Information Request Report. Amazon report. http://d0
.awsstatic.com/certifications/Transparency_Report.pdf
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
244
Daphne Keller & Paddy Leerssen
Anderson, J., Carlson, K., Stender, M., ... | Social_Media_and_Democracy |
The rapid advancement of conversational and chat-based language models has led
to remarkable progress in complex task-solving. However, their success heavily
relies on human input to guide the conversation, which can be challenging and
time-consuming. This paper explores the potential of building scalable techniques
to... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
discrete random variables. In International Conference on Learning Representations, 2017.
Omid Nejati Manzari, Hamid Ahmadabadi, Hossein Kashiani, Shahriar B Shokouhi, and Ahmad Ayatollahi.
Medvit: A robust vision transformer for generalized medical image classification. Computers in Biology
and Medicine, 157:106791, ... | BiomedGPT |
semantics”)
Table 8: Example of correct chains of thought produced by the model for the GSM8K dataset. | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
where b−(cid:96)(o) =(cid:80)
h(cid:96)
minimize
˜v(cid:96)∈V (cid:96),h(cid:96)∈R
subject to F|aj · (b−(cid:96) + ˜v(cid:96)) − ψ(aj) ≤ h(cid:96)
(cid:96)(cid:54)=(cid:96)(cid:48)∈[n] b(cid:96)(cid:48)
(o) ∀o ∈ O.
(LP1)
∀j ∈ [q],
Proof. The linear program above is solvable in polynomial time since it consists of... | Incomplete Information VCG Contracts for Common Agency |
adversarial network for artifact-free vocoder. arXiv preprint arXiv:2206.13404 (2022).
[33] Jon Barker, Shinji Watanabe, Emmanuel Vincent, and Jan Trmal. 2018. The fifth’CHiME’speech separation and
recognition challenge: dataset, task and baselines. arXiv preprint arXiv:1803.10609 (2018).
[34] Murali Karthick Baskar... | AReviewofDeepLearningTechniquesforSpeechProcessing |
2.2 TEXT-TO-VIDEO GENERATION WITH BIDIRECTIONAL TRANSFORMERS
In this stage, the text-to-video task can be formulated as a sequence-to-sequence problem to predict
video tokens given the paired text embeddings. Most of recent methods [38, 65, 60, 22] adopt a
transformer model for these sequence-to-sequence tasks. In the... | PHENAKI- VARIABLE LENGTH VIDEO GENERATION FROM OPEN DOMAIN TEXTUAL DESCRIPTIONS |
April 30, 2021
Search Meta AI
Who We Are
Latest Work
https://ai.facebook.com/blog/imagebind-six-modalities-binding-ai/
12/13
11/05/2023, 05:04
ImageBind: Holistic AI learning across six modalities
Our Actions
Newsletter
Privacy Policy
Terms
Cookies
Meta © 2023
https://ai.facebook.com/blog/imagebind-six-mod... | ImageBind_ Holistic AI learning across six modalities |
At the same time, the space of possible models within that intersection is vast, perhaps
even infinite; saying that the right architecture is there is a start, but only a start,
something like saying that a web browser probably ought to be written in a language
49
THE NEXT DECADE IN AI / GARY MARC... | The Next Decade in AI- |
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
Amendment of Section 230
259
perceived to possess the resources and technical competence to effectively create
effective systems of detection and mitigation.3
Without the cooperation of these platforms, proposals to address politi... | Social_Media_and_Democracy |
[99] J. Feldman, The neural binding problem(s), Cogn. Neurodyn. 7 (1) (2013) 1–11.
[100] H. Paulheim, Knowledge graph refinement: a survey of approaches and evaluation methods, Semant. Web 8 (3) (2017) 489–508.
[101] W. Beek, L. Rietveld, S. Schlobach, F. van Harmelen, Lod laundromat: why the semantic web needs centrali... | Knowledge graphs as tools for explainable machine learning: A survey |
LIF(θ) = − M(cid:88)
Nm(cid:88)
m=1
j=1
log pθ(zm,j|˜x, z<m, zm,<j)
(6)
3.4
Instruction Tuning | DOCLLM |
Wei Dong, Charikar Moses, and Kai Li. 2011. Efficient
k-nearest neighbor graph construction for generic
similarity measures. In Proceedings of the 20th in-
ternational conference on World wide web, pages
577–586.
Nan Du, Yanping Huang, Andrew M Dai, Simon Tong,
Dmitry Lepikhin, Yuanzhong Xu, Maxim Krikun,
Yanqi Zhou, ... | DataManagementForLargeLanguageModels-ASurvey |
[Yu et al., 2023b] Zichun Yu, Chenyan Xiong, Shi Yu, and
Zhiyuan Liu. Augmentation-adapted retriever improves
generalization of language models as generic plug-in.
arXiv preprint arXiv:2305.17331, 2023.
[Zhang et al., 2019] Zhengyan Zhang, Xu Han, Zhiyuan Liu,
Xin Jiang, Maosong Sun, and Qun Liu. Ernie: Enhanced
langu... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
49Richard Ngo suggested this point in conversation; see also his discussion of the “generalization-based
approach” here. The training and fine-tuning used for GPT-3 may also be suggestive of patterns of this type.
14
broad-scope world-modeling are very useful types of cognition, we might expect to see
them cropping ... | Is Power-Seeking AI an Existential Risk? |
Internet Platforms and Content Moderation
245
Center for Democracy and Technology. (2017). Mixed Messages? The Limits of
Automated Social Media Content Analysis. Center for Democracy and Technology
report. https://cdt.org/insight/mixed-messages-the-limits-of-automated-social-media-
content-analysis/
Chang, B. (2018)... | Social_Media_and_Democracy |
Jurassic-X: Crossing the neuro-symbolic chasm with the MRKL system
https://www.ai21.com/blog/jurassic-x-crossing-the-neuro-symbolic-chasm-with-the-mrkl-system
13/13
T
e
r
m
s
o
f
U
s
e
P
r
i
v
a
c
y
P
o
l
i
c
y
S
t
a
A
I
2
1
S
t
u
d
i
o
W
o
r
d
t
u
n
e
W
o
r
d
t
u
n
e
R
e
a
d
C
o
m
p
a
n
y | Jurassic-X_ Crossing the neuro-symbolic chasm with the MRKL system |
6 CONCLUSION
Large language models have disrupted the technology market within a few months, enabling novel kinds of AI support
in a variety of professional and non-professional contexts alike. The simple chat-style interfaces promise equitable
access. However, young males are currently the predominant user group of LL... | Adoptionand AppropriationofLLMs |
l
y
n
o
e
x
p
l
a
i
n
a
b
i
l
i
t
y
m
e
t
h
o
d
t
h
a
t
a
i
m
s
t
o
i
l
l
u
m
i
n
a
t
e
t
h
i
s
h
i
g
h
l
y
c
o
m
p
l
e
x
p
r
o
c
e
s
s
.
I
n
a
d
d
i
t
i
o
n
,
e
x
i
s
t
i
n
g
a
p
p
r
o
a
c
h
e
s
a
r
e
u
n
a
b
l
e
t
o
p
r
o
d
u
c
e
d
i
v
e
r
s
e
e
x
p
l
a
n
a
t
i
o
n
s
,
g
e
a
... | PhD Fellow in Explainable Natural Language Understanding |
7
One possible solution to this problem could be to increase the size of the model, based on the assumption that the model
can effectively memorize more of the training data. However, some facts change over time, like the answers to ‘How
old is Rafael Nadal?’ or ‘What time is it in California?’. Lazaridou et al. (202... | LaMDA- Language Models for Dialog Applications |
2.2 Training Details
We adopt most of the pretraining setting and model architecture from Llama 1. We use the standard
transformer architecture (Vaswani et al., 2017), apply pre-normalization using RMSNorm (Zhang and
Sennrich, 2019), use the SwiGLU activation function (Shazeer, 2020), and rotary positional embeddings
(... | Llama2 |
[10] John Brooke. 1995. SUS: A quick and dirty usability scale. Usability Evaluation in Industry 189 (Nov. 1995).
[11] Paul-Christian Bürkner. 2017. brms: An R package for Bayesian multilevel models using Stan. Journal of statistical
[12] Astrid Carolus, Martin Koch, Samantha Straka, Marc Latoschik, and Carolin Wienri... | AI enhance sour performance |
6.4. Removing redundant actions | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
Figure 19. An illustration of synthesized texture maps. For each
row, the leftmost and the rightmost textures are from 3DBiCar,
while the other three textures are interpolated results generated by
RaBit under different weights.
decoder branches are adopted to learn the appearances of
these local regions from the input... | RaBit- Parametric Modeling of 3D Biped Cartoon Characters with a Topological-consistent Dataset |
tracking data has been available for two decades, systematic commercial
tracking of digital advertising is very new, and until the aftermath of the 2016 | Social_Media_and_Democracy |
(obtained by minimizing the right-hand side over all v(cid:96) ∈ V (cid:96) and dropping the ∀v(cid:96) ∈ V (cid:96) quanti-
fier): h(cid:96)(b−(cid:96)) − Wela∗(b)(b−(cid:96), 0) ≤ min˜v(cid:96)∈V (cid:96) Wela∗(b−(cid:96),˜v(cid:96))(b−(cid:96), ˜v(cid:96)) − Wela∗(b)(b−(cid:96), 0) ∀(cid:96) ∈ [n], b ∈ V.
That is, by... | Incomplete Information VCG Contracts for Common Agency |
13
Figure 3: Performance of models from the GPT family on the various tasks used for evaluating emergence tested on
the closed prompting strategy. Models which are instruction-tuned (IT) and those which are non-instruction-tuned
(non-IT) are evaluated in the few-shot (FS), zero-shot (ZS) settings using BERTScore accu... | AreEmergentAbilitiesinLarge Language Models just In-Context |
One major challenge is that LLM-based agents, when performing non-textual actions, require an
intermediate step of generating thoughts or formulating tool usage in textual form before eventually
translating them into concrete actions. This intermediary process consumes time and reduces the
response speed. However, this... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
estimates news accounts for just 3 percent of time spent; see Hindman 2018.)
As attention is going elsewhere, advertising is following, especially as people
increasingly embrace platform products and services like search, social media,
messaging applications, and the like that offer cheap, targeted advertising at
scale... | Social_Media_and_Democracy |
traffic and, | Social_Media_and_Democracy |
proved the quality of audio encoders dramatically, the lack
of an equivalently high-quality pre-trained decoder, com-
bined with a recommended protocol of dataset-specific fine-
tuning, is a crucial weakness which limits their usefulness
and robustness. The goal of a speech recognition system
should be to work reliably “... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
Tool Use Several approaches aim to equip LMs
with the ability to use external tools such as search
engines (Komeili et al., 2022; Thoppilan et al.,
2022; Lazaridou et al., 2022; Shuster et al., 2022;
Yao et al., 2022), web browsers (Nakano et al.,
2021), calculators (Cobbe et al., 2021; Thoppilan
et al., 2022), transla... | Toolformer |
transparency often needs to be backed with regulatory oversight, as scholars
critical about corporate transparency in practice emphasize the possible
existence of “opaque” forms of transparency that do not actively make the
democratically relevant information visible but rather can be used to obfuscate
processes and pr... | Social_Media_and_Democracy |
is a music that is ....’
– Complete the answer based on the music descrip-
4.3 MUImage Dataset
We present the MUImage dataset for generating appropri-
ate music for input images. The MUImage dataset is as-
sembled by obtaining music samples from the AudioSet
with paired videos. A random frame is selected from each
v... | M2UGen |
model outputs.
17
√
Figure 13 These figures show training curves in the
KL vs PM score plane, exhibiting the approximate
linear relationship between these variables, especially in the left-hand plot using the more highly-performing
52B PMs. We observe some instability in the smaller models, likely because the traini... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
[13] Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, brian ichter, Fei Xia, Ed Chi,
Quoc V Le, and Denny Zhou. Chain of Thought Prompting Elicits Reasoning in Large Language
Models. In Conference on Neural Information Processing Systems (NeurIPS), 2022.
[14] Takeshi Kojima, Shixiang (Shane) Gu, Machel Reid, Yu... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
For the pretraining tasks, we consider text-to-video
(T2V), text-to-image (T2I), and four self-supervised learn-
ing (SSL) tasks: frame prediction (FP), central inpainting
and central outpainting (Painting) [74] and audio-video con-
tinuation (AVCont) where the model is provided with the
first frame and its correspondin... | VideoPoet |
If fully onchain games are not a viable approach, the reasons to be excited about
them could get expressed in less “onchain” ways. The games that work may use
smart contracts minimally, or not at all. O | The Open Problems of Onchain Games |
# of pairs
24M
60M
69M
19M
90M
3M
6M
∼ 270M
B Implementation Details
We list the hyperparameters in Table 11. Since some evaluation datasets have long texts, we freeze
the position embeddings during both pre-training and fine-tuning and set the maximum text length to
512 for evaluation.
For the Quora duplicate ... | E5 |
1
2
3
dgrid = DistributionGrid(expert_parallel_group_size = dist.get_world_size())
encoder = TransformerMoEEncoderLayer(256, 4, nexperts=8, distribution_grid = dgrid)
encoder = ORTModule(encoder)
Looking Forward
We are able to scale various language, speech and vision models using the Mixture Of Experts technique by
... | Scaling Speech, Language and Vision Models with Mixture of Experts Technique - Microsoft Community Hub |
els. In ICLR, 2020.
[58] Chenfei Wu, Lun Huang, Qianxi Zhang, Binyang Li, Lei Ji, Fan Yang, Guillermo Sapiro, and
Nan Duan. Godiva: Generating open-domain videos from natural descriptions. arXiv preprint
arXiv:2104.14806, 2021.
[59] Chenfei Wu, Jian Liang, Xiaowei Hu, Zhe Gan, Jianfeng Wang, Lijuan Wang, Zicheng Liu,... | PHENAKI- VARIABLE LENGTH VIDEO GENERATION FROM OPEN DOMAIN TEXTUAL DESCRIPTIONS |
issue among contemporary liberal democracies, and in many respects the new
internet regulations, actual and proposed, are extensions of existing practices.
We conclude that, in the US case, content regulation will be very difficult to
achieve politically and that antitrust should be considered as an alternative. | Social_Media_and_Democracy |
Figure 16: Example of standard python bugs found and explained by Code Llama - Instruct.
43
Prompt: I have a pandas dataframe with the columns "decoding", "Capabilities", "Fine-tuning", "Model size", "HE pass@1",
"MBPP pass@1". I want a seaborn figure with two scatterplots side-by-side. The two plots show "HE pass@1... | CodeLlama2 |
Limitations and Future Works. Future directions are
abundant and worth exploring. Our dataset and method
mainly focus on pop music, while other music genres, such
as classical music, can further be explored. Since we only
generate symbolic music with piano tracks, other instru-
ments (e.g. drum, guitar, and string) can... | VideoBackgroundMusicGeneration |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.