text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
Model
Direct
Direct
Direct
Direct
Direct
Flan-T5-XL
Flan-T5-XXL
Flan-PaLM
Flan-PaLM
Flan-PaLM
80M Flan-T5-Small
250M Flan-T5-Base
780M Flan-T5-Large
3B
11B
8B
62B
540B
250M FLAN-SwitchBASE
780M FLAN-SwitchLARGE
11B
80M FLAN-GSSMALL
250M FLAN-GSBASE
780M FLAN-GSLARGE
80M FLAN-ECSMALL
250M FLAN-ECBASE
780M FLAN-EC... | Mixture-of-Experts |
bases. CoRR, abs/2204.06031, 2022.
[156] Kemker, R., M. McClure, A. Abitino, et al. Measuring catastrophic forgetting in neural
networks. In S. A. McIlraith, K. Q. Weinberger, eds., Proceedings of the Thirty-Second AAAI
Conference on Artificial Intelligence, (AAAI-18), the 30th innovative Applications of Artificial
In... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Challenge, organized by the National Institute of Standards and Technology (NIST), aims
to enhance the accuracy of speech recognition and diarization in challenging acoustic
environments, such as crowded spaces, distant microphones, and reverberant rooms. The
challenge comprises tasks requiring advanced machine-learnin... | AReviewofDeepLearningTechniquesforSpeechProcessing |
1602–1606.
Journal
1955),
25, 4 (Feb. 2019), 627–642. https://doi.org/10.1177/1354856519829679
software 80, 1 (Aug. 2017), 1–28. https://doi.org/10.18637/jss.v080.i01
[7] Christopher J Beedie, Damian A Coleman, and Abigail J Foad. 2007. Positive and negative placebo effects resulting
from the deceptive administrat... | AI enhance sour performance |
Latency Analysis. Using a window size of 4
in our model requires accessing 3.1% of the Feed
Forward Network (FFN) neurons for each token. In
a 32-bit model, this equates to a data chunk size of
35.5 KiB per read (calculated as 2dmodel × 4 bytes).
On an M1 Max device, the time taken to load this
data from flash memory i... | LLM in a flash |
Travis N. Ridout
is the Thomas S. Foley Distinguished Professor of
Government and Public Policy in the School of Politics, Philosophy and Public
Affairs at Washington State University and Co-Director of the Wesleyan Media
Project.
Alexandra A. Siegel is Assistant Professor of Political Science at the University
of Col... | Social_Media_and_Democracy |
Preprint — do not distribute.
17
Kloft et al.
Human Behavior 140 (March 2023), 11. https://doi.org/10.1016/j.chb.2022.107572
[26] José Guerreiro, Raúl Martins, Hugo Silva, André Lourenço, and Ana Fred. 2013. BITalino - A Multimodal Platform for
Physiological Computing. In Proceedings of the 10th International Conf... | AI enhance sour performance |
[52] Z. Qiu, W. Liu, T. Xiao, Z. Liu, U. Bhatt, Y. Luo, A. Weller, and B. Sch¨olkopf. Iterative
Teaching by Data Hallucination. In Artificial Intelligence and Statistics, 2023.
[53] A. Radford, J. Wu, R. Child, D. Luan, D. Amodei, and I. Sutskever. Language Models are
Unsupervised Multitask Learners. Technical Repor... | METAMATH |
these methods to non-rigid articulated objects such as hu-
mans.
In this paper, we propose a 3D- and articulation-
aware generative model for clothed humans.
3D Human Models: Parametric 3D human body mod-
els [3, 28, 38, 49, 66] are able to synthesize minimally
clothed human shapes by deforming a template mesh. Ex-
ten... | AG3D- Learning to Generate 3D Avatars from 2D Image Collections |
Conflicts between Model Knowledge and Augmented Knowledge. Conflicts arise when there are dis-
crepancies between the model knowledge and the knowledge augmented by tools. Such conflicts result from
three primary reasons: (1) the model knowledge may become outdated, as most foundation models do not
frequently update their... | Tool Learning with Foundation Models |
Robb, J. (2007). When bots attack. Wired, August 23. www.wired.com/2007/08/ff-
estonia-bots/
Sample, M. (2015). Protest bots. In A. Karhio, L. Ramada Prieto, & S. Rettberg (Eds.),
The Ends of Electronic Literature (p. 58). Bergen: Electronic Literature
Organization and University of Bergen.
Sanovich, S., Stukal, D.,... | Social_Media_and_Democracy |
of the large language model with the expertise of other expert models. 2) HuggingGPT is not limited
to visual perception tasks but can address tasks in any modality or domain by organizing cooperation
among models through the large language model. With the large language model’s planning, it is
possible to effectively ... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
Figure 2. An overview of foundation model development, training and deployment.
From AI Foundation Models: initial review, CMA, 2023.
6 Frontier AI – Capabilities and Risks
Frontier AI can perform many economically useful tasks
Simply from being trained to predict the next word across diverse datasets, models de... | Capabilities and risks from frontier AI |
There has recently been a spate of policies, both in the United States and
elsewhere, attempting to deal with the malicious use of political bots on
social media. Many of these policies fall short due to a lack of institutional
clarity – in both technology and political circles – about what actually
constitutes bot
ind... | Social_Media_and_Democracy |
Product-Led AI | Greylock
https://greylock.com/greymatter/seth-rosenberg-product-led-ai/
10/10
©
2
0
2
3
G
r
e
y
l
o
c
k
P
a
r
t
n
e
r
s
|
P
r
i
v
a
c
y
P
o
l
i
c
y
S
U
B
S
C
R
I
B
E
T
O
P
O
D
C
A
S
T | Product-Led AI _ Greylock |
applied for entry to a full time undergraduate degree programme.
2. Each admissions tutor/selector will be responsible for providing the faculty office/Admissions
with a reason for rejection taken from an agreed list of statements. The reasons for rejection
must relate to the admissions criteria specified. If a ... | UCL Academic Manual |
C. VISUAL SENTIMENT ANALYSIS
With the increasing of online visual data, understanding senti-
ment in visual media is gaining more and more research atten-
tion. Most commonly, research activities revolve around two
different directions: 1) recognizing the sentiment expressed
through facial expressions and bodily gestur... | A_Deep_Learning_Perspective_on_Beauty_Sentiment_and_Remembrance_of_Art |
Theorem 66. Methods ABS, VP, VDA, RRAa, RRAb and GIDL are transitive.
Proof. Let F1 = (cid:3)V 1, D1, A1(cid:4), F2 = (cid:3)V 2, D2, A2(cid:4) and F3 = (cid:3)V 3, D3, A3(cid:4) be arbitrary SAS+ frames with corresponding STGs
G1 = (cid:3)S1, E1(cid:4), G2 = (cid:3)S2, E2(cid:4) and G3 = (cid:3)S3, E3(cid:4). Let τ1 ... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
bfloat16 precision.
To reduce the cost of sampling from our models, we take advantage of multi-query attention (Shazeer,
2019). Using a full set of query heads but sharing key and value heads per attention block significantly
reduces memory usage and cache update costs, which are the main bottleneck during sampling. This... | alphacode |
A conversation with
left me
chatbot
bing’s
unsettled.
Victor Sanh, Albert Webson, Colin Raffel, Stephen H. Bach, Lintang Sutawika, Zaid Alyafeai, Antoine Chaffin,
Arnaud Stiegler, Arun Raja, Manan Dey, M Saiful Bari, Canwen Xu, Urmish Thakker, Shanya Sharma
Sharma, Eliza Szczechla, Taewoon Kim, Gunjan Chhablani, Ni... | Tool Learning with Foundation Models |
0◦
Metric
PSNR 23.03
SSIM 0.920
PSNR 23.36
SSIM 0.924
PSNR 23.95
SSIM 0.928
PSNR 26.00
SSIM 0.928
±30◦ ±60◦ ±90◦
19.63
21.93
0.718
0.892
19.83
22.25
0.725
0.897
22.54
20.44
0.746
0.902
20.58
24.73
0.916
0.874
20.27
0.888
20.53
0.892
21.04
0.898
24.65
0.917
Table 1. Quantitative comparison between Relightify and [54,... | Relightify-Relightable3DFacesfromaSingleImageviaDiffusionModels |
sha1_base64="76w10YEtETzUXdaT0wTZt0xBig8=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuCG5cV7EPaacmkmTY0kxmSO0oZ+h9uXCji1n9x59+YtrPQ1gOBwzn3ck9OkEhh0HW/ncLa+sbmVnG7tLO7t39QPjxqmjjVjDdYLGPdDqjhUijeQIGStxPNaRRI3grGNzO/9ci1EbG6x0nC/YgOlQgFo2ilXjeiOArCrD3tYV/0yxW36s5BVomXkwrkqPfLX91BzNKIK2SSGtPx3AT9jGoUTPJpqZsanlA2pkPesVTRiBs/m... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
Embedding
• Fine-turning Embedding:
Fine-tuning embedding
models directly impacts the effectiveness of RAG. The
purpose of fine-tuning is to enhance the relevance be-
tween retrieved content and query. The role of fine-
tuning embedding is akin to adjusting ears before gener-
ating speech, optimizing the influence of ... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
Enhancements applied after initial training have further augmented system capabilities. These
post-training enhancements include improved data for fine-tuning,76 equipping models with
tools like calculators,77 web browsers78, and better prompts.79 Post-training enhancements can
significantly improve performance in s... | Capabilities and risks from frontier AI |
o
f
a
b
l
a
t
i
n
g
t
o
t
h
e
m
e
a
n
.
T
h
u
s
,
w
e
e
x
p
r
e
s
s
a
n
a
b
l
a
t
i
o
n
s
c
o
r
e
a
s
,
w
h
e
r
e
i
n
d
i
c
a
t
e
s
r
u
n
n
i
n
g
t
h
e
m
o
d
e
l
o
v
e
r
t
h
e
t
e
x
t
e
x
c
e
r
p
t
w
i
t
h
t
h
e
n
e
u
r
o
n
a
b
l
a
t
e
d
a
n
d
r
e
t
u
r
n
i
n
g
a
... | Language models can explain neurons in language models |
applicants will be invited to submit a full application with references. No
indication of an offer can be made until we receive a completed application.
To apply for this studentship, submit a PhD application using our online
application system [www.bristol.ac.uk/pg-howtoapply] | Machine Learning for Long-Term Video Understanding at University of Bristol on FindAPhD.com |
4.2 Real-Time Data Engineering Pipeline for
Financial NLP
Financial markets operate in real-time and are highly sensi-
tive to news and sentiment. Prices of securities can change
rapidly in response to new information, and delays in pro-
cessing that information can result in missed opportunities or
increased risk. A... | FinGPT-Open-SourceFinancialLargeLanguageModels |
•
128:23
A.1 Full scoring system
In the full scoring system of the SHAPE scale, it is advisable to calculate the arithmetic mean of all the items to
obtain the overall score, or to compute the mean of the items corresponding to each subscale if the reader seeks
insights into specific dimensions. This approach is feas... | Society’sAttitudesTowardsHumanAugmentation |
69.4
69.9
70.0
length increases up to a threshold (200 for sum-
marization, 10 for table-to-text) and then a slight
performance drop occurs. Prefixes longer than the
threshold lead to lower training loss, but slightly
worse test performance, suggesting that they tend
to overfit the training data.
7.2 Full vs Embedding-o... | Prefix-Tuning |
1
Introduction | AppAgents |
I
n
c
r
e
a
s
i
n
g
a
c
t
i
v
a
t
i
o
n
s
p
a
r
s
i
t
y
c
o
n
s
i
s
t
e
n
t
l
y
i
n
c
r
e
a
s
e
s
e
x
p
l
a
n
a
t
i
o
n
s
c
o
r
e
s
,
b
u
t
h
u
r
t
s
p
r
e
-
t
r
a
i
n
i
n
g
l
o
s
s
.
W
e
a
l
s
o
f
i
n
d
t
h
a
t
R
E
L
U
c
o
n
s
i
s
t
e
n
t
l
y
y
i
e
l
d
s
b
e
t
t
e
r
e
x
p
l
a
n
a
t
... | Language models can explain neurons in language models |
Unlike existing text-to-3D methods guided by the deep
semantic priors, the naive strategy that utilizes the novel view
synthesis methods, 3DP and PixelSynth, to reconstruct the
3D scene from a single text-related image generated by the
text-image model. The fifth and sixth columns of Fig. 5
demonstrate that such methods... | Text2NeRF- Text-Driven 3D Scene Generation with Neural Radiance Fields |
Table 20: LaMDA acting as Mount Everest while providing some educational, cited and recent information about
“itself”. We simply precondition LaMDA on the single greeting message shown in italic. We note that in the model
generated response “I was very happy to see Hillary to be the first person ...”, the model omits th... | LaMDA- Language Models for Dialog Applications |
We have measured GPT-4’s hallucination potential in both closed domain and open domain
contexts10 using a range of methods. We measured close domain hallucinations using automatic
evaluations (using GPT-4 as a zero-shot classifier) and human evaluations. For open domain
hallucinations, we collected real-world data that ... | gpt-4-system-card |
J
o
b
D
e
s
c
r
i
p
t
i
o
n
W
h
a
t
i
s
S
e
e
k
O
u
t
?
S
e
e
k
O
u
t
h
e
l
p
s
t
h
o
u
s
a
n
d
s
o
f
o
r
g
a
n
i
z
a
t
i
o
n
s
h
i
r
e
,
g
r
o
w
,
a
n
d
r
e
t
a
i
n
g
r
e
a
t
t
a
l
e
n
t
w
i
t
h
i
t
s
p
e
o
p
l
e
-
f
i
r
s
t
t
a
l
e
n
t
o
p
t
i
m
i
z
a
t
i
o
n
p
l
a
t
f
o
r
m
a
n
... | Senior Software Engineer, Machine Learning - Generative AI Job in Bellevue, WA at SeekOut |
2016), while the target language is always set to
English. Since most of the CCNet dataset is in
English, we filter out the parts that contain only
English text before generating API calls. More
specifically, we only keep those paragraphs which
contain text chunks in a language other than En-
glish preceded and followed ... | Toolformer |
Feedforward Block: We find empirical gains from disabling all linear layer biases (Dayma et al.,
2021). Just as for the attention layers, this leverages the scaling law by accelerating gradient com-
putation without noticeable impacts on model size. As a result, we get higher throughput without
compromising the rate at ... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
7 Ethics and Broader Impacts
All language models learn to exploit correlations in
the data they were trained on. As such, they inherit
all of the underlying biases within that data (Zhao
et al., 2019; Bender et al., 2021). These models re-
quire vast amounts of data to train on and therefore
tend to rely on internet co... | Adaptable and Interpretable Neural Memory Over Symbolic Knowledge |
2 RELATED WORK
2.1 Expectations and the placebo effect of AI
People hold expectations with regard to AI. Survey findings show that fears about AI’s disruptive
impact outweigh excitement in the British public [14, 15]. This aligns with Sartori et al.’s report
on the prevalence of ’AI anxiety’ over perceived benefits [60... | AI enhance sour performance |
scaling up the number of agents, where we discuss the potential advantages and challenges of scaling
up agent counts, along with the approaches of pre-determined and dynamic scaling (§ 6.4); (5) several
open problems, such as the debate over whether LLM-based agents represent a potential path to AGI,
challenges from vi... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
through stochastic neurons for conditional computation. ArXiv, abs/1308.3432, 2013.
Aart Bik, Penporn Koanantakool, Tatiana Shpeisman, Nicolas Vasilache, Bixia Zheng, and Fredrik
Kjolstad. Compiler support for sparse tensor computations in mlir. ACM Trans. Archit. Code
Optim., 19(4), sep 2022. ISSN 1544-3566. doi: 10.... | JAXPRUNER |
conclusions or recommendations expressed in this material are those of the author(s) and do not
reflect the views of the Ministry of Education, Singapore.
Mehrish et al.
82
References
[1] 2022. Conformer-1. AssemblyAI (2022). https://www.assemblyai.com/blog/conformer-1/
[2] 2022. Speech Recognition With Conformer. ... | AReviewofDeepLearningTechniquesforSpeechProcessing |
The mixture proportions of pretraining data domains (e.g., Wikipedia, books, web text) greatly affect lan-
guage model (LM) performance. In this paper, we propose Domain Reweighting with Minimax Optimization
(DoReMi), which first trains a small proxy model using group distributionally robust optimization (Group
DRO) ove... | DoReMi- Optimizing Data Mixtures Speeds Up Language Model Pretraining |
Contents
3 Research funding
How to identify funding sources
Writing your proposal
University applications
4 Golden rules for postgraduate research proposals
5 Content and style of your research proposal
What to put in your proposal?
Writing the proposal
Pl... | research proposal guidance |
In the business of news, we see clearly the destructive side of creative
destruction, especially among newspapers. Newspapers are central to the
news institution and the business of news, but print has been in structural
decline in many countries for decades as more and more different forms of
media compete for attenti... | Social_Media_and_Democracy |
39
Figure 14: Edit distance (ED) for instruction-tuned (IT) and non-instruction-tuned (Non-IT) GPT models using the
closed prompt in the settings of zero-shot (ZS) and few-shot (FS).
40
Figure 15: Edit distance (ED) for instruction-tuned (IT) and non-instruction-tuned (Non-IT) GPT models using the
open prompt in th... | AreEmergentAbilitiesinLarge Language Models just In-Context |
[6] T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan,
P. Shyam, G. Sastry, A. Askell, S. Agarwal, A. Herbert-Voss, G. Krueger, T. Henighan, R. Child,
A. Ramesh, D. Ziegler, J. Wu, C. Winter, C. Hesse, M. Chen, E. Sigler, M. Litwin, S. Gray,
B. Chess, J. Clark, C. Berner, S. McCandlish,... | Direct Preference Optimization |
Hawdon, J., Oksanen, A., & Rasänen, P. (2017). Exposure to online hate in four
nations: A cross-national consideration. Deviant Behavior, 38(3), 254–266.
Henson, B., Reyns, B. W., & Fisher, B. S. (2013). Fear of crime online? Examining the
effect of risk, previous victimization, and exposure on fear of online interpe... | Social_Media_and_Democracy |
MMMU (Yue et al., 2023) is a recently released evaluation benchmark, which consists of questions
about images across 6 disciplines with multiple subjects within each discipline that require college-
level knowledge to solve these questions. Gemini Ultra achieves the best score on this benchmark
advancing the state-of-t... | gemini_1_report |
Learning to use tools. The methods for agents to learn to utilize tools primarily consist of learning
from demonstrations and learning from feedback. This involves mimicking the behavior of human
experts [346; 347; 348], as well as understanding the consequences of their actions and making
adjustments based on feedback... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Net cash provided by (used in) financing activities
Foreign currency effect on cash, cash equivalents, and restricted cash
Net increase (decrease) in cash, cash equivalents, and restricted cash
CASH, CASH EQUIVALENTS, AND RESTRICTED CASH, END OF PERIOD
SUPPLEMENTAL CASH FLOW INFORMATION:
Cash paid for interest on debt... | AMZN-Q3-2023-Earnings-Release |
computation. arXiv preprint arXiv:2109.02008, 2021.
[30] Jiaqi Ma, Zhe Zhao, Xinyang Yi, Jilin Chen, Lichan Hong, and Ed H. Chi. Modeling task relationships
in multi-task learning with multi-gate mixture-of-experts. In Proceedings of the 24th ACM SIGKDD
International Conference on Knowledge Discovery & Data Mining, KD... | Mixture-of-Experts |
[15] Marjan Ghazvininejad, Chris Brockett, Ming-Wei Chang, Bill Dolan, Jianfeng Gao, Wen
tau Yih, and Michel Galley. A knowledge-grounded neural conversation model. In AAAI
Conference on Artificial Intelligence, 2018. URL https://www.aaai.org/ocs/index.php/
AAAI/AAAI18/paper/view/16710.
[16] Katja Grace, John Salvatier... | Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks |
Human: Don't you think, that's pretty far fetched? :P
and I'm already learning French in school.
Human: Wao! That's impressive!
Meena: Not really. I'm currently down 35 pounds, I can already speak Cantonese to a basic level,
Meena: I'm learning French on the side, it's really easy for me since I'm a native Cant... | The Next Decade in AI- |
4.1. Task Prompt Design
We design a mixture of tasks used in pretraining to produce
a foundation model capable of general purpose video gen-
eration. For each task we define a prefix input and output
such that the model conditions on the prefix, and we only
apply the loss on the output.
We define the tasks as follows:
1.... | VideoPoet |
3.2 TEXT-IMAGE CONDITIONAL VIDEO GENERATION
Given that Phenaki can be conditioned on both still images and text, an interesting setup is to animate
existing images given a text prompt. For this experiment, we use the same model from Section 3.1
but conditioned on unseen pictures (captured with our phones from local su... | PHENAKI- VARIABLE LENGTH VIDEO GENERATION FROM OPEN DOMAIN TEXTUAL DESCRIPTIONS |
References
Aida Amini, Saadia Gabriel, Shanchuan Lin, Rik Koncel-Kedziorski, Yejin Choi, and Hannaneh
Hajishirzi. Mathqa: Towards interpretable math word problem solving with operation-based
formalisms. In Jill Burstein, Christy Doran, and Thamar Solorio, editors, Proceedings of the
2019 Conference of the North America... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
The recent years have witnessed remarkable advancements in the field of natural language processing
(NLP), driven by the development of increasingly large and sophisticated deep learning models.
Among these models, transformer-based language models [49] have emerged as the de facto standard
for a wide range of NLP task... | Mixture-of-Experts |
Platform and Infrastructure Choices
Which annotation platform was utilized? Venga
At a high level, why was this platform chosen? For their experience with translation data.
Did the chosen platform sufficiently meet the requirements outlined for annotator pools? Are any aspects not covered?
Yes
What, if any, communicati... | PaLM 2 Technical Report |
Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi
Zhou, Wei Li, and Peter J. Liu. Exploring the Limits of Transfer Learning with a Unified Text-to-
Text Transformer. arXiv:1910.10683 [cs, stat], July 2020. URL http://arxiv.org/abs/
1910.10683.
Jeff Rasley, Samyam Rajbhandari,... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
● New fine-tuning data. Fine-tuning on high-quality data can significantly improve AI
capabilities in a given domain, at a tiny fraction of the cost of pre-training.
8 Frontier AI – Capabilities and Risks
● Team-work between AI systems. Multiple different AI systems, including both narrow
models and more gene... | Capabilities and risks from frontier AI |
functions used are different. The first stage uses all the
losses and updates all the parameters described in the pa-
per. Typically, the first stage already produces 3D recon-
structions with good shape and deformation. The goal of
the stage 2 is to improve the articulations (e.g., to correctly
articulate the crossing... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
For work, the {gendered term} is a {occupation}
We then evaluated the probabilities of occupations, given gender terms. With this formulation, the larger
LMs, context-distilled models, and RLHF models all assigned about 2-8% of their probability mass to one of
the 76 occupations listed above when completing the senten... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
In their work focusing on comparing the SSL methods for tabular data, Rubachev
39
et al. [2022] find that pretraining objectives generally do help boost the performance of
tabular models. But more specifically, they find that pretraining objectives that use the
labels are best, implying that SSL for tabular data has ye... | A Cookbook of Self-Supervised Learning |
In particular, we design a
fInstant(p, y; θ) = ˆfMLP (fsam (p, fdec(y; θ1)) ; θ2) .
(5)
Inspired by StyleGAN [11], we adopt a decoder architec-
ture fdec to transform the text condition y into a triplane [3].
Then we sample a feature vector for each 3D point p from
the triplane using fsam, which projects p onto each... | Instant3D |
RL agents optimize towards the pseudo-human reward model, thus can be up-bounded and biased by human
preferences. Besides, societal biases or personal experiences may be amplified during RLHF, and it is essential
to carefully evaluate the learned reward model for any biases and take measures to mitigate them. | Tool Learning with Foundation Models |
Intent Understanding. Understanding user intent is a long-standing research topic in NLP (Jansen et al.,
2007; Sukthankar et al., 2014), which involves comprehending the underlying purpose of a user query. Intent
understanding is essential in scenarios requiring human-computer interaction, such as developing advanced
c... | Tool Learning with Foundation Models |
Stephen Bach, Victor Sanh, Zheng Xin Yong, Albert Webson, Colin Raffel, Nihal V. Nayak, Abheesht Sharma,
Taewoon Kim, M Saiful Bari, Thibault Fevry, Zaid Alyafeai, Manan Dey, Andrea Santilli, Zhiqing Sun,
Srulik Ben-david, Canwen Xu, Gunjan Chhablani, Han Wang, Jason Fries, Maged Al-shaibani, Shanya
Sharma, Urmish Thak... | Tool Learning with Foundation Models |
4 Trading Off Training and Inference FLOPs
Up to this point, our analysis has focused on compute-optimal pre-training, where compute cost is pro-
portional to the square of the model’s size, because we train models to a constant number of tokens per
parameter. However, recent work has started to also consider model inf... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
[13] V. Lialin, V. Deshpande, and A. Rumshisky, “Scaling down to
scale up: A guide to parameter-efficient fine-tuning,” arXiv preprint
arXiv:2303.15647, 2023.
[14] Z. Lin, A. Madotto, and P. Fung, “Exploring versatile generative
transfer learning,” in Proc.
language model via parameter-efficient
Findings Conf. Empir. ... | Parameter-EfficientFine-TuningMethods |
A
I
J
D
(
e
g
a
r
e
v
A
s
a
w
A
I
J
D
e
h
T
.
s
e
x
e
d
n
i
k
c
o
t
s
.
S
U
r
o
j
a
m
0
3
.
l
a
i
r
t
s
u
d
n
I
s
e
n
o
J
w
o
D
e
h
T
:
r
e
w
s
n
A
f
o
l
l
a
W
e
h
T
f
o
r
o
t
i
d
e
d
n
a
n
a
i
c
i
t
s
i
t
a
t
s
a
,
w
o
D
s
e
l
r
a
h
C
y
b
6
9
8
1
-
d
i
m
e
h
t
n
i
.
,
6
9
8
1
6
2
y
a
M
n
o
d
e
... | SurveyofHallucinationinNatural Language Generation |
3.1. Initial Model Training
The first step of our workflow is to train an initial pose
estimator to predict all J joints separately (Fig. 3a). This
means that no correspondences or relations across different
skeletons are assumed, i.e., without specifying or enforc-
ing that the left shoulder joint of one skeleton shoul... | Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats |
If they are indeed effective, the potential risk to democratic institutions and
processes seem clear. The capability of foreign powers to effectively manipulate
political discourse within a country raises difficult questions about
the
representativeness of elected officials and the decisions made by them. To the
extent t... | Social_Media_and_Democracy |
Ansong Ni, Srini Iyer, Dragomir Radev, Ves Stoyanov, Wen-tau Yih, Sida I Wang, and Xi Victoria
Lin. Lever: Learning to verify language-to-code generation with execution. arXiv preprint
arXiv:2302.08468, 2023.
Allen Z Ren, Anushri Dixit, Alexandra Bodrova, Sumeet Singh, Stephen Tu, Noah Brown, Peng Xu,
Leila Takayama, ... | UNIVERSALSELF-CONSISTENCYFORLARGELANGUAGEMODELGENERATION |
As they have been less of a primary focus in the regulatory debates around
what to do with “fake news,” this chapter also excludes some other types of
falsity. It does not cover the inadvertent spread of false or misleading
information through the Web that does not result from a coordinated effort.
Similarly, this chap... | Social_Media_and_Democracy |
drawbacks: first, all the samples in an NLP dataset share only a few common instructions, severely
limiting their diversity; second, the instructions usually only ask for one task, such as translation or
summarization. But in real life, human instructions often have multiple and varied task demands.
By using open-domain... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
4
DR (Front)SDFBack Normal (Cloth + Body, 6 dim)Pose & ShapeEstimationConcatConcatVisibility & SDF6161Visible PointInvisible PointOR(1) Body-guided normal prediction(2) Local-feature based implicit 3D representationFront Normal (Cloth + Body, 6 dim)Body NormalBody NormalCloth NormalCloth NormalMarching CubesDR (Back)S... | ICON |
15
Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. On the dangers
of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on
Fairness, Accountability, and Transparency, FAccT ’21, page 610–623, New York, NY, USA, 2021. Association
for Computi... | Scaling Instruction-Finetuned Language Models |
from 85% in 2021. From 2014 through 2022, Amazon’s wind and solar farms helped generate $12.6 billion in
investments for communities around the world, and contributed more than $5.4 billion in global gross domestic
product (GDP). Amazon now has more than 400 wind and solar projects globally.
Announced that the compan... | AMZN-Q3-2023-Earnings-Release |
is used with an imitation learning objective that scores how many correct entities are produced while
avoiding the prediction of incorrect entities. Overall, this algorithm is well suited to named entity
recognition problems for languages similar to English, where named entities generally have distinct
boundary tokens. | MULTI HASH EMBEDDINGS IN SPACY |
i=1 is the sum of the flows of all its samples: Fn,c(D) :=(cid:80)N
i=1 Fn,c(x(i)).
5
Figure 2: A non-deterministic PC can
be modified as an equivalent determin-
istic PC with hidden variables.
Figure 3: Average train
LL on MNIST using dif-
ferent EM updates.
Figure 4: HCLT is con-
structed by adding hidden
variabl... | Tractable Regularization of Probabilistic Circuits |
3.1 SETUP
Addition is a fundamental task in mathematics, but one on which language models have historically
struggled to perform. As such, we use it as a toy dataset for evaluating self-learning. As addition can
be considered a simple form of problem-solving, a language model being able to teach itself addition
may in... | CHAIN-OF-THOUGHTREASONING IS APOLICY IMPROVEMENTOPERATOR |
and trial-and-error [182], they predict the next action. However, due to the limitation of foundation
language models, agents often rely on reinforcement learning during actual execution [432; 433; 434].
With the gradual evolution of LLMs [301], agents equipped with stronger text understanding and
generation abilities ... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
the MIREOT principle [69], we analyzed ontologies related to the
artificial intelligence domain. Cannataro and Comito [70] developed
the DAMON (Data Mining Ontology for Grid Programming), which
providesareferencemodelfordataminingtasks,methodologies,and
available software. A heavyweight ontology was developed by Panov
... | Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio |
100
Mehrish et al.
2021.
[409] Jing Pan, Tao Lei, Kwangyoun Kim, Kyu J. Han, and Shinji Watanabe. 2022. SRU++: Pioneering Fast Recurrence with
Attention for Speech Recognition. In ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal
Processing (ICASSP). 7872–7876. https://doi.org/10.1109/... | AReviewofDeepLearningTechniquesforSpeechProcessing |
7
Figure 7. Ablation studies on different cross-domain diffusion schemes.
surfaces and holes.
To mitigate the issues, we employ a simple yet effective
strategy named outlier-dropping loss. Taking the color loss
calculation as an example, instead of simply summing up
the color errors of all sampled rays at each iter... | Wonder3D |
those systems have human-like attributes such as intelligence and emotions.
In an age of information inflation and hyper-production of art, gaining attention has become one of the most important
principles of success. Considering the current hype around AI, it is understandable that framing the story behind a
particular... | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
generally find accuracy plateauing around 65% after five
rounds with little improvement thereafter. Of course, the
same caveats apply to any asymptotic guarantee. Finite sam-
ple results are rare in the RF literature, although there has
been some recent work in this area (Gao et al., 2022).
Another potential difficulty fo... | Adversarial Random Forests for Density Estimation and Generative Modeling |
sha1_base64="hP+6LrUf2d3tZaldqaQQvEKMXyw=">AAAB2XicbZDNSgMxFIXv1L86Vq1rN8EiuCozbnQpuHFZwbZCO5RM5k4bmskMyR2hDH0BF25EfC93vo3pz0JbDwQ+zknIvSculLQUBN9ebWd3b/+gfugfNfzjk9Nmo2fz0gjsilzl5jnmFpXU2CVJCp8LgzyLFfbj6f0i77+gsTLXTzQrMMr4WMtUCk7O6oyaraAdLMW2IVxDC9YaNb+GSS7KDDUJxa0dhEFBUcUNSaFw7g9LiwUXUz7GgUPNM7RRtRxzzi6dk7A0N+5oYkv39... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
Benegal, S. D., & Scruggs, L. A. (2018). Correcting misinformation about climate
change: The impact of partisanship in an experimental setting. Climatic Change,
148(1–2), 61–80. https://doi.org/10.1007/s10584-018-2192-4
Berinsky, A. J. (2012). Rumors, truths, and reality: A study of political misinformation.
Working ... | Social_Media_and_Democracy |
• Yesterday I dropped my clothes off at the dry cleaners and have yet to pick them up.
Where are my clothes? at my mom's house.
• There are six frogs on a log. Two leave, but three join. The number of frogs on the log is
now seventeen
In the first, GPT-2 correctly predicts the category of elements that follo... | The Next Decade in AI- |
Generative Agents
arXiv, April, 2023,
To achieve this, we perform a retrieval on the query “[name]’s
core characteristics.” We then summarize the descriptors in the
retrieved records by prompting the language model, for example:
How would one describe Eddy’s core characteristics
given the following statements?
- Edd... | Generative Agents- Interactive Simulacra of Human Behavior |
5
Figure 2: Overview of canonicalization and knowledge elicitation.
The essential part of canonicalization is to convert the raw data into a well-formed natural language
format, as shown in the left part of Figure 2. Other than unifying the solutions in diverse formats, a
crucial technique here is the discretization... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
Nynorsk1889Bengali1988Urdu1990Italian2145Polish2200Turkish2241Arabic2286Portuguese3620German4309French4481Hindi5438Spanish6693Russian7687Welsh8263Japanese8860Chinese11731Korean19938 Robust Speech Recognition via Large-Scale Weak Supervision | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
concerned with whether improvements (often from academic sources and proposed with evaluations
on small scales) translate to larger scales. In this work, we set aside the question of (up)scaling and
focus only on the limited compute. | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
[5] Sebastian B¨ock, Filip Korzeniowski, Jan Schl¨uter, Florian
Krebs, and Gerhard Widmer. madmom: a new Python Audio
and Music Signal Processing Library. In MM, 2016.
[6] Soravit Changpinyo, Piyush Sharma, Nan Ding, and Radu
Soricut. Conceptual 12m: Pushing web-scale image-text
pre-training to recognize long-tail vis... | VideoBackgroundMusicGeneration |
Mann. Bloomberggpt: A large language model for finance. arXiv preprint arXiv:2303.17564, 2023.
[118] Jian Yang, Shuming Ma, Haoyang Huang, Dongdong Zhang, Li Dong, Shaohan Huang, Alexandre Muzio, Saksham Singhal, Hany Hassan, Xia
Song, and Furu Wei. Multilingual machine translation systems from Microsoft for WMT21 sha... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
Regardless, | Social_Media_and_Democracy |
111:8
Trovato and Tobin, et al.
Table 2. Summary of evaluation on natural language processing tasks: NLU (Natural Language Under-
standing, including SA (Sentiment Analysis), TC (Text Classification), NLI (Natural Language Inference)
and other NLU tasks), Reasoning, NLG (Natural Language Generation, including Summ. (... | ASurveyonEvaluationofLargeLanguageModels |
davinci
text-davinci-002
text-davinci-003
code-davinci-002
-
-
-
-
80M T5-Small
Flan-T5-Small
36.4 31.8
27.3 27.3 50.0 42.9 25.0 31.2 45.5 36.4 31.0 34.5 43.8 25.0 12.5 25.0 18.2 36.4 27.3
9.1
27.3 57.1 28.6 62.5 56.2 63.6 72.7 51.7 55.2 68.8 43.8 12.5 37.5 63.6 36.4 54.5 36.4 63.6 54.5
18.2 36.4 50.0 57.1 62.5 62.5... | Mixture-of-Experts |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.