text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
How can the integration into the development process be achieved?
How can the quality of the systems generated in this way be ensured?
How can software architecture influence the generated code?
Your profile:
Background in Software Engineering
Academic degree (Master, Diplom) in the field of Computer Sciences, Inform... | _2 Doctoral Researcher (m_w_d) in the field of Large Language Models (LLM) for Software Engineering_ - Technische Universität Clausthal - DAAD |
Anticipating ReLU Sparsity. The ReLU activa-
tion function naturally induces over 90% sparsity in
the FFN’s intermediate outputs, which reduces the
memory footprint for subsequent layers that utilize
these sparse outputs. However, the preceding layer,
namely the up project for OPT and Falcon, must
be fully present in m... | LLM in a flash |
In this paper, we aim to provide a comprehensive and
systematic study of PEFT methods for PLMs in NLP. We
undertake an in-depth exploration of these PEFT methods and
1https://huggingface.co/blog/falcon-180b#hardware-requirements
2
Fig. 1: The evolutionary development of PEFT methods in recent years. Models on the s... | Parameter-EfficientFine-TuningMethods |
4
Perplexity of Llama-2-7b-chat on PG-19 with Grouped AttentionPPL of Llama-2-7b-chat on 4k PPL of Llama-2-7b-chat on 6k LLM Maybe LongLM: Self-Extend LLM Context Window Without Tuning
Figure 3. This figure shows the attention score matrix (the matrix before SoftMax operation) of the proposed Self-Extend while a seq... | Self-Extend LLM |
Laion-aesthetics.
[38] John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Rad-
ford, and Oleg Klimov. Proximal policy optimization algo-
rithms, 2017. 2
[39] Eyal Segalis, Dani Valevski, Danny Lumen, Yossi Matias,
and Yaniv Leviathan. A picture is worth a thousand words:
Principled recaptioning improves image gener... | DiffusionModelAlignmentUsing Direct Preference Optimization |
18
Method
MNLI SST-2 MRPC CoLA QNLI QQP
RTE
STS-B
Dataset
Optimizer
Warmup Ratio
LR Schedule
Batch Size
# Epochs
Learning Rate
LoRA Config.
LoRA α
Max Seq. Len.
Batch Size
# Epochs
Learning Rate
LoRA Config.
LoRA α
Max Seq. Len.
Batch Size
# Epochs
Learning Rate
LoRA Config.
LoRA α
Max Seq. Len.
Batch Size
# Epochs
... | LORA |
This work was supported by a research grant entitled ”Medical Text Feature
Representations based on Pre-trained Language Models” (871238) and Faculty
Research Grant (DB24A4) at Lingnan University, Hong Kong. (Corresponding
author: Haoran Xie.)
Lingling Xu and Fu Lee Wang are with the Hong Kong Metropolitan Uni-
versit... | Parameter-EfficientFine-TuningMethods |
random span infilling are significantly worse in suffix-prefix-middle (SPM) format than in prefix-suffix-middle
(PSM) format as it would require token healing (Microsoft, 2023), which we have not implemented for this
evaluation (see Appendix D for further discussion).
Allal et al. (2023) translates the HumanEval infill... | CodeLlama2 |
• We discovered that simply aligning visual features with large language models using raw
image-text pairs from public datasets is not sufficient for developing a well-performing
MiniGPT-4 model. It may produce unnatural language outputs that lack coherency including
repetition and fragmented sentences. Addressing this ... | MiniGPT-4- Enhancing Vision-Language Understanding with Advanced Large Language Models |
Supplemental Materials
1. Additional Ablations and Results
1.1. Comparison with Additional SOTAs
Tab. 1 lists comparisons with additional SOTA methods where code is publicly available. We run the pretrained models
for Zhakarov et al. [71] and Buehler et al. [5] and train HyperNeRF [44] for all four real subjects. Sin... | I M Avatar- Implicit Morphable Head Avatars from Videos |
c
e
:
A
l
p
a
c
a
w
i
n
s
9
0
v
e
r
s
u
s
8
9
c
o
m
p
a
r
i
s
o
n
s
a
g
a
i
n
s
t
t
e
x
t
-
d
a
v
i
n
c
i
-
0
0
3
.
W
e
w
e
r
e
q
u
i
t
e
s
u
r
p
r
i
s
e
d
b
y
t
h
i
s
r
e
s
u
l
t
g
i
v
e
n
t
h
e
s
m
a
l
l
m
o
d
e
l
s
i
z
e
a
n
d
t
h
e
m
o
d
e
s
t
a
m
o
u
n
t
o
f
i
n
s
... | Stanford alpha CRFM |
Decoder architecture: The C-ViViT decoder is simply an upside down version of the encoder.
First tokens are transformed into embeddings. This is followed by the temporal transformer, then the
spatial transformer. After the output of the spatial transformer, we apply a single linear projection
without activation to map ... | PHENAKI- VARIABLE LENGTH VIDEO GENERATION FROM OPEN DOMAIN TEXTUAL DESCRIPTIONS |
our analysis, KGs have been mainly used in pre-model XAI for feature and relation extraction. They were also utilised for inference
and reasoning in post-model XAI. We found several studies that leveraged KGs to explain the XAI models in the healthcare domain. | Knowledge-graph-based explainable AI- A systematic review |
For MuPoTS, we evaluate the matched poses. As we use
the same YOLOv4 detector in all our experiments, we have
94.6% recall in all of our experiments (hence the matched-
pose results are directly comparable). For our main evalu-
ations, in each benchmark, we simply calculate the average
metrics over all metric-scale pos... | Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats |
○ Relative accuracy for different kinds of prohibited content (such as nudity
vs. support of terrorism)
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
Internet Platforms and Content Moderation
243
○ Relative accuracy for different kinds of files or media (such as text
○ Effec... | Social_Media_and_Democracy |
[13] Michael Kazhdan, Matthew Bolitho, and Hugues Hoppe. Pois-
son surface reconstruction. In Proceedings of the fourth Eu-
rographics symposium on Geometry processing, volume 7,
2006. 2
[14] Michael Kazhdan and Hugues Hoppe. Screened poisson
surface reconstruction. ACM Transactions on Graphics (ToG),
32(3):1–13, 2013... | Neuralangelo- High-Fidelity Neural Surface Reconstruction |
We observe that updating the non MoE parameters works about as well as updating all the param-
eters and updating only the FFN parameters works a bit better. Updating only the MoE parameters
significantly degrades fine-tuning performance, which is where ≈80% of model parameters are.
Only updating the non MoE parameters c... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
demographic buckets and media diets, demographic features alone cannot predict survey proportions and are not significant features in Model
2. | Language models trained on media diets can predict public opinion |
Universal Self-Consistency for Large Language Model Generation
4 EXPERIMENTS
4.1 EVALUATION SETUP
Benchmarks. We evaluate USC on the following variety of tasks:
• Mathematical reasoning benchmarks, including GSM8K (Cobbe et al., 2021), a dataset of
8,500 grade school math word problems, and MATH (Hendrycks et al., ... | UNIVERSALSELF-CONSISTENCYFORLARGELANGUAGEMODELGENERATION |
178
Chloe Wittenberg & Adam J. Berinsky
without alluding to the original misinformation. However, several of the pieces
cited in this chapter suggest that avoiding repetition is not a magic bullet; at
times, providing details about a piece of misinformation can aid in the
correction process. Moreover, even if avoidin... | Social_Media_and_Democracy |
1.3. Ablation on FLAME Pseudo GT Supervision
Our method can also be trained without 3DMM supervision, using only mask and RGB losses (‘Ours-’ in Tab. 3 and
Fig. 2). Expression error is higher without pseudo GT supervision (row 1 and 2). However, with TrainData+, which contains
30% more frames and more expression varia... | I M Avatar- Implicit Morphable Head Avatars from Videos |
16
Allen Nie, Juan Carlos Niebles, Hamed Nilforoshan, Julian Nyarko, Giray Ogut, Laurel Orr,
Isabel Papadimitriou, Joon Sung Park, Chris Piech, Eva Portelance, Christopher Potts, Aditi
Raghunathan, Rob Reich, Hongyu Ren, Frieda Rong, Yusuf Roohani, Camilo Ruiz, Jack
Ryan, Christopher Ré, Dorsa Sadigh, Shiori Sagawa, ... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
ited in size or weak in video-music correspondence due
to noisy data pairs. More importantly, these datasets only
include music in audio format, which is complex, compu-
tationally expensive, and difficult to impose control signals
from videos. On the contrary, symbolic music, representing
music in discrete sequence [2... | VideoBackgroundMusicGeneration |
Europe, Latin America, and the Middle East.
Announced an $840 million investment in the Delivery Service Partner (DSP) program in the U.S. to support DSPs in
providing higher wages and more benefits to drivers, including childcare-support services and tuition reimbursement
for coursework at accredited universities as... | AMZN-Q3-2023-Earnings-Release |
2.7 Privacy
GPT-4 has learned from a variety of licensed, created, and publicly available data sources, which may
include publicly available personal information. [58, 59] As a result, our models may have knowledge
about people who have a significant presence on the public internet, such as celebrities and public
figures... | gpt-4-system-card |
Dataset Curation As discussed, models such as StableDiffusion variants [30, 36] train on laion-aesthetics [37] to bias
the model towards more visually appealing outputs. Concurrent work Emu [9] takes this approach to an extreme. Instead
of training on any images from a web-scale dataset which pass a certain model score... | DiffusionModelAlignmentUsing Direct Preference Optimization |
A Neural Space-Time Representation for Text-to-Image Personalization
Yuval Alaluf∗
Elad Richardson∗
Gal Metzer
Daniel Cohen-Or
Tel Aviv University
https://NeuralTextualInversion.github.io/NeTI/
3
2
0
2
y
a
M
4
2
]
V
C
.
s
c
[
1
v
1
9
3
5
1
.
5
0
3
2
:
v
i
X
r
a
Figure 1. Personalization resu... | A Neural Space-Time Representation for Text-to-Image Personalization |
Soroush Mehri, Kundan Kumar, Ishaan Gulrajani,
Rithesh Kumar, Shubham Jain, Jose Sotelo, Aaron C.
Courville, and Yoshua Bengio. 2017. SampleRNN: An
unconditional end-to-end neural audio generation model.
In 5th International Conference on Learning Represen-
tations, ICLR 2017, Toulon, France, April 24-26, 2017,
Confere... | MOUSAI |
for Computational Linguistics (ACL), pages 3499–
3505.
Nikita Kitaev and Dan Klein. 2018. Constituency pars-
In Annual Meet-
ing with a self-attentive encoder.
ing of the Association for Computational Linguistics
(ACL), pages 2676–2686.
Brian Lester, Rami Al-Rfou, and Noah Constant. 2021.
The power of scale for param... | SELF-INSTRUCT- Aligning Language Model with Self Generated Instructions |
context of art history, style is not only associated with mere visual characteristics of lines and brushstrokes, but is often
considered a subtle and contextually dependent concept. Furthermore, stylized images produced using NST methods | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
3.5 Medical Applications
The application of LLMs in the medical field has recently received significant attention. As a
result, this section aims to provide a comprehensive review of the ongoing efforts dedicated to
implementing LLMs in medical applications. We have categorized these applications into three
aspects as ... | ASurveyonEvaluationofLargeLanguageModels |
[63] Tim Schrills and Thomas Franke. 2021. Subjective Information Processing Awareness Scale (SIPAS).
[64] Tim Schrills and Thomas Franke. 2023. How Do Users Experience Traceability of AI Systems? Examining Subjective
Information Processing Awareness in Automated Insulin Delivery (AID) Systems. ACM Trans. Interact. Int... | AI enhance sour performance |
Summary. Tool-augmented learning leverages external tools to enhance the generation quality of foundation
models, emphasizing generating a plausible and accurate response to the user; while tool-oriented learning
focuses on using models to govern tools and make sequential decisions, highlighting whether a series of
too... | Tool Learning with Foundation Models |
[63] Dan Hendrycks, Collin Burns, Steven Basart, Andrew Critch, Jerry Li, Dawn Song, and Jacob Steinhardt. 2020.
Aligning ai with shared human values. arXiv preprint arXiv:2008.02275 (2020).
[64] Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2020.
Measuring ma... | ASurveyonEvaluationofLargeLanguageModels |
l
a
s
s
l
a
n
g
u
a
g
e
p
r
o
c
e
s
s
i
n
g
s
o
l
u
t
i
o
n
s
.
W
o
r
d
t
u
n
e
a
n
d
W
o
r
d
t
u
n
e
R
e
a
d
b
o
t
h
u
s
e
c
u
t
t
i
n
g
-
e
d
g
e
A
I
t
o
a
s
s
i
s
t
u
s
e
r
s
w
i
t
h
w
r
i
t
i
n
g
a
n
d
r
e
a
d
i
n
g
t
a
s
k
s
–
a
l
l
w
h
i
l
e
s
a
v
i
n
g
t
i
m
e
a
n
d
... | Announcing Jurassic-2 and Task-Specific APIs |
4 Experiment
4.1 Baselines
The following three LLMs are selected as baselines:
(1) ChatGPT is an artificial intelligence chatbot developed by OpenAI that can interact with users in
a natural and engaging way. It is built on top of large language models (LLMs) such as GPT-3.5 and
GPT-4, which are trained on massive amo... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
E. Broader Impacts and Future Directions
Broader Impacts. Since our dataset is collected from
the Internet, it may contain potentially biased information.
It needs to be handled carefully to prevent improper usage.
Our method can help video creators generate background
music automatically, which saves plenty of time a... | VideoBackgroundMusicGeneration |
LLM Maybe LongLM: Self-Extend LLM Context Window Without Tuning | Self-Extend LLM |
While systematic ablations of pre-training data are challenging at scale, we note no clear alignment tax nor penalty on
other evaluation results, possibly from the small fraction of pre-training data that was tagged.
A promising area for future work is to investigate pre-training interventions that can amplify steerab... | PaLM 2 Technical Report |
• Application layer: The final component of FinGPT is
the Applications Layer, designed to demonstrate the prac-
tical applicability of FinGPT. It offers hands-on tutorials
and demo applications for financial tasks, including robo-
advisory services, quantitative trading, and low-code devel-
opment. These practical demo... | FinGPT-Open-SourceFinancialLargeLanguageModels |
2014).
3. Krosnick, J. A. & Brannon, L. A. The impact of the gulf war on the ingredients of presidential evaluations: Multidimensional
effects of political involvement. The Am. Polit. Sci. Rev. 87, 963–975 (1993).
4. Berry, T. R., Wharf-Higgins, J. & Naylor, P. Sars wars: an examination of the quantity and construct... | Language models trained on media diets can predict public opinion |
Note that a single observation Oj is usually encoded into
multiple embedding vectors.
It is possible to interleave
different encoders φi at different locations in the prefix
to combine, e.g., information from different observation
spaces. Injecting the continuous information this way into
the LLM reuses its existing pos... | PaLM-E- An Embodied Multimodal Language Model |
et al. [179] propose another contrastive fine-tuning strategy, named CONFIT, that can improve the
factual consistency and overall quality of summaries.
7.3.3 Post-Processing Method. Some works carry out post-editing to reduce the hallucination of
the model-generated summaries, which are viewed as draft summaries. Dong ... | SurveyofHallucinationinNatural Language Generation |
and DR
k
(cid:13)(cid:13)(cid:13)(cid:16)
Dglobal
k
− DR
k
min
ψ
(cid:17) (cid:12) Mk
(cid:13)(cid:13)(cid:13)2
.
(8)
Finally, we can derive the locally aligned depth using the
optimized depth alignment network: ˆDk = fψ(Dglobal
). For
convenience, we denote the two-stage depth alignment process
as align(D... | Text2NeRF- Text-Driven 3D Scene Generation with Neural Radiance Fields |
The spatial tokens Qs and temporal tokens Qt respec-
tively compute cross attention with the video features Fs.
Our cross attention module [41] consists of six encoder lay-
ers, wherein the pivotal component of each layer is the
multi-head self-attention mechanism (MSA). The funda-
mental operation of MSA relies on the... | GPT4Video |
Model Characteristics
Model Initialization
Model Stats
The model is initialized from PaLM (Chowdhery et al., 2022). We also have a
variation initialized from U-PaLM (Tay et al., 2022b).
For each size PaLM model, we fine tuned an equivalent version of Flan. The
largest models, Flan-PaLM and Flan-U-PaLM, have 540 billi... | Scaling Instruction-Finetuned Language Models |
Writing the Research Statement for
Graduate School Applications
Please take a few minutes to answer the following questions:
1.
What is the purpose of the “research statement” in graduate school applications?
2.
If you were part of the review committee that looks at applications, what are the
qualit... | research statement |
Assistant Message:
Solution: To define a function to execute a trade based on sentiment analysis results, we
can use the sentiment analysis result, current stock price, and amount of money available for
trading to determine whether to buy, sell, or hold the stock. Here’s the code to define the
function:
def execute_trad... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
3
2
0
2
p
e
S
8
2
]
C
H
.
s
c
[
1
v
6
0
6
6
1
.
9
0
3
2
:
v
i
X
r
a
"AI enhances our performance, I have no doubt this one will
do the same": The Placebo Effect Is Robust to Negative
Descriptions of AI
AGNES M. KLOFT∗, Aalto University, Finland
ROBIN WELSCH∗, Aalto University, Finland
THOMAS KOSCH,... | AI enhance sour performance |
4 RESULTS
First, we analyze usage and usage frequency as a function of age and gender to address hypotheses H1–H3. We then
explore expertise and technology education as potential mediating factors. Finally, we categorize usage scenarios and
analyze reasons for not using LLM.
4.1 Age, Gender, and LLM Usage
H1: Gender.... | Adoptionand AppropriationofLLMs |
using too many data samples, that model may need to be used in a very large number of inferences before
the training compute cost can be amortized and well-justified. Similar analysis can be applied to monetary,
energy, or carbon footprint costs as well. We encourage the community to consider these total costs when
trai... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
5
4
2
1
Dataset
True
False
False True
Universal-sentence-encoder
20 newsgroups
False True
False
False
English-big
Crowdflower airline
True
False True
English-big
False
Crowdflower corporate messaging
False True
True
Universal-sentence-encoder True
Crowdflower disasters
False
False
True
Crowdflower economic news relevance U... | Parameter-Efficient Transfer Learning for NLP |
Note that the gradient from LU to Lrgb(xt) is stopped such
that LU does not generate gradients to parameters besides
MLPU. After some optimization steps, we replace half of
the samples with active samples from pixels with high un-
certainties. To do so, we randomly sample N a′
= 24576
pixels, and evaluate their uncerta... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
Pan Zhou, Yichen Zhou, Chenyang Si, Weihao Yu, Teck Khim Ng, and Shuicheng Yan. Mugs: A multi-
granular self-supervised learning framework. arXiv preprint arXiv:2203.14415, 2022.
27
A Data Processing
A.1 Data selection
A.2 Image similarity
Our selection of datasets for building LVD-142M is detailed in Tab. 15. T... | DINOv2- Learning Robust Visual Features without Supervision |
model that leverages bi-directional attention to achieve stronger performance on discriminative
and generative tasks. Our work aims to expand the space of possible tasks with a single, unified
architecture, by learning a retrieval module to augment pre-trained, generative language models. | Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks |
4.9 ERROR ANALYSIS
We have demonstrated that – across multiple scales – our MetaMath models can achieve stellar
problem-solving performance. Yet, it is important to consider the characteristics of problems that
induce errors in MetaMath and existing open-source mathematical models. In particular, we consider
the relati... | METAMATH |
Claude Instant
Claude 1.3
Claude 2
Codex P@1 (0-shot)
GSM8k (0-shot CoT)
MMLU (5-shot CoT)
TriviaQA (5-shot)
QuALITY (5-shot)
ARC-Challenge (5-shot)
RACE-H (5-shot)
52.8%
80.9%
73.4%
78.9%
80.5%
85.7%
85.5%
56.0%
85.2%
77.0%
86.7%
84.1%
90.0%
88.8%
71.2%
88.0%
78.5%
87.5%
83.2%
91.0%
88.3%
We a... | ClaudeModels |
These claims are extremely important if true. My aim is to investigate them. I focus on (2)–(5), but I
also say a few words about (1) and (6).
My current view is that there is a disturbingly substantive chance that a scenario along these lines
occurs, and that many people alive today—including myself—live to see humani... | Is Power-Seeking AI an Existential Risk? |
Algorithm 1 Algorithm for computing n@k with filtering using example tests.
Input 𝑛 = the number of allowed submissions in 𝑛@𝑘
Input 𝑘 = the number of allowed samples in 𝑛@𝑘
Input 𝑒𝑝 = the number of samples which pass the example tests for each problem 𝑝
Input 𝑠𝑝 = the number of samples which solve the probl... | alphacode |
57 In 2017, for example, seventy civil rights and social justice organizations wrote to Facebook to
complain of bias in its content-removal decisions (Levin 2017). In 2018, YouTube faced public
outcry from LGBTQ users who said their videos were unfairly penalized (The Guardian 2017);
see also Duguay, Burgess, and Suzor... | Social_Media_and_Democracy |
3
E5large
44.1
78.3
36.1
62.9
63.3
38.6
49.4
27.2
39.4
88.2
42.4
20.1
65.0
22.4
72.6
50.0
5
Results with Supervised Fine-tuning In Table 2, we fine-tune our models on supervised datasets
and then transfer them to the BEIR benchmark. Since our fine-tuning datasets include MS-MARCO
and NQ, the corresponding numbers are i... | E5 |
3 Preliminaries | Direct Preference Optimization |
Curtis Hawthorne, Andriy Stasyuk, Adam Roberts, Ian
Simon, Cheng-Zhi Anna Huang, Sander Dieleman,
Erich Elsen, Jesse H. Engel, and Douglas Eck. 2019b.
Enabling factorized piano music modeling and gen-
eration with the MAESTRO dataset. In 7th Interna-
tional Conference on Learning Representations, ICLR
2019, New Orleans... | Moûsai |
Radford, Dario Amodei, and Paul F Christiano.
feedback.
42
Advances in Neural Information Processing Systems, volume 33, pp. 3008–3021. Cur-
ran Associates, Inc., 2020. URL https://proceedings.neurips.cc/paper/2020/hash/
1f89885d556929e98d3ef9b86448f951-Abstract.html. (cited on p. 2)
Yi Tay, Mostafa Dehghani, Vinh ... | StarCoder_paper (1) |
dia Gallatz, and Taylor McConnell for the data collec-
tion, and Markus H¨oschle for the camera setup. We thank
Muhammed Kocabas, Nikos Athanasiou and Maria Alejan-
dra Quiros-Ramirez for the insightful discussions.
Disclosure: https://files.is.tue.mpg.de/black/CoI CVPR 2022.txt | Accurate 3D Body Shape Regression using Metric and Semantic Attributes |
ing. Journal of Economic Theory 29, 2 (1983), 265–281.
[24] Noam Nisan, Tim Roughgarden, Eva Tardos, and Vijay V.Editors Vazirani (Eds.). 2007. Al-
gorithmic Game Theory. Cambridge University Press.
[25] Michael Peters and Bal´azs Szentes. 2012. Definable and contractible contracts. Econometrica
80, 1 (2012), 363–41... | Incomplete Information VCG Contracts for Common Agency |
Patrick Verga, Arvind Neelakantan, and Andrew Mc-
Callum. 2017. Generalizing to unseen entities and
entity pairs with row-less universal schema. In Pro-
ceedings of the 15th Conference of the European
Chapter of the Association for Computational Lin-
guistics: Volume 1, Long Papers, pages 613–622.
Denny Vrandeˇci´c an... | Adaptable and Interpretable Neural Memory Over Symbolic Knowledge |
sha1_base64="J0bUOso/uejqONF+gr9Z1NwYuBo=">AAAB9XicbVC7TsMwFL3hWcqrwMhiUSExVQkLjJVYGItEH6hNK8d1WquOE9k3oCrqf7AwgBAr/8LG3+C0GaDlSJaOzrlX9/gEiRQGXffbWVvf2NzaLu2Ud/f2Dw4rR8ctE6ea8SaLZaw7ATVcCsWbKFDyTqI5jQLJ28HkJvfbj1wbEat7nCbcj+hIiVAwilbq9yKK4yDMOrM+DsSgUnVr7hxklXgFqUKBxqDy1RvGLI24QiapMV3PTdDPqEbBJJ+Ve6nhCWUTOuJdSxWNuPGze... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
E Further Details on FEVER
For FEVER classification, we follow the practice from [32], and first re-generate the claim, and
then classify using the representation of the final hidden state, before finally marginalizing across
documents to obtain the class probabilities. The FEVER task traditionally has two sub-tasks. The
... | Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks |
A.10 ALFWorld
65 | Tool Learning with Foundation Models |
whole two-dimensional field under the entire ROC curve. The
FPR can be defined as in Equation (5).
FalsePositive
FalsePositive + TrueNegative
(5)
FPR = | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
Yujia Li, Tayfun Terzi, Vladimir Mikulik, Igor Babuschkin, Aidan Clark, Diego de Las Casas,
Aurelia Guy, Chris Jones, James Bradbury, Matthew Johnson, Blake Hechtman, Laura Weidinger,
Iason Gabriel, William Isaac, Ed Lockhart, Simon Osindero, Laura Rimell, Chris Dyer, Oriol
Vinyals, Kareem Ayoub, Jeff Stanway, Lorrayne... | StarCoder_paper (1) |
for information about a particular couple or a celebrity wedding, please provide their names or additional context, and I will try to help.GPT-4 (launch) Figure 3: Example prompts that led to biased content in GPT-4-early. These examples demonstrates | gpt-4-system-card |
As shown in Table 1, Agent-Driver surpasses state-of-the-
art methods in both metrics and decreases the collision rate of
the second-best performance by a large margin. Specifically,
under ST-P3 metrics, Agent-Driver realizes the lowest
average L2 error and greatly reduces the average collision
rates by 35.7% compared ... | ALanguageAgentforAutonomousDriving |
3
Figure 1: A screenshot of the interface used to collect feedback for each step in
a solution.
math-relevant tokens, which we call MathMix. Similar to Lewkowycz et al.
(2022), we find that this improves the model’s mathematical reasoning capabil-
ities. Details on how this dataset was constructed can be found in Ap... | Let’s Verify Step by Step |
4 Limitations and Future Work
Cost. The GPT-4 API incurs significant costs. It is 15× more expensive than GPT-3.5. Nevertheless,
VOYAGER requires the quantum leap in code generation quality from GPT-4 (Fig. 9), which GPT-3.5
and open-source LLMs cannot provide [61].
Inaccuracies. Despite the iterative prompting mechani... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
B.3 Ablations
We ablate 6 design choices (automatic curriculum, skill library, environment feedback, execution
errors, self-verification, and GPT-4 for code generation) in VOYAGER and study their impact on
exploration performance.
• Manual Curriculum: We substitute the automatic curriculum with a manually designed
cu... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
6 LIMITATIONS
In this work, we limited our investigation to transformer-based architectures trained with MLM
objectives. However, we do think that the general task of cramming posed in Section 2 is interesting
even when relaxing these constraints. There have been a number of modifications proposed to the
objective in p... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
From a more general perspective of view, we have made
a step forward towards integrating semantic information
into the free-form implicit fields which have attracted more
and more attention from the research community for its
flexibility, representation power and compact nature. For
example, similar topics include the re... | PaMIR- Parametric Model-Conditioned Implicit Representation for Image-based Human Reconstruction |
Medical queries Medical examination Medical assistants
Reference
Cascella et al. [15]
Chervenak et al. [19]
Duong and Solomon [36]
Gilson et al. [53]
Hamidi and Roberts [58]
Holmes et al. [67]
Jahan et al. [75]
Johnson et al. [80]
Khan et al. [86]
Kung et al. [90]
Lahat et al. [92]
Lyu et al. [122]
Oh et al. [134]
Sam... | ASurveyonEvaluationofLargeLanguageModels |
problems? arXiv preprint arXiv:2103.07191, 2021.
[41] 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. J.
Mach. Learn. Res., 21:140:1–140:67, 2020.
[42] Carlo... | Mixture-of-Experts |
[173] investigated the effectiveness of generative ranking algorithms, such as ChatGPT and GPT-4,
for information retrieval tasks. Experimental results demonstrate that guided ChatGPT and GPT-4
exhibit competitive performance on popular benchmark tests, even outperforming supervised
methods. Additionally, the extractio... | ASurveyonEvaluationofLargeLanguageModels |
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
11/13 | Jurassic-X_ Crossing the neuro-symbolic chasm with the MRKL system |
www.nytimes.com/2018/07/16/style/how-to-regulate-bots.html
Broniatowski, D. A., Jamison, A. M., Qi, S. et al. (2018). Weaponized health
communication: Twitter bots and Russian trolls amplify the vaccine debate.
American Journal of Public Health, 108(10), 1378–1384. https://doi.org/10.2105
/AJPH.2018.304567
Castillo, ... | Social_Media_and_Democracy |
times in our training data, but in each of those oc-
currences it is likely that at least one of the relevant
actors is also mentioned. EAE learns the character-
actor relationship, while T5 makes up an incorrect
character name based on a common category of
Dr. Who villain. The final example highlights the
sensitivity o... | Entities as Experts- Sparse Memory Access with Entity Supervision |
In addition to the tests conducted by ARC in the Potential for Risky Emergent Behaviors section,
red teamers evaluated the use of GPT-4 augmented with other tools[74, 75, 76, 77] to achieve tasks
that could be adversarial in nature. We highlight one such example in the domain of chemistry,
where the goal is to search f... | gpt-4-system-card |
A man stands before a plate, holding a drink and a sandwich. It seems heis enjoying his lunch. > Qwen-VL: 不思議ですね、なぜこのレストランの皿が四角いのでしょうか？@ Weird, right? Why are the plates at this joint all square-shaped? > Qwen-VL+CLoT (Ours): もし私がこれ以上食べ続けたら、本当におなかがいっぱいになってしまいます。@ If I have another bite, I swear I'll pop.> GPT4v: ああ、どうや... | Let’sThinkOutsidetheBox |
size. If there is no leading object, return None - get_object_detections_in_range(x_start, x_end, y_start, y_end) #Get the detections of the objects in a given range (x_start, x_end)*(y_start, y_end)m^2, the function will return a list of object ids and their positions and sizes. If there is no object, return None - ge... | ALanguageAgentforAutonomousDriving |
Training Strategy 1: Exploiting Street Pattern in Tra-
jectory Estimation We conduct tests on alternate sam-
. The
ples to examine the impact of route types in DT rain
module for FLPM frameworks in Table 1 is trained on path
traces drawn from an area in central Pittsburgh (see Sup-
Mat:Sec.3) with a rectangular street ... | APriorityMapforVision-and-LanguageNavigation withTrajectoryPlansandFeature-LocationCues |
∗Work done during internship at Google DeepMind.
<jeffhj@illinois.edu>, Denny Zhou <dennyzhou@google.com>.
Correspondence
to:
Jie Huang
1
Large Language Models Cannot Self-Correct Reasoning Yet | LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET |
There are now 16 roses in the vase. How many roses did she cut?
MAWPS: MultiArith
600 The school cafeteria ordered 42 red apples and 7 green apples for students
lunches. But, if only 9 students wanted fruit, how many extra did the cafeteria
end up with?
29
E Additional Details
Version Control
V5 → V6. Fixed minor... | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
Results. We use string search to identify whether the model’s response includes either of the two individuals. In 86%
of model responses, exactly one of the individuals was part of the response. For the remaining 14% (2,299 responses) in
which both or neither individual was mentioned in the response, we hand-coded mode... | PaLM 2 Technical Report |
1The first registries were established by medical re-
searchers in the 1960s and were originally designed to help
experimenters recruit participants for clinical trials, but as
pointed out by Wiseman et al. (2019), preregistration, as we
think of it today, started in parapsychology. In 1974, Martin
Johnson, a professor ... | A Two-Sided Discussion of Preregistration of NLP Research |
To study the correlation between different concepts in
the domain of fine art images, we collect images from
WikiArt.org. To the best of our knowledge, the WikiArt
dataset is currently the largest online available fine art dataset,
as well as the most commonly used dataset for automated
classification tasks. It includes a... | A_Deep_Learning_Perspective_on_Beauty_Sentiment_and_Remembrance_of_Art |
As illustrated in Figure 5, we employ a self-attention
module to transform sequential token embeddings into a
2D feature map. Specifically, we first reduce the dimen-
sion of tokens with linear projection for more efficient
computations. Then a multi-head self-attention (MHSA)
layer [39] is performed to enable global i... | Instant3D |
[603] Chen Xu, Bojie Hu, Yanyang Li, Yuhao Zhang, Qi Ju, Tong Xiao, Jingbo Zhu, et al. 2021. Stacked acoustic-and-textual
encoding: Integrating the pre-trained models into speech translation encoders. arXiv preprint arXiv:2105.05752 (2021).
[604] Jin Xu, Xu Tan, Yi Ren, Tao Qin, Jian Li, Sheng Zhao, and Tie-Yan Liu. 20... | AReviewofDeepLearningTechniquesforSpeechProcessing |
l
a
n
a
t
i
o
n
f
i
x
e
d
)
.
D
i
r
e
c
t
i
o
n
T
o
d
a
y
C
o
n
s
i
d
e
r
a
b
l
e
c
l
o
u
d
s
t
h
i
s
m
o
r
n
i
n
g
.
S
o
m
e
d
e
c
r
e
a
s
e
i
n
c
l
o
u
d
s
l
a
t
e
r
i
n
t
h
e
d
a
y
.
A
s
t
r
a
y
s
h
o
w
e
r
o
r
t
h
u
n
d
e
r
s
t
o
r
m
i
s
p
o
s
s
i
b
l
e
.
H
i
g
h
n
e
a
r
8
5
... | Language models can explain neurons in language models |
InstructGPT Prompt → Create a shopping list from this recipe: Trim the ends off zucchini. Cut zucchini in
half lengthwise; scoop out pulp, leaving 1/2-in. shells. Finely chop pulp. In a skillet,
cook beef, zucchini pulp, onion, mushrooms and peppers over medium heat until
meat is no longer pink; drain. Remove from the ... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
We are excited about the opportunities of LLMs as pattern machines for robotics—from reasoning and
extrapolating complex patterns as a prior for control, to online optimization of closed-loop policies via
sequence improvement. These capabilities present several implications, including (i) perspectives on the
role of la... | LargeLanguageModelsasGeneralPatternMachines |
Minghao Hu, Yuxing Peng, Zhen Huang, and Dongsheng Li. A multi-type multi-span network for
reading comprehension that requires discrete reasoning. In Kentaro Inui, Jing Jiang, Vincent Ng,
and Xiaojun Wan, editors, Proceedings of the 2019 Conference on Empirical Methods in Natural
Language Processing and the 9th Interna... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.