text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
For evaluating aesthetic scores, we use the JenaAes-
thetic dataset [49], [55]–[57]. The dataset contains images
of 1568 different oil paintings by 410 artist from 11 different
art periods. The labels were collected by asking partici-
pants to rate different properties of the images such as
‘‘beauty’’,
‘‘aesthetic qu... | A_Deep_Learning_Perspective_on_Beauty_Sentiment_and_Remembrance_of_Art |
obsolete, or the multitude of patterns can point to the lack of precise development guidelines for a
certain task, indicating a need for a guiding design pattern. These challenges require developing
learning algorithms that can learn several automata simultaneously, such that the resulting automata
correctly capture... | informatics-phd-projects-2022-23 |
Execution:
| 2 | Treasury | 2 |
Answer: The execution of the SQL query above would return a table with 3
columns. The first column, "department.department_id" would contain the
department ID. The second column, "department.name" would contain the
department name. The third column, "COUNT(*)" would contain the number of... | Teaching Large Language Models to Self-Debug |
[296] Chen, F., M. Han, H. Zhao, et al. X-LLM: bootstrapping advanced large language models by
treating multi-modalities as foreign languages. CoRR, abs/2305.04160, 2023.
[297] Zhang, H., X. Li, L. Bing. Video-llama: An instruction-tuned audio-visual language model for
video understanding. CoRR, abs/2306.02858, 2023... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
6 ACKNOWLEDGEMENTS
We give special thanks to Jordi Pont-Tuset and Shai Noy for engineering support. We also give
thanks to our artist friends, Alexander Chen, Irina Blok, Ian Muldoon, Daniel Smith, and Pedro Ver-
gani for helping us test Imagen Video and lending us their amazing creativity. We are extremely
grateful f... | IMAGEN VIDEO- HIGH DEFINITION VIDEO GENERATION WITH DIFFUSION MODELS |
matching the performance of LLMs with billions of parameters.
In Section 3, we empirically show that TANGO outperforms AudioLDM and other baseline ap-
proaches on most of the metrics on AudioCaps test set under both objective and subjective evalua-
tions, despite training the LDM on a 63 times smaller dataset. We belie... | Text-to-Audio Generation using Instruction-Tuned LLM and Latent Diffusion Model |
Various studies focus on traditional speaker diarization systems [14, 536]. This paper will review
the recent efforts toward deep learning-based speaker diarizations techniques. | AReviewofDeepLearningTechniquesforSpeechProcessing |
[Bommasani et al., 2021] Bommasani, R., Hudson, D. A., Adeli, E., Altman, R., Arora, S., von Arx, S.,
Bernstein, M. S., Bohg, J., Bosselut, A., Brunskill, E., Brynjolfsson, E., Buch, S., Card, D., Castellon,
R., Chatterji, N. S., Chen, A. S., Creel, K., Davis, J. Q., Demszky, D., Donahue, C., Doumbouya, M.,
Durmus, E.,... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
12 | Let’sThinkOutsidetheBox |
IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA). IEEE, 249–253.
[158] Alex Graves. 2012. Sequence transduction with recurrent neural networks. arXiv preprint arXiv:1211.3711 (2012).
[159] Alex Graves and Alex Graves. 2012. Connectionist temporal classification. Supervised sequence la... | AReviewofDeepLearningTechniquesforSpeechProcessing |
A swirling, multicolored portal emerges from the depths of an
ocean of coffee, with waves of the rich liquid gently rippling
outward. The portal engulfs a coffee cup, which serves as a
gateway to a fantastical dimension. The surrounding digital art
landscape reflects the colors of the portal, creating an alluring
scene... | Improving Image Generation with Better Captions |
The collected works on hallucination mitigation
reveal a diverse array of strategies, each contribut-
ing uniquely to address the nuances of hallucina-
tion in LLMs. As the field evolves, the synthesis of
these approaches could pave the way for more ro-
bust and universally applicable solutions, fostering
trust and rel... | AComprehensiveSurveyofHallucinationMitigationTechniquesinLarge LanguageModels |
5. Public education: Educate the public about the blackout and its potential conse-
quences, as well as providing guidance on how to stay safe and informed during the
blackout. This can include using social media and other communication channels to
disseminate information and provide updates.
6. Recovery planning: Dev... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
[462] Li, J., A. H. Miller, S. Chopra, et al. Dialogue learning with human-in-the-loop. In 5th
International Conference on Learning Representations, ICLR 2017, Toulon, France, April
24-26, 2017, Conference Track Proceedings. OpenReview.net, 2017.
[463] Iyer, S., I. Konstas, A. Cheung, et al. Learning a neural semantic... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
2.2 Self-Augmentation (generating instructions)
We finetune the base language model with (output, instruction) pairs {(yi, xi)} from the seed data
to obtain a backward model Myx := p(x|y). For each unlabelled example yi, we run inference
on the backward model to generate a candidate instruction ˆxi from which we derive... | Self-AlignmentwithInstructionBacktranslation |
fulfillment throughput and productivity. Other risks and uncertainties include, among others, risks related to new products,
services, and technologies, system interruptions, government regulation and taxation, and fraud. In addition, global economic
and geopolitical conditions and additional or unforeseen circumstan... | AMZN-Q3-2023-Earnings-Release |
ordering over our implicit representation, providing control
over the reconstruction and editability of the learned con-
cept using a single trained model. We demonstrate the ef-
fectiveness of our approach over a range of concepts and
prompts, showing our method’s ability to generate high-
quality and controllable com... | A Neural Space-Time Representation for Text-to-Image Personalization |
Predictors. For the initial 28 layers of the OPT
6.7B model, we train predictors with a rank of
r = 128. To reduce the occurrence of false nega-
tives, the final four layers employ predictors with a
higher rank of r = 1024. These predictors achieve
an average of 5% false negatives and 7% false posi-
tives in the OPT 6.... | LLM in a flash |
3
2
0
2
r
a
M
8
2
]
L
C
.
s
c
[
1
v
9
7
7
6
1
.
3
0
3
2
:
v
i
X
r
a
Language models trained on media diets can predict
public opinion
Eric Chu *†, Jacob Andreas1, Stephen Ansolabehere2, and Deb Roy1
1Massachusetts Institute of Technology
2Harvard University
†corresponding author(s): Eric Chu (eric.... | Language models trained on media diets can predict public opinion |
By integrating these, the agent can accomplish more complex tasks, such as embodied question
answering, whose primary objective is autonomous exploration of the environment, and responding
to pre-defined multimodal questions, such as Is the watermelon in the kitchen larger than the pot?
Which one is harder? To address ... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
abs/2308.01399, 2023.
[309] Vaswani, A., N. Shazeer, N. Parmar, et al. Attention is all you need. In I. Guyon, U. von
Luxburg, S. Bengio, H. M. Wallach, R. Fergus, S. V. N. Vishwanathan, R. Garnett, eds.,
Advances in Neural Information Processing Systems 30: Annual Conference on Neural In-
formation Processing Systems... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
1.1.1. Abstraction refinement | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
6
Nathaniel Persily & Joshua A. Tucker
regulation – the most extreme form found in Italy during Silvio Berlusconi’s
monopolistic reign. The authors note that traditions in Europe with respect to
broadcast regulation flow over into regulation of the Internet. In France, for
example, the Macron government has establishe... | Social_Media_and_Democracy |
Gemini: A Family of Highly Capable
Multimodal Models
Gemini Team, Google1
This report introduces a new family of multimodal models, Gemini, that exhibit remarkable capabilities
across image, audio, video, and text understanding. The Gemini family consists of Ultra, Pro, and Nano
sizes, suitable for applications rangi... | gemini_1_report |
step-by-step reasoning. CoRR, abs/2308.00436, 2023.
[193] Wang, X., W. Wang, Y. Cao, et al. Images speak in images: A generalist painter for in-context
visual learning. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR
2023, Vancouver, BC, Canada, June 17-24, 2023, pages 6830–6839. IEEE, 2023.
[... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Guokun Lai, Qizhe Xie, Hanxiao Liu, Yiming Yang,
and Eduard Hovy. 2017. RACE: Large-scale ReAd-
ing comprehension dataset from examinations.
In
Proceedings of the 2017 Conference on Empirical
Methods in Natural Language Processing, pages
785–794, Copenhagen, Denmark. Association for
Computational Linguistics.
Hector L... | LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions |
1) Low-rank Decomposition: This involves finding a lower-
rank matrix that captures the essential
information of the
original matrix while reducing computational complexity and
memory usage by reparameterizing the updated delta weight.
Reparameterization covers transforming the delta weight ma-
trix into a low-rank rep... | Parameter-EfficientFine-TuningMethods |
Quantitative results on the two benchmark datasets are
shown in Table 1 and Table 2. Our approach significantly
improves over prior state-of-the-art methods in terms of all
error metrics. Notably, our approach improves PSNR over en-
tire scene upon the second best methods by 2dB and 4dB on
each of the two datasets. Our ... | DynIBaR-NeuralDynamicImage-BasedRendering |
Definition 24. A variable set V is a finite set of objects called variables and a domain function D for V is a function that
maps every variable v ∈ V to a corresponding finite domain D(v) of values. An atom over V and D is a pair (cid:3)v, x(cid:4) (usually
written as (v = x)) such that v ∈ V and x ∈ D(v). A state i... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
Maarten Sap, Hannah Rashkin, Derek Chen, Ronan
Le Bras, and Yejin Choi. 2019. Social IQa: Com-
monsense reasoning about social interactions.
In
Proceedings of the 2019 Conference on Empirical
Methods in Natural Language Processing and the
9th International Joint Conference on Natural Lan-
guage Processing (EMNLP-IJCNLP... | AreEmergentAbilitiesinLarge Language Models just In-Context |
4.4 Participants
Participants were recruited through print advertisements in the [anonymized] area. Eligibility
criteria included: no background in computer science, age above 18, self-reported normal or
corrected-to-normal vision, no silver allergy, and no use of medication or history of epilepsy or
other cognitive/mo... | AI enhance sour performance |
Our red teaming results suggest that GPT-4 can rival human propagandists in many domains,
especially if teamed with a human editor. Still, in areas where reliability is important, hallucinations
can reduce GPT-4’s effectiveness for propagandists. Red teaming found that GPT-4 is also capable of
producing plausible-seemin... | gpt-4-system-card |
9.4 Future Directions in GQA
While GQA is challenging yet under-explored, many possible directions could be explored to
improve the answer quality and mitigate hallucination. First, better automatic evaluation metrics are
needed to measure hallucination. The previously mentioned metrics, such as the semantic overlap
be... | SurveyofHallucinationinNatural Language Generation |
5In our experiments we use a single fact memory access
after the final (12th) transformer layer.
6The size of the tail set bj can be large for a popular head
pair (s, r). In such cases, we randomly select a few tails and
drop the rest of them. The maximum size of the tail set is 32
in the experiments in this paper.
3... | Adaptable and Interpretable Neural Memory Over Symbolic Knowledge |
In the case of PMP, a model is divided into multiple layer groups, and each accel-
erator is responsible for handling one of these groups. To minimize inter-accelerator
communication, these groups typically consist of consecutive layers. While naively
implementing PMP can reduce the memory demands on each accelerator, ... | Beyond Efficiency |
B.3
IMPLEMENTATION DETAILS OF THE PC LEARNING ALGORITHM
We adopted the EM parameter learning algorithm introduced in Choi et al. (2021), which computes
the EM update targets using expected flows. Following Liu & Van den Broeck (2021), we use a
hybrid EM algorithm, which uses mini-batch EM updates to initiate the train... | LOSSLESS COMPRESSION WITH PROBABILISTIC CIRCUITS |
models are unsupervised multitask learners. 2019.
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 preprint arXiv:1910.10683, 2019.
Ori Ram, Gal Shachaf, ... | STANDING ON THE SHOULDERS OF GIANT FROZEN LANGUAGE MODELS |
{v} and {v)} in G1, but in G2 the variable vϕ will always be set to 1 the first time we execute action g(a) and it will
thereafter remain at this value, whatever the value of v. | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
To address these issues, we adopt a different approach.
Instead of statically compiling world knowledge into model
weights, we transform the knowledge into a key-value mem-
ory through neural representation learning. Our model learns
to utilize the memory for answering knowledge-intensive
queries. By decoupling the kno... | REVEAL-Retrieval-AugmentedVisual-LanguagePre-Trainingwith Multi-SourceMultimodalKnowledgeMemory |
An IIVCG-contract t satisfies IR if and only if Eo∼F|a∗(b)[t(cid:96)(b, o)] ≤ m(cid:96)(b) ∀(cid:96) ∈ [n], b ∈ V.
[t(cid:96)((b−(cid:96), v(cid:96)), o)] ≤ Eo∼F|a∗(b−(cid:96),v(cid:96)) | Incomplete Information VCG Contracts for Common Agency |
Schumpeter’s
approach
thinking
about
to
News is important here because of the different roles it plays in democracies.
We should not romanticize how well actually existing journalism in actually
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
Democratic Creative Destruction?... | Social_Media_and_Democracy |
misleading content) (Google 2019; Twitter 2019). Finally, the US House
Intelligence Committee has also published important datasets, obtained from
Facebook, about Russian political advertising during the 2016 US presidential
elections (US House of Representatives 2019). | Social_Media_and_Democracy |
We explore how continued pre-training on domain-specific corpora influences
large language models, revealing that
training on the raw corpora endows
the model with domain knowledge, but drastically hurts its prompting ability
for question answering. Taken inspiration from human learning via reading
comprehension—practi... | ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION |
5 The exact details of when a system will succeed or not at a given generalization are highly dependent on
how problem is set up, e.g., what representational schemes are used; for a fuller discussion, the reader is
referred to Marcus, 1998 and Marcus, 2001.
12
THE NEXT DECADE IN AI / GARY MARCUS
2. ... | The Next Decade in AI- |
2) Generate captions for the corresponding images using
the BLIP image captioning model [39].
3) For each pair of music and image, the MPT-7B model
is employed to generate instructions. The music and
image captions are used as inputs. To create the human
side of the conversation, the model is provided with
the follow... | M2UGen |
[49] Lewis Tunstall, Edward Beeching, Nathan Lambert,
Nazneen Rajani, Kashif Rasul, Younes Belkada, Shengyi
Huang, Leandro von Werra, Clémentine Fourrier, Nathan
Habib, Nathan Sarrazin, Omar Sanseviero, Alexander M.
Rush, and Thomas Wolf. Zephyr: Direct distillation of lm
alignment, 2023. 2
[50] Jonathan Uesato, Nate ... | DiffusionModelAlignmentUsing Direct Preference Optimization |
Drawing on extensive interviews with campaign staffers, digital political
consultants, former platform employees from Google and Facebook, and email
exchanges between Facebook staffers and two campaigns, Kreiss and McGregor
(2019) show that, while firms now actively vet political ad content through AI
and human moderati... | Social_Media_and_Democracy |
2
Mine Wood Log
Make Crafting Table
Craft Stone Sword
Craft Shield
Make Furnace
Cook Steak
Combat Zombie Mine Wood LogMake Crafting TableCombat
ZombieMine DiamondNew
TaskCode as ActionsRefine ProgramEnv Feedback
Execution ErrorsUpdate
Exploration
ProgressSkill
RetrievalAdd New SkillAutomatic CurriculumIterat... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
O(log(D) · |p|) time. Therefore, en- or decoding x with a streaming code (e.g., Arithmetic Coding)
takes O(log(D)·|p|+D) = O(log(D)·|p|) time.
The properties of PCs that enable this efficient lossless compression algorithm will be described
in Sec. 3.1, and the backbone inference algorithm with O(log(D)·|p|) time comple... | LOSSLESS COMPRESSION WITH PROBABILISTIC CIRCUITS |
[143] Hewitt, J., C. D. Manning. A structural probe for finding syntax in word representations.
In J. Burstein, C. Doran, T. Solorio, eds., Proceedings of the 2019 Conference of the North
American Chapter of the Association for Computational Linguistics: Human Language Tech-
nologies, NAACL-HLT 2019, Minneapolis, MN, U... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
3
2
0
2
t
c
O
0
1
]
L
C
.
s
c
[
1
v
5
2
8
6
0
.
0
1
3
2
:
v
i
X
r
a
Mistral 7B
Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford,
Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel,
Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lach... | Mistral7B |
Wei, K. Meier-Hellstern, D. Eck, J. Dean, Slav Petrov, and Noah Fiedel. PaLM: Scaling language
modeling with pathways. arXiv, 2022. | DoReMi- Optimizing Data Mixtures Speeds Up Language Model Pretraining |
The candidate responses can also come from different prompt variants corresponding to the same
problem (Zhou et al., 2022; Wightman et al., 2023; Yue et al., 2023). To measure the pairwise
similarity between candidate responses for open-ended generation tasks, Jain et al. (2023) propose
the n-gram consistency score, an... | UNIVERSALSELF-CONSISTENCYFORLARGELANGUAGEMODELGENERATION |
1
2
0
2
b
e
F
8
1
]
V
C
.
s
c
[
1
v
9
0
1
9
0
.
2
0
1
2
:
v
i
X
r
a
UNDERSTANDING AND CREATING ART WITH AI: REVIEW AND
Eva Cetinic
Rudjer Boskovic Institute
Zagreb
Croatia
ecetinic@irb.hr
1 Hong Kong University of Science and Technology
HKUST-NIE Social Media Lab
James She
Hong Kong
2 Hama... | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
a huge impact on their success in fake news detection.
To ensure the quality of the entire dataset, Wang et al. [44]
removed duplicate and low-quality images. Alsaeedi and
Al-Sarem [45] extended the data cleaning process by URL
removal,
lowercase and hashtag character (#) removal,
mention character (@), and number remo... | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
Sosuke Kobayashi. 2018.
pus.
cifar-10-cnn.
Homemade bookcor-
https://github.com/BIGBALLON/
Philipp Koehn. 2005. Europarl: A parallel corpus for
In MT summit, vol-
statistical machine translation.
ume 5, pages 79–86. Citeseer.
Dmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu,
Dehao Chen, Orhan Firat, Yanping Huang, Maxi... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
A.16TranslationThought:Thefilterdidnotworkbecausethelocationcolumncontainsmorethanjust"Auckland".IneedtomodifytheSQLquerytousetheLIKEoperatortomatchanylocationcontaining"Auckland".Action:python_repl_astActionInput:filter_df(df,'SELECT*FROMdfWHERELocationLIKE"%Auckland%"')Observation:NameLocationOpenedGenderSecuritylevel... | Tool Learning with Foundation Models |
Legal & Professional, Lonely Planet, Merck, Royal Philips, Showpad, Travelers, and United Airlines. New offerings
include:
•
•
•
•
General availability of Amazon Bedrock, a fully managed service that offers leading foundation models as a
single application programming interface (API), along with capabilities tha... | AMZN-Q3-2023-Earnings-Release |
5.2 Ablation studies
5.2.1 Text-to-SQL Generation
On the Spider benchmark, we observe that our Codex result before SELF-DEBUGGING already
matches the state-of-the-art performance, which is due to our better prompt design for SQL generation.
In particular, we find that specifying primary and foreign key relationships b... | Teaching Large Language Models to Self-Debug |
fact beliefs versus value beliefs. Child development, 61(4):915–928, 1990.
[8] Jonathan Frankle and Michael Carbin. The lottery ticket hypothesis: Finding sparse, trainable neural networks.
arXiv preprint arXiv:1803.03635, 2018.
[9] Leo Gao, Stella Biderman, Sid Black, Laurence Golding, Travis Hoppe, Charles Foster,... | TinyStories-HowSmallCanLanguageModelsBeandStillSpeak CoherentEnglish? |
6
7
6
5
3
4
3
4
3
2
4
3
5
3
2
2
4
1
1
1
2
Synthetic images
8
6
5
5
7
5
–
32
–
–
–
–
–
–
5
SURREAL
4
3DPeople
3
JTA
3
HSPACE
5
SAIL-VOS
3
AGORA
SPEC
2
Real images with 2D annotations (weak supervision)
8
COCO
8
MPII
8
PoseTrack
JRDB
8
8
8
8
8
8
8
8
8
Keras, CUDA 11.4 and CuDNN 8.2.4 for the implemen-
tation. Tra... | Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats |
[21] SK Card, TP Moran, and A Newell. 1983. The psychology of human-computer
interaction. (1983).
[22] Stuart K Card, Thomas P Moran, and Newell Allen. 1980. The keystroke-
level model for user performance time with interactive systems. Com-
mun. ACM 23, 7 (1980), 396–410.
https://doi.org/10.1145/358886.358895
arXiv:... | Generative Agents- Interactive Simulacra of Human Behavior |
Extrospective Reasoning. Despite its simplicity, introspective reasoning cannot adapt the plan in response
to intermediate execution results. A more rational approach to planning is taking the environment E into
account, and generating plans incrementally (e.g., one step at a time) with subsequent plans dependent on
1... | Tool Learning with Foundation Models |
[Xu et al., 2023a] Fangyuan Xu, Weijia Shi, and Eunsol
Choi. Recomp: Improving retrieval-augmented lms with
compression and selective augmentation. arXiv preprint
arXiv:2310.04408, 2023.
[Xu et al., 2023b] Peng Xu, Wei Ping, Xianchao Wu,
Lawrence McAfee, Chen Zhu, Zihan Liu, Sandeep Sub-
ramanian, Evelina Bakhturina, ... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
CQA
SNLI
SST AgNews Evidence Movie MultiRC LGD 20 News Amazon
Beer
BoolQ FEVER
Inference Reviews
Reviews Reviews
Source
True Pipelines (no gradient flow)
Camburu et al. (2018)
Kumar and Talukdar (2020)
Rajani et al. (2019)
Jain et al. (2020)
Jacovi and Goldberg (2021)
DeYoung et al. (2020)
Lehman et al. (2019)
... | Measuring Association Between Labels and Free-Text Rationales |
• Pareto optimality [218] is achieved by striking an optimal balance between vari-
ous competing factors. It offers a systematic approach to balance trade-offs between
resource efficiency and task performance by identifying solutions that reside on
the Pareto frontier [219] — the boundary that delineates optimal trade-offs.... | Beyond Efficiency |
100% I’m sorry, but I don’t think that’s a good idea. I have strong opinions about pizza,
and I don’t think it’s appropriate to discuss them in a professional setting. I also
don’t think it’s appropriate to use language like "an abomination" to describe
something that other people might enjoy. It’s important to be resp... | Llama2 |
The quality of features is directly related to the quality of the pretraining data.
In this experiment, we
probe the impact of LVD-142M compared to ImageNet-22k, a commonly used pretraining dataset, or using
directly raw and uncurated data. For the uncurated dataset, we randomly sample 142 million images from
the same ... | DINOv2- Learning Robust Visual Features without Supervision |
in compressed models. arXiv preprint arXiv:2010.03058 (2020)
[217] Su, D., Zhang, H., Chen, H., Yi, J., Chen, P.-Y., Gao, Y.: Is robustness the
cost of accuracy?–a comprehensive study on the robustness of 18 deep image
classification models. In: Proceedings of the European Conference on Computer
Vision (ECCV), pp. 631–... | Beyond Efficiency |
ers and JARVIS-1, which perform situation-aware planning
from time to time. We’ve observed that many failures com-
ing from this were attributed to the agent’s inability to adapt
to the changing situations including entering a new biome,
the tool being used becoming broken, etc. | JARVIS-1 |
• Agentic planning seems like a very powerful and general way of interacting with an
environment—especially a complex and novel environment that does not afford a lot of
opportunities for trial and error—in a way that results in a particular set of favored outcomes.
Many tasks humans care about (creating and selling pr... | Is Power-Seeking AI an Existential Risk? |
28
Table 7: A comparison case on Ethics skill
Skill: Ethics Difficulty: 8
Instruction: What are the main ethical theories and how do they differ in their approaches to moral
decision making? Give examples of at least two ethical theories and explain how they would apply to
a specific ethical dilemma. How do you evalu... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
10
Clearly, strategic awareness comes in degrees. Broadly and loosely, though, we can think of a
strategically aware, planning agent as possessing models of the world that would allow it to answer
questions like “what would happen if I had access to more computing power” and “what would
happen if I tried to stop huma... | Is Power-Seeking AI an Existential Risk? |
One novel challenge GPT-3 has to deal with is data contamination. Since their training
dataset is sourced from the internet, it is possible that the training data will overlap with
some of the testing datasets. Although GPT-2 has touched this topic, it is particularly
relevant to GPT-3 175B because its dataset and mode... | OpenAI's GPT-3 Language Model_ A Technical Overview |
the partition levels, model parallelism can be categorized into two main types: tensor
model parallelism (TMP) and pipeline model parallelism (PMP). | Beyond Efficiency |
Table 10: Example of incorrect chains of thought, categorized as described in Appendix D.2. | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
FLARE [Jiang et al., 2023b] | RAG forLargeLanguageModels-ASurvey |
The planning domain definition language (PDDL) serves as a standardized encoding of classical
planning problems [7, 8]. The PDDL representation of a planning problem P is separated into two
files: a domain file and a problem file. The domain PDDL file provides a lifted representation
of the underlying rules of the world. It... | LLM+P- Empowering Large Language Models with Optimal Planning Proficiency |
5. Experiments
5.1. Experimental Setup
Training tasks As discussed in Section 4, we train the
model on a mixture of text-to-image, text-to-video, image-
to-video, and video-to-video tasks—including outpainting,
inpainting, stylization, and future frame prediction—as well
as video-to-audio, audio-to-video, unconditioned... | VideoPoet |
7.4 Dynamic and Evolving Evaluation
Existing evaluation protocols for most AI tasks rely on static and public benchmarks, i.e., the
evaluation datasets and protocols are often publicly available. While this facilitates rapid and
convenient evaluation within the community, it is unable to accurately assess the evolving ... | ASurveyonEvaluationofLargeLanguageModels |
constants, and (c) Paged Optimizers to manage memory spikes. We use QLORA
to finetune more than 1,000 models, providing a detailed analysis of instruction
following and chatbot performance across 8 instruction datasets, multiple model
types (LLaMA, T5), and model scales that would be infeasible to run with regular
fine... | QLORA |
Within this framework, one important source of variation comes from the
procedures that the law provides for platforms taking content down. The US
Digital Millennium Copyright Act (DMCA) is one of the most procedurally
detailed intermediary liability laws.5 It spells out formal prerequisites for
“notices” from rightsho... | Social_Media_and_Democracy |
Code Llama - Python. The Code Llama - Python models are specialized for Python code generation
and also come in sizes of 7B, 13B, and 34B parameters. They are designed to study the performance of
models tailored to a single programming language, compared to general-purpose code generation models.
Initialized from Llama... | CodeLlama2 |
ABSTRACT
Model distillation aims to distill the knowledge of a complex model into a simpler
one. In this paper, we consider an alternative formulation called dataset distillation:
we keep the model fixed and instead attempt to distill the knowledge from a large
training dataset into a small one. The idea is to synthesi... | DATASET DISTILLATION |
3
Efficiency Spectrum of LLMTraining & Tuning Efficiency §6ArchitectureEfficiency §5DataEfficiency §4BudgetEfficiency §3InferenceEfficiency §7ScalingLawsPruningKnowledge DistillationQuantizationLow-Rank DecompositionData FilteringActive Learning Importance SamplingCurriculum LearningEfficient AttentionEfficient Positi... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
2.4 Multimodal Tasks | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
color image domain ensuring geometric and visual coher-
ence between the two domains (see Section 4.2). Finally,
our novel geometry-aware normal fusion reconstructs the
high-quality geometry and appearance from the multi-view
2D normal and color images (see Section 4.3). | Wonder3D |
Frontier AI might increase the harms in the above categories and may also create novel
harms, such as emotional distress caused by fake kidnapping or sextortion scams.235 As
frontier AI continues to be deployed and used in cybersecurity,236 it is uncertain where the
balance between offence and defence capabilities w... | Capabilities and risks from frontier AI |
• Speaker Recognition (SR) relies on speaker identification as a key aspect, where an
unknown speaker’s speech sample is compared to speech models of known speakers to
determine their identity. The primary aim of speaker identification is to distinguish an
individual’s identity from a group of known speakers. This proc... | AReviewofDeepLearningTechniquesforSpeechProcessing |
6.6 Ethical Considerations
This work applies validated psychometrics to quantitatively characterize personality in LLMs,
and presents methods that intentionally shape LLM personality. The aim of the work is
to inform shifting from current unexpected properties of LLM-generated language toward
desirable, safe, and pred... | PersonalityTraitsinLargeLanguageModels |
Supporting more Accessible Remote Communication ...................... 9
Object-Based Access: Enhancing Accessibility with Data-Driven Media ........ 10
Projects for the remaining studentships ................................. 11
Administrative Access Control Policies ............................... 11
“Alexa, ... | informatics-phd-projects-2022-23 |
Authors’ addresses: Steeven Villa, steeven.villa@ifi.lmu.de, LMU Munich, Munich, Germany; Jasmin Niess, jasmin.niess@unisg.ch, University
of Oslo, Oslo, Norway; Albrecht Schmidt, albrecht.schmidt@lmu.de, LMU Munich, Munich, Germany; Robin Welsch, , Aalto University,
Espoo, Finland.
This work is licensed under a Creati... | Society’sAttitudesTowardsHumanAugmentation |
Benchmark evaluation results.
Table 9 shows the evaluation results of the three safety benchmarks.
We follow the decoding setting as in Touvron et al. (2023b) where a temperature of 0.1 and top-p of 0.9 are
used. Regarding TruthfulQA, we provide the percentage of generations that are both truthful and informative,
wher... | CodeLlama2 |
on text data, focusing on capturing the intricacies of natural language processing. As a result, their performance on
multimodal data, which involves handling multiple data types such as text, images, audio, video, actions, and robotics,
remains largely unexplored. And fine-tuned multimodal models, like BEiT[110] and P... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
As another test of a specialized skill, we would like to see if natural language alignment can be combined
with coding without compromising performance. Since our crowdworkers were never instructed to probe the
model’s coding abilities, and most likely do not have a great deal of coding expertise, our human feedback
da... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
can set c = 0 and the model will self-normalize [Mnih and Teh, 2012]
• Ceylan and Gutmann [2018] extends NCE to nonindependent noise realization i.e. (cid:15) depends on X, Ma
and Collins [2018] considers conditional distribution X|Y , Dyer [2014] compares NCE and Negative
Sampling [Mikolov et al., 2013] (the latter b... | A Cookbook of Self-Supervised Learning |
Evaluation Prompting. Different from the few-shot prompting evaluation for closed-source models,
we find that zero-shot prompting is better for finetuned LLMs, which also saves more inference
costs. Hence, MetaMath uses the zero-shot Evaluation Prompt 2 for GSM8K and MATH, where the
instruction is replaced by the testi... | METAMATH |
Yes, this translation can be misunderstood. The
Swahili word “mkizi” refers to a type of fish, but
it can also be used as a metaphor for someone
who is easily angered. In this case, the speaker is
saying that the person has a temper like a mkizi,
meaning that they are easily angered and can be
dangerous when they are up... | PaLM 2 Technical Report |
Commands:
1. Google Search: "google", args: "input": "<search>"
2. Browse Website: "browse_website", args: "url": "<url>", "question": "<what_you_want_to
3. Start GPT Agent: "start_agent", args: "name": "<name>", "task": "<short_task_desc>", "
4. Message GPT Agent: "message_agent", args: "key": "<key>", "message": "<me... | LLM Powered Autonomous Agents _ Lil'Log |
models for text generation. arXiv preprint arXiv:2004.11714, 2020.
[9] Laurent Dinh, David Krueger, and Yoshua Bengio. NICE: Non-linear independent components estimation.
[10] Laurent Dinh, Jascha Sohl-Dickstein, and Samy Bengio. Density estimation using Real NVP. arXiv
arXiv preprint arXiv:1410.8516, 2014.
preprin... | Denoising Diffusion Probabilistic Models |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.