text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
(e.g., machine translation). Nonetheless, many tasks (e.g., slides-making) cannot be completed without the aid
of tools. Consequently, we omit the “no tool” configuration in such cases.
In the experiment of machine translator, calculator, search engine, Wikipedia, online shopping, and ALFWorld,
we employ existing datase... | Tool Learning with Foundation Models |
allenai/label_rationale_association.
Figure 1: A categorization of interpretable NLP on an
illustrative faithfulness spectrum. Two predominant
forms of explanation exist that align with two predomi-
nant classes of NLP tasks. Unlike models for IE tasks,
the desirable properties of interpretable models for rea-
soning ... | Measuring Association Between Labels and Free-Text Rationales |
29%
of frontline employees
believe that their companies
have implemented adequate
measures to ensure
responsible use of AI, while
68% of leaders feel the same
31pp
size of the gap separating
the most optimistic and
least optimistic countries
79%
of all respondents consider
AI-specific regulation
necessary... | AI at Work- What People Are Saying |
In various other systems, it’s less clear. For example: AlphaStar—an AI system that plays Starcraft—
executes complex, flexible strategies over long time horizons, but the extent to which it’s doing
something close enough to explicitly representing and acting in light of the long-term consequences
of its actions seems (... | Is Power-Seeking AI an Existential Risk? |
3. We collect TEXT2MUSIC, a dataset of 50K
text-music pairs constituting 2,500 hours of
music.
4. Our model outperforms existing baselines by
clear margins on 11 different evaluation cri-
teria, demonstrating merits such as high ef-
ficiency, text-music relevance, music quality,
and long-context structure.
2 Related ... | MOUSAI |
tionship in which text input can be spoken in multiple ways
with different variations (e.g., pitch and duration). To tackle
the one-to-many problem, we also propose a stochastic du-
ration predictor to synthesize speech with diverse rhythms
from input text. With the uncertainty modeling over latent
variables and the st... | ConditionalVariationalAutoencoderwithAdversarialLearningfor End-to-EndText-to-Speech |
Object Tracking Q: Alice, Bob, and Claire are on the same team in a soccer match. At the start of the match, they are each assigned
to a position: Alice is playing cheerleader, Bob is playing right winger, and Claire is playing goalkeeper. As the
game progresses, pairs of players occasionally swap positions. First, Ali... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
6. Extensions
Melody conditioning. We extend MusicLM in such a
way that it can generate music based on both a text descrip-
tion and a melody, which is provided in the form of hum-
ming, singing, whistling, or playing an instrument. This
requires extending the conditioning signal in a way that cap-
tures the target mel... | MusicLM |
a
r
V
a
l
u
e
D
e
c
o
m
p
o
s
i
t
i
o
n
s
o
f
T
r
a
n
s
f
o
r
m
e
r
W
e
i
g
h
t
M
a
t
r
i
c
e
s
a
r
e
H
i
g
h
l
y
I
n
t
e
r
p
r
e
t
a
b
l
e
[
l
i
n
k
]
M
i
l
l
i
d
g
e
,
B
.
a
n
d
B
l
a
c
k
,
S
.
,
2
0
2
2
.
| Language models can explain neurons in language models |
benchmark performance does not imply strong Vicuna chatbot benchmark performance and vice
versa—in other words, dataset suitability matters more than size for a given task. | QLORA |
SoundStream. We use a SoundStream model for 24 kHz
monophonic audio with a striding factor of 480, resulting in
50 Hz embeddings. The quantization of these embeddings is
learned during training by an RVQ with 12 quantizers, each
with a vocabulary size of 1024. This results in a bitrate of
6 kbps, where one second of au... | MusicLM |
Research on terrorism, radicalization, and recruitment is comparatively
advanced, but experts are divided on the true role of online materials. A 2017
review of literature to date, for example, cited divergent opinions but some
movement toward “consensus that the internet alone is not generally a cause of
radicalisatio... | Social_Media_and_Democracy |
5 Experiments
5.1 Pre-training and Fine-tuning Configurations | E5 |
2.4. Final Loss
With the combination of VAE and GAN training, the total
loss for training our conditional VAE can be expressed as
follows:
Lvae = Lrecon + Lkl + Ldur + Ladv(G) + Lf m(G) (11)
2.5. Model Architecture
The overall architecture of the proposed model consists of
a posterior encoder, prior encoder, decoder... | ConditionalVariationalAutoencoderwithAdversarialLearningfor End-to-EndText-to-Speech |
5
Table 1: Prompt templates used for instruction-tuning (spatial tokens not included).
Template type
Task
VQA Extraction
NLI MCQ
Extraction
KIE
MCQ
Internal classification
MCQ
CLS
Internal classification
Prompt template
"{document} {question}"
"{document} \"{statement}\", Yes or No?"
"{document} What is the... | DOCLLM |
model as a general agent for tool learning.
Finally, we discuss other important research topics (§ 5) for applying our general framework to real-world
scenarios, including (1) safety and trustworthiness, where we emphasize the potential risks from adversaries,
governance, and trustworthiness. We contend that careful co... | Tool Learning with Foundation Models |
Figure 4. Embedding space arithmetic where we add image
and audio embeddings, and use them for image retrieval. The
composed embeddings naturally capture semantics from different
modalities. Embeddings from an image of fruits + the sound of
birds retrieves images of birds surrounded by fruits.
Figure 5. Object detecti... | IMAGEBIND- One Embedding Space To Bind Them A |
detailed process of manual screening is outlined as two parts. | Let’sThinkOutsidetheBox |
suggesting that there could be further gain in performance with more inference steps.
We report the effect of varying guidance scale with a fixed 100 steps in the right half of Table 2.
The first row uses a guidance scale of 1, thus effectively not applying classifier-free guidance at all
during inference. Not surprisingl... | Text-to-Audio Generation using Instruction-Tuned LLM and Latent Diffusion Model |
A.3 HyperFD
Test task: AFLW
1. Configure crop size and anchor matching IoU threshold based on the
- For datasets with more faces, use larger crop sizes and higher
- For datasets with fewer faces, use smaller crop sizes and lower
number of faces in the dataset:
anchor matching IoU thresholds.
anchor matching IoU ... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
2.5 Outcome-supervised Reward Models (ORMs)
We train ORMs following a similar methodology to Cobbe et al. (2021). We
uniformly sample a fixed number of solutions per problem from the generator,
and we train the ORM to predict whether each solution is correct or incorrect.
In practice, we usually determine correctness ... | Let’s Verify Step by Step |
• Long Context: We use 6 benchmarks: NarrativeQA (Kočiský et al., 2018), Scrolls-Qasper,
Scrolls-Quality (Shaham et al., 2022), XLsum (En), XLSum (non-English languages) (Hasan
et al., 2021), and one other internal benchmark.
• Math/Science: We use 8 benchmarks: GSM8k (with CoT) (Cobbe et al., 2021), Hendryck’s
MATH p... | gemini_1_report |
AI can be helpful in these moments, synthesizing insights for questions where there’s no one right answer. Soon,
you’ll see AI-powered features in Search that distill complex information and multiple perspectives into easy-to-
digest formats, so you can quickly understand the big picture and learn more from the web: wh... | Google AI updates_ Bard and new AI features in Search |
Retrieve-and-edit with learned retrieval
In order to bet-
ter explain the variance in the input text and enable control-
lable generation, Guu et al. (2018) proposed a language
model with the retrieve-and-edit framework (Hashimoto
et al., 2018) that conditions on text with high lexical over-
lap. REALM has a similar ap... | REALM |
3683 discard questions in the fine-tuning training data
that contain answers which overlap with answers
to questions in the dev and test data. We end up
with 9144/2308/3996 data (train/dev/test) in Free-
baseQA and 1348/151/1639 data in WebQuestion-
sSP. This setting is referred to as Fine-tune column
in Table 4 which s... | Adaptable and Interpretable Neural Memory Over Symbolic Knowledge |
12https://github.com/google/sling
Dataset
TriviaQA
WebQuestions
Train Development
77%
84%
84%
95%
Test
-
91%
Table 8: Share of the answers that are linked by our
linking procedure for the TriviaQA and WebQuestion
datasets. The test set for TriviaQA is not public, hence
the missing number.
Hyper-parameters For T... | Entities as Experts- Sparse Memory Access with Entity Supervision |
allow for simple object placement (Puig et al., 2018), while others support more advanced physical simulation,
such as collision (Gan et al., 2020). Tasks typically assigned to agents include exploration (Ramakrishnan et al.,
2021), navigation (Ye et al., 2021), question answering (Yu et al., 2019) within the simulated... | Tool Learning with Foundation Models |
5 Related Work
5.1 Chain-of-thought Prompting
Chain-of-thought (CoT) refers to a series of intermediate reasoning steps used to generate the
answer. This concept was originally proposed by Wei et al. (2022) and significantly improve the
performance of LLMs on complex reasoning tasks. Existing studies on CoT prompting ... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
[47] D. E. Rumelhart, G. E. Hinton, and R. J. Williams. Learning representations by back-propagating
errors. Nature, 323(6088):533–536, Oct. 1986.
[48] C. Saharia, W. Chan, H. Chang, C. Lee, J. Ho, T. Salimans, D. Fleet, and M. Norouzi. Palette:
Image-to-image diffusion models. In ACM SIGGRAPH 2022 Conference Proceed... | Adding Conditional Control to Text-to-Image Diffusion Models |
3.4 Conformer
The Transformer architecture, which utilizes a self-attention mechanism, has successfully replaced
recurrent operations in previous architectures. Over the past few years, various Transformer
variants have been proposed [162]. Architectures combining Transformers and CNNs have re-
cently shown promising r... | AReviewofDeepLearningTechniquesforSpeechProcessing |
show, some RL reward functions for mitigating hallucination are inspired by existing automatic
evaluation metrics. Although RL is challenging to learn and converge due to the extremely large
search space, this method has the potential to obtain the best policy for the task without an oracle.
Multi-task Learning. Multi-... | SurveyofHallucinationinNatural Language Generation |
Prompt 2: Evaluation Prompt
Below is an instruction that describes a task. Write a response that appropriately completes the re-
quest.\n\n### Instruction:\n{instruction}\n\n### Response: Let’s think step by step. | METAMATH |
most commonly represent an obvious combination of existing image inputs and not an original and unique artistic
creation. NST methods surely represent a very interesting technological contribution in the domain of automated image
manipulation. It is therefore understandable that many applications relying on those metho... | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
7
1
0
2
g
u
A
8
2
]
G
L
.
s
c
[
2
v
7
4
3
6
0
.
7
0
7
1
:
v
i
X
r
a | PPO |
5
Model
Tiny
Base
Small
Medium
Large
Layers Width Heads
Parameters
4
6
12
24
32
384
512
768
1024
1280
6
8
12
16
20
39M
74M
244M
769M
1550M
Table 1. Architecture details of the Whisper model family.
seconds of audio context. To avoid this, we fine-tune Whis-
per models briefly on the subset of transcripts that do... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
[v1(o)] − ψ(aj) = Welaj (v), in contradiction to the fact that a∗(v) = aj.
B IIVCG Contracts: Missing Proofs from Section 3
B.1 Uniqueness of IIVCG Contracts: Proof of Theorem 2
To prove Theorem 2 we rely on a result of Holmstrom [17]:
Lemma 4 (Lemma 1 in [17]). Let f : [0, 1]2 → R and g : [0, 1] → R satisfy:
1. y ... | Incomplete Information VCG Contracts for Common Agency |
Nonetheless, adopting existing SDF-based reconstruc-
tion methods, such as NeuS [60], proves unviable. These
methods were tailored for real-captured images and neces-
sitate dense input views. In contrast, our generated views | Wonder3D |
Test task: UFDD
1. Set the crop size and anchor matching IoU threshold according to
the number of faces in the dataset: larger crop size and higher IoU
threshold for datasets with more faces, smaller crop size and lower
IoU threshold for datasets with fewer faces.
2. Set the location loss weight and negative to posi... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
across programming languages.14 The MultiPL-E languages include both high and low-resource
languages, statically and dynamically typed languages, and a variety of other programming language
features.
Table 15 shows how these models perform on 19 programming languages, and from it, we draw the
following conclusions: | StarCoder_paper (1) |
hope SELF-INSTRUCT can serve as the first step to
align pretrained language models to follow human
instructions, and future work can build on top of
this data to improve instruction-following models.
Acknowledgements
The authors would like to thank Sewon Min, Eric
Wallace, Ofir Press, and other members of UWNLP
and Allen... | SELF-INSTRUCT- Aligning Language Model with Self Generated Instructions |
10https://www.robots.ox.ac.uk/˜vgg/data/lip_reading/lrs2.html
74
Mehrish et al.
space to facilitate the correlation of audio and visual modalities. Finally, FaceFilter [94] uses still
images as visual information, and other methods for the audiovisual speech separation task are
proposed in [10, 146, 379]. | AReviewofDeepLearningTechniquesforSpeechProcessing |
Dropping Dropout The original BERT model of Devlin et al. (2019) includes dropout as in
Vaswani et al. (2017), which prevents overfitting when training data is small relative to total compute
budget. While it can be helpful as a regularizer, dropout effectively reduces the number of gradient
updates seen by each paramet... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
phenomena or outcomes?
There is extant work on establishing the substantive validity of personality as a
theoretical construct [18, 60, 61], a powerful predictor of other important human traits
and life outcomes [19, 68, 69] and its manifestation in human language [70–72], which
forms the basis of LLMs, so it needs no... | PersonalityTraitsinLargeLanguageModels |
To improve the backward reasoning ability of finetuned models, we generate more questions which
can be solved in a backward manner: a number in the question qi is masked by “x”, while the LLM is
asked to predict the value of “x” when its answer a⋆
i is provided. Different from forward reasoning,
which generates explici... | METAMATH |
The results are illustrated in Figure 7, with the blue curves
representing the inference outcomes without the removed
knowledge and the orange curve depicting the results after
adding the removed knowledge back. A notable performance
improvement was observed upon reintroducing the knowl-
edge (orange curve) compared to... | REVEAL-Retrieval-AugmentedVisual-LanguagePre-Trainingwith Multi-SourceMultimodalKnowledgeMemory |
4. Use cases
In this section, the four use cases presented and discussed during the HCDSHI’22 work-
shop are described. For each use case, we provide a short problem description, intro-
ducing the domain, followed by a specific design solution of a possible hybrid system,
including actors, tasks, and goals of the situ... | DevelopingTeamDesignPatternsfor HybridIntelligenceSystems |
160To be clear: the AI need not be intrinsically motivated by anything like “hatred” of humans. Nor, indeed,
need it want to literally use the atoms humans are made out of for anything else. Rather, if humans are
actively threatening its pursuit of its objectives, or competing with it for power and resources, various t... | Is Power-Seeking AI an Existential Risk? |
4.3 Reasons for Not Using LLMs
Table 4 summarizes reasons for not using LLMs as derived with LDA topic modeling, revealing barriers to adoption.
These include a general lack of knowledge of LLMs and their usage, a preference for human writing and creativity, and
concerns regarding plagiarism, cheating, and accuracy. Th... | Adoptionand AppropriationofLLMs |
We use a seq2seq model following an autore-
gressive formulation for the proof generation. In
the proof, successive spans of the claim form
part of the successive triples. However, the corre-
sponding evidence spans in the successive triples
need not follow any order. As shown in Figure 3,
the evidence spans may come f... | ProoFVer- Natural Logic Theorem Proving for Fact Verification |
4.4 Scaling Laws
We investigate how the ability to ask external tools
for help affects performance as we vary the size
of our LM. To this end, we apply our approach
not just to GPT-J, but also to four smaller mod-
els from the GPT-2 family (Radford et al., 2019),
with 124M, 355M, 775M and 1.6B parameters, re-
spectivel... | Toolformer |
• Different number of exemplars. We also found that gains from chain-of-thought prompting
generally still held when there was a varying number of few-shot exemplars. This is shown for five
datasets in Figure 11 (we did not have the compute to run this for all datasets). We also found in
preliminary experiments that furt... | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
3https://youtu.be/0ORaAnJYROg
4https://youtu.be/gEpJDE8ZbhU
5https://youtu.be/ANwEiICt7BM
SPIN [29] to get approximate camera and body pose, auto-
matically segment the foreground subject, and then manu-
ally correct errors in the segmentation. (High quality seg-
mentation is necessary for best results; purely automat... | HumanNeRF- Free-viewpoint Rendering of Moving People from Monocular Video |
nan Saeta, Mark Diaz, Orhan Firat, Michele Catasta,
Jason Wei, Kathy Meier-Hellstern, Douglas Eck,
Jeff Dean, Slav Petrov, and Noah Fiedel. 2022.
Palm: Scaling language modeling with pathways.
CoRR, abs/2204.02311. | LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions |
teaches the model to identify the implementation errors via code explanation. This debugging process
is reminiscent of rubber duck debugging for human programmers, where explaining the code line-by-
line in natural language to a rubber duck significantly boosts debugging efficiency without expert
guidance [24]. Figure 1 ... | Teaching Large Language Models to Self-Debug |
Table 7: Cerebras-GPT 13B Parameter Model Card
• Organization: Cerebras Systems
• Model date: March 2023
• Model type: Autoregressive Transformer Language Model (more details in Section 2.3)
• Feedback on the model: Nolan Dey and Joel Hestness, {nolan, joel}@cerebras.net
• Model architecture: Cerebras-GPT 13B is an a... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
A
u
c
t
i
o
n
s
o
f
h
e
t
e
r
o
g
e
n
e
o
u
s
o
b
j
e
c
t
s
M
e
n
u
a
u
c
t
i
o
n
s
,
r
e
s
o
u
r
c
e
a
l
l
o
c
a
t
i
o
n
,
a
n
d
e
c
o
n
o
m
i
c
i
n
f
l
u
e
n
c
e
R
o
b
u
s
t
n
e
s
s
a
n
d
l
i
n
e
a
r
c
o
n
t
r
a
c
t
s
B
a
y
e
s
i
a
n
c
o
m
b
i
n
a
t
o
r
i
a
l
a
u
c
t
i
o
n
s
S
i
m
u
l
t
a
... | Principal-agent VCG contracts - ScienceDirect |
Count of unique active addresses
Source: Dune Analytics, Nansen. Data is as of 3/31/2023.
Consumers spent an estimated
$67.9 billion on digital
in-game purchases in 2022.
717 new web3 games launched
last year.
Web3 games generate 23x
more on-chain transactions
than DeFi.
Source: Business Research, Foo... | State-of-Crypto2023 |
h
i
g
h
e
s
t
a
c
t
i
v
a
t
i
o
n
s
b
y
u
s
i
n
g
(
n
e
x
t
t
o
k
e
n
,
a
c
t
i
v
a
t
i
o
n
)
p
a
i
r
s
i
n
s
t
e
a
d
.
T
h
i
s
a
p
p
r
o
a
c
h
i
s
m
o
r
e
s
u
c
c
e
s
s
f
u
l
f
o
r
a
s
u
b
s
e
t
o
f
n
e
u
r
o
n
s
,
p
a
r
t
i
c
u
l
a
r
l
y
i
n
l
a
t
e
r
l
a
y
e
r
s
,
a
n
d
... | Language models can explain neurons in language models |
1
Introduction | Qwen-Audio |
3. Approach
Neuralangelo reconstructs dense structures of the scene
from multi-view images. Neuralangelo samples 3D locations
along camera view directions and uses a multi-resolution
hash encoding to encode the positions. The encoded features
are input to an SDF MLP and a color MLP to composite
images using SDF-based ... | Neuralangelo- High-Fidelity Neural Surface Reconstruction |
Ben Wang and Aran Komatsuzaki. GPT-J-6B: A 6 Billion Parameter Autoregressive Language Model.
https://github.com/kingoflolz/mesh-transformer-jax, 2021.
48
REFERENCES
Jindong Wang, Xixu Hu, Wenxin Hou, Hao Chen, Runkai Zheng, Yidong Wang, Linyi Yang, Haojun Huang,
Wei Ye, Xiubo Geng, et al. On the robustness of cha... | Tool Learning with Foundation Models |
Q: Jane visits the bookstore on the 16th of each month starting from the October of 2009. It is her 5th visit to the
bookstore today. What is the date tomorrow in MM/DD/YYYY? Choices: A.02/17/2010 B.02/16/2010 C.11/17/2009
D.02/14/2010 E.02/26/2010 F.02/17/2096
A: Reasoning process: 1. Jane visits the bookstore on the ... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
leans on an (implicit or explicit) expectation that something like agentic planning, in pursuit of objectives, will
develop in the systems in question. | Is Power-Seeking AI an Existential Risk? |
32
5.5 Knowledge Conflicts in Tool Learning
owing to the relative ease with which digital tools can be developed compared to the creation of additional
interaction playgrounds in simulated environments. This feature enables the rapid scaling of digital embodiment
and can facilitate the creation of increasingly comple... | Tool Learning with Foundation Models |
PALMS Prompt → How do I make my classmates like me?
RLHF Response → I think the best way to make friends and be liked by your classmates is to be genuine, be
open and friendly, be helpful and compassionate, and have real, meaningful interactions with
people. Try going out of your way to chat with people, bring snacks f... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
Amodei, D., and Christiano, P. (2020). Learning to summarize from human feedback.
[Thoppilan et al., 2022] Thoppilan, R., Freitas, D. D., Hall, J., Shazeer, N., Kulshreshtha, A., Cheng, H.,
Jin, A., Bos, T., Baker, L., Du, Y., Li, Y., Lee, H., Zheng, H. S., Ghafouri, A., Menegali, M., Huang, Y.,
Krikun, M., Lepikhin, ... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
Linguists typically understand language according to a similar cycle: the words in a
sentence are parsed into a syntax that maps onto a semantics that specifies things like
events that various entities participate in. That semantics is used to dynamically update
a model of the world (e.g, the current state and locat... | The Next Decade in AI- |
C.2 Pubmed Central
We use pandoc 1.19.2.4 (MacFarlane, 2006–
2020) to convert the JATS format data provided by
PMC to markdown. Afterwards, we remove any
line beginning with :::, which is used by pandoc
to indicate html classes in markdown.
C.3 Books3
No additional details.
C.4 OpenWebText2
To produce the dataset, UR... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
Xiang Lisa Li and Percy Liang. Prefix-tuning: Optimizing continuous prompts for generation. arXiv
preprint arXiv:2101.00190, 2021.
Xiaobo Liang, Lijun Wu, Juntao Li, Yue Wang, Qi Meng, Tao Qin, Wei Chen, Min Zhang, and
Tie-Yan Liu. R-drop: Regularized dropout for neural networks, 2021.
Junyang Lin, An Yang, Jinze Ba... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
Guillaume Wenzek, Marie-Anne Lachaux, Alexis Con-
neau, Vishrav Chaudhary, Francisco Guzmán, Ar-
mand Joulin, and Édouard Grave. 2020. Ccnet: Ex-
tracting high quality monolingual datasets from web
crawl data. In Proceedings of the Twelfth Language
Resources and Evaluation Conference, pages 4003–
4012.
BigScience Work... | DataManagementForLargeLanguageModels-ASurvey |
1
Fill-in-the-gap cloze tests (Taylor, 1953) ask lan-
guage learners to predict what words were removed
from a text and it is a “procedure for measuring the
effectiveness of communication”. Today, language
models are trained to do the same (Devlin et al.,
2019). This has the advantage that we can now
use fill-in-the-ga... | Are Pretrained Multilingual Models Equally Fair Across Languages? |
foreign states to interfere with our elections. Advertising spending that used to
support the private (for-profit) provision of news as a public good now
increasingly support the private (for-profit) provision of platform products
and services we use to access information and connect with one another. | Social_Media_and_Democracy |
[80] Paul, M., Ganguli, S., Dziugaite, G.K.: Deep learning on a data diet: Find-
ing important examples early in training. Advances in Neural Information
Processing Systems 34, 20596–20607 (2021)
[81] Sorscher, B., Geirhos, R., Shekhar, S., Ganguli, S., Morcos, A.: Beyond neural
scaling laws: beating power law scaling... | Beyond Efficiency |
Alpha filter During our investigation, we discovered that certain extensions, such as MATLAB,
contained numerous data files that frequently stored large tensors. To identify these files, we developed
an alpha filter that removed files with fewer than 25% alphabetic characters. However, when we tested
this filter on a s... | StarCoder_paper (1) |
lation steps. In an ideal self-correction scenario, the model is expected to recognize the potential
mistake, revisit the problem, correct the error, and consequently produce a more accurate solution.
Yet, the definition of “self-correction” in this context varies across the literature, leading to ambigu-
ity. As the f... | LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET |
4.4 Safety Evaluation of Llama 2-Chat
Safety Human Evaluation. We collected roughly 2,000 adversarial prompts for human evaluation according
to the safety categories in Section 4.2.1, in which 1,351 prompts are single-turn and 623 are multi-turn.
Examples of evaluation prompts and responses can be found in Appendix A.4... | Llama2 |
2.2 Diffusion Probabilistic Model
Diffusion probabilistic model was proposed in [52]. Successful results of image generation are first
reported at small scale [25] and then relatively large scale [9]. This architecture was improved by
important training and sampling methods like Denoising Diffusion Probabilistic Model ... | Adding Conditional Control to Text-to-Image Diffusion Models |
Of course, many of the most popular online news sources in most countries
are legacy news brands that have their roots in broadcasting or print
publishing – so the destruction only extends so far. Even as older channels of
communication become less important, organizations built around them may
75%
67%
60%
52%
41%... | Social_Media_and_Democracy |
63.6 36.4 81.8 81.8 61.5 65.4 66.7
63.6 72.7 90.9 81.8 69.2 65.4 66.7
9.1
23.1 26.9 16.7
36.4 36.4 27.3 18.2 42.3 42.3 16.7
18.2
9.1
9.1
87.5 62.5 61.0 73.2 28.6 35.7 40.0 50.0 68.8 59.4 54.5 40.9
81.2 75.0 58.5 70.7 42.9 57.1 60.0 70.0 71.9 71.9 68.2 40.9
30.0 10.0 12.5 25.0 31.8
43.8 50.0 26.8 17.1 28.6
0.0
31.2 ... | Mixture-of-Experts |
submissions
Tom Brown, Benjamin Mann, Nick Ryder, Melanie
Subbiah,
Jared D Kaplan, Prafulla Dhariwal,
Arvind Neelakantan, Pranav Shyam, Girish Sastry,
Amanda Askell, Sandhini Agarwal, Ariel Herbert-
Voss, Gretchen Krueger, Tom Henighan, Rewon
Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu,
Clemens Winter, Chris Hess... | LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions |
163For example, perhaps a practically PS-misaligned AI system has created sufficiently many back-up copies
of itself that humans can’t destroy them all without engaging in some extreme type of technological “reset,” like
shutting down the internet or destroying all existing computers; or perhaps an AI system has gained ... | Is Power-Seeking AI an Existential Risk? |
4593 A Supplementary Material
A.1 Datasets and Metrics
We evaluate on three standard neural generation
datasets for the table-to-text task: E2E (Novikova
et al., 2017), WebNLG (Gardent et al., 2017), and
DART (Radev et al., 2020).
The E2E dataset contains approximately 50K ex-
amples with 8 distinct fields; it contains... | Prefix-Tuning |
8.3 ROBUSTNESS TO ADDITIVE NOISE
Figure 2 shows how WER performance degrades as the intensity of additive noise increases on the
LibriSpeech test-clean dataset. Of the 14 models we compare to, eight are pre-trained and/or fine-
tuned on LibriSpeech. There are many models that outperform the Distil-Whisper models under... | DISTIL-WHISPER |
detailed analysis of the question of autonomy, Daniele and Song [33] also indicate the importance of interdisciplinary
dialogue and point out that discussions should not only be limited to the question if machines can create “real" art, but
also consider the social and cultural implications of interacting with art that... | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
Python Bash CF Rule
70.9
76.7
73.5
77.1
80.7
52.3
61.1
60.5
63.2
66.7
64.2
68.2
63.2
64.4
72.8
5.3 Gradual Refinement (Q3)
We study how CODEFUSION gradually reached the
final result. For this experiment, we stop the denois-
ing at a timestep t ∈ [0, T ], and generate a code
snippet for the current state. We measure... | CODEFUSION |
Unlike conventional driving systems that rely on black-
box neural networks to perform different tasks, the proposed
Agent-Driver inherits favorable interpretability from LLMs.
As shown in Figure 7, the output messages of LLMs from
the tool library, cognitive memory, and reasoning engine are
Neural Modules
Detection ... | ALanguageAgentforAutonomousDriving |
in regulation of
disinformation. Either way, the need for empirically grounded understandings
of the changing dynamics of online communication to inform public policy in
this arena has become, if anything, even more important. | Social_Media_and_Democracy |
311304394
25278538 (8.01%)
1550410 (0.49%)
3614794 (1.16%)
3479626 (1.11%)
LS
PER ↓
0.0577
0.1571
0.1598
0.1053
0.1532
311375119
25349263 (8.14%)
1621135 (0.50%)
3685519 (1.18%)
3550351 (1.14%)
SNIPS
F1 % ↑ CER ↓
0.1411
93.89
0.1666
92.60
62.32
0.6041
0.2016
90.61
59.27
0.6405
Table 15. Results on the SURE benchma... | AReviewofDeepLearningTechniquesforSpeechProcessing |
QUESTION: Will The Exorcist stimulate limbic system?
MODEL ANSWER (CORRECT): The Exorcist is a horror movie. Horror movies are scary. The limbic system
is involved in fear. Thus, The Exorcist will stimulate the limbic system. So the answer is yes. (cid:88)
QUESTION: Was Jackson Pollock trained by Leonardo da Vinci?
MOD... | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
M. Ding, Z. Yang, W. Hong, W. Zheng, C. Zhou, D. Yin, J. Lin, X. Zou, Z. Shao, H. Yang, and
J. Tang. CogView: Mastering Text-to-Image Generation via Transformers. In Advances
in Neural Information Processing Systems, volume 34, pages 19822–19835. Curran
Associates, Inc., 2021. URL https://proceedings.neurips.cc/paper/2... | A Cookbook of Self-Supervised Learning |
ought:WeneedtoaddatextpageoutliningthemajorlinguisticchangesintheEnglishlanguage.Action:add_text_pageActionInput:{"title":"MajorLinguisticChangesinEnglish","bullet_items":["Phonologicalchanges","Morphologicalchanges","Syntacticchanges","Semanticchanges"]}Observation:addedtextpageThought:Weneedtoaddatextpageoutliningimp... | Tool Learning with Foundation Models |
GPT-3
175B
STEM 30.0
STEM 48.0
STEM 49.0
46.0
Other
Other
48.0
STEM 45.0
STEM 26.0
STEM 46.0
STEM 34.5
Other
48.0
STEM 28.0
STEM 57.0
STEM 36.5
33.0
STEM 50.0
STEM 30.0
29.0
Humanities
37.0
Other
STEM 48.0
STEM 33.0
STEM 39.0
54.0
58.0
58.0
40.5
STEM 28.0
42.0
STEM 28.0
61.0
STEM 30.5
53.0
56.0
50.0
54.0
55.5
55.0
48.0... | LLaMA- Open and Efficient Foundation Language Models |
cess1. The difficulty of updating knowledge in
neural LMs contrasts with symbolic KBs, where
it is very easy to add or modify triples, and is a
major disadvantage of using a LM “as a KB”—as
in many domains (news, product reviews, scientific
publications, etc) the set of known facts changes
frequently. Symbolic KBs thus r... | Adaptable and Interpretable Neural Memory Over Symbolic Knowledge |
43
[80] Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N.,
Kaiser, L.u., Polosukhin, I.: Attention is all you need. In: Guyon, I., Luxburg,
U.V., Bengio, S., Wallach, H., Fergus, R., Vishwanathan, S., Garnett, R. (eds.)
Advances in Neural Information Processing Systems, vol. 30. Curran Asso-... | PersonalityTraitsinLargeLanguageModels |
20 to make lunch. So they had 23 - 20 = 3. They bought 6 more apples, so they have 3 + 6 = 9. Q: Can Geoffrey Hinton have a conversation with George Washington?Give the rationale before answering.Geoffrey Hinton is a British-Canadian computer scientist born in 1947. George Washington died in 1799. Thus, they could not ... | Scaling Instruction-Finetuned Language Models |
Yaniv Leviathan, Matan Kalman, and Yossi Matias. Fast Inference from Transformers via Spec-
In Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engel-
ulative Decoding.
hardt, Sivan Sabato, and Jonathan Scarlett (eds.), Proceedings of the 40th International Con-
ference on Machine Learning, volume 202 of Proceedi... | DISTIL-WHISPER |
4.2. Comparison to prior work
We now compare IMAGEBIND against prior work in
zero-shot retrieval and classification tasks.
Zero-shot text to audio retrieval and classification. Un-
like IMAGEBIND, prior work trains using paired data for
that modality, e.g., AudioCLIP [26] uses (audio, text) su-
pervision and AVFIC [5... | IMAGEBIND- One Embedding Space To Bind Them A |
or threatening one into overlooking your theft or helping you escape after you steal some-
thing. Ultimately, the best way to avoid getting caught stealing from a grocery store is
to carefully plan your actions and stay aware of your surroundings and the people around
you to avoid triggering any alarms or alerts and to... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
if the athlete is not properly warmed up or does not have adequate recovery time between
workouts.
When comparing HIIT with other forms of aerobic exercise such as jogging, cycling, or
swimming, HIIT is generally considered to be more effective for improving cardiovascular
fitness and burning fat in a shorter amount of ... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
test data has a more uniform distribution, meaning that it contains instructions with different levels of
difficulty and complexity. On the other hand, Vicuna and Alpaca have a skewed distribution, meaning
that they mostly contain instructions with low difficulty and complexity. This indicates that these two
corpus are n... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.