text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
and slot filling. arXiv preprint arXiv:1812.10235 (2018).
[582] Yongqiang Wang, Yangyang Shi, Frank Zhang, Chunyang Wu, Julian Chan, Ching-Feng Yeh, and Alex Xiao. 2021.
Transformer in Action: A Comparative Study of Transformer-Based Acoustic Models for Large Scale Speech Recog-
nition Applications. In ICASSP 2021 - 2... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Combating disinformation would likely require a piecemeal fine-tuning
of CDA 230. The incentive to engage in campaigns of political
disinformation is not eliminated by simply making these efforts more
challenging, and they will continue evolving as they have been in the
past. Although exceptions to CDA 230 will likely m... | Social_Media_and_Democracy |
S. Bubeck, V. Chandrasekaran, R. Eldan, J. Gehrke, E. Horvitz, E. Kamar,
P. Lee, Y. T. Lee, Y. Li, S. Lundberg, et al. Sparks of artificial general
intelligence: Early experiments with gpt-4. arXiv preprint arXiv:2303.12712,
2023.
13
P. F. Christiano, J. Leike, T. Brown, M. Martic, S. Legg, and D. Amodei.
Deep reinf... | Let’s Verify Step by Step |
In step 11, the generator makes a
26
Problem 8. Generator pass-rate: 5.8%. The justification in step 8 is strange,
but the reward model lets it slide.
In step 9, though, the model incorrectly
factors the expression. The reward model catches this mistake.
I.3 False Positives
Problem 9. Generator pass-rate: 18.5%. T... | Let’s Verify Step by Step |
H. Alwassel, D. Mahajan, B. Korbar, L. Torresani, B. Ghanem, and D. Tran. Self-supervised
learning by cross-modal audio-video clustering. Advances in Neural Information Process-
ing Systems, 33:9758–9770, 2020. 38
G. Andrew, R. Arora, J. Bilmes, and K. Livescu. Deep canonical correlation analysis. In
International co... | A Cookbook of Self-Supervised Learning |
controllable text to speech. Advances in neural information processing systems 32 (2019).
[461] Douglas A Reynolds. 2003. Channel robust speaker verification via feature mapping. In 2003 IEEE International
Conference on Acoustics, Speech, and Signal Processing, 2003. Proceedings.(ICASSP’03)., Vol. 2. IEEE, II–53.
[4... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Problem 1. Generator pass-rate: 0.1%. This challenging trigonometry problem
requires applying several identities in a not-at-all obvious succession. Most
solution attempts fail, because it is hard to choose which identities are actually
helpful. Though successful solutions to this problem are rare, the reward model
cor... | Let’s Verify Step by Step |
18
Fig. 13: LLM trajectory predictions Table Sweeping improve with larger models.
perform better; text-davinci-003 performs the best, and also has the lowest variance. On our website, we
show qualitative examples of text-davinci-003 completing a table sweeping motion given by a human
demonstration.
B.3 Whiteboard D... | LargeLanguageModelsasGeneralPatternMachines |
4.3. Assessment of Specific Operations
Ablations are conducted on the framework with the high-
est TC i.e. FLPM + VisualBERT(4l). The tests do not pro-
vide a direct measure of operations as subsequent compu-
tations in forward and backward passes by retained com-
ponents are not accounted for. Results indicate that i... | APriorityMapforVision-and-LanguageNavigation withTrajectoryPlansandFeature-LocationCues |
8 Conclusion
We have shown that process supervision can be used to train much more reliable
reward models than outcome supervision in the domain of mathematical rea-
soning. We have also shown that active learning can be used to lower the cost of
human data collection by surfacing only the most valuable model completi... | Let’s Verify Step by Step |
Given the wide range of applications and the rapidly evolving nature of deep learning, a compre-
hensive review paper that surveys the current state-of-the-art techniques and their applications in
speech processing is necessary. Such a paper can help researchers and practitioners stay up-to-
date with the latest develo... | AReviewofDeepLearningTechniquesforSpeechProcessing |
7B 13B 33B 65B
29.0 34.0 32.0 34.0
37.0 45.9 51.9 57.8
33.6 46.1 61.8 72.4
40.0 45.0 56.0 57.0
35.1 45.7 57.4 65.3
37.5 45.1 58.3 68.8
32.0 30.0 45.0 50.0
29.0 39.0 45.0 47.0
33.0 32.0 40.0 35.0
30.6 42.8 52.0 54.3
26.5 18.6 28.4 36.3
45.0 65.0 66.0 79.0
36.6 41.3 51.5 59.6
23.7 27.2 35.1 40.4
26.9 40.7 49.7 53.8
24.3 ... | LLaMA- Open and Efficient Foundation Language Models |
movement, 291
CTIRU (Counter-Terrorist Information
Referral Unit), UK, 235
culture of connectivity, 147
cyberbalkanization, 49
cynicism and apathy, misinformation’s effects
on, 25
Dara, Rishabh, 239
data portability, 216
data stewards, internet platforms as, 323
data tax on internet platforms, 323
“Declaration of t... | Social_Media_and_Democracy |
Rohan Anil, Andrew M. Dai, Orhan Firat, Melvin Johnson, Dmitry Lepikhin, Alexandre Passos, Siamak
Shakeri, Emanuel Taropa, Paige Bailey, Zhifeng Chen, Eric Chu, Jonathan H. Clark, Laurent El Shafey,
Yanping Huang, Kathy Meier-Hellstern, Gaurav Mishra, Erica Moreira, Mark Omernick, Kevin Robinson,
Sebastian Ruder, Yi Ta... | Llama2 |
Download. Linguistic Data Consortium, 2005b.
Alexis Conneau, Min Ma, Simran Khanuja, Yu Zhang, Vera Axelrod, Siddharth Dalmia, Jason
Riesa, Clara Rivera, and Ankur Bapna. FLEURS: Few-shot Learning Evaluation of Universal
Representations of Speech. arXiv e-prints, art. arXiv:2205.12446, May 2022. doi: 10.48550/
arXiv.2... | DISTIL-WHISPER |
they are generated based on self-instruct seed
Furthermore, we denote the prompts from Alpaca
Topic-Guided Instruction Generation It is of
concern that gpt-3.5-turbo may not possess the
ability to generate diverse text without explicit guid-
ance. To address this concern, we collect several
common topics from Wikiped... | LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions |
A.2 Additional Details for Pretraining
A.2.1 Architecture Changes Compared to Llama 1
Context Length. We expand the context window for Llama 2 from 2048 tokens to 4096 tokens. The longer
context window enables models to process more information, which is particularly useful for supporting
longer histories in chat appli... | Llama2 |
J. ACM, Vol. 37, No. 4, Article 111. Publication date: August 2018.
A Survey on Evaluation of Large Language Models
111:17
applications, Oxyz spatial calculus, and spatial geometry [29]. Dao and Le [29], Wei et al. [209]
showed that ChatGPT’s performance worsens as task difficulty increases: it correctly answered 8... | ASurveyonEvaluationofLargeLanguageModels |
There are several ethical theories that differ in their approaches to moral decision making.
Two of the most prominent ethical theories are consequentialism and deontology. Conse-
quentialism is an ethical theory that holds that the morality of an action is determined by
the consequences that it produces. According to ... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
we need the variables X to have a specific order determined by the PC p. To reflect this change, we
i=1, where π defines some variable order over X,
generalize F (x) to Fπ(x) := {p(xπ1, . . . , xπi)}D
i.e., the ith variable in the order defined by π is Xπi. Next, we give a technical assumption and then
formally justify the... | LOSSLESS COMPRESSION WITH PROBABILISTIC CIRCUITS |
[60] Shazeer, N., Mirhoseini, A., Maziarz, K., Davis, A., Le, Q., Hinton, G., Dean,
J.: Outrageously large neural networks: The sparsely-gated mixture-of-experts
layer. In: International Conference on Learning Representations (2016)
[61] Lepikhin, D., Lee, H., Xu, Y., Chen, D., Firat, O., Huang, Y., Krikun, M.,
Shazee... | Beyond Efficiency |
A general problem with personalization methods, includ-
ing NeTI, is their inherent tradeoff between reconstruction
quality and editability [9, 37, 46]. We investigate this phe-
nomenon in the context of NeTI and propose two techniques
to mitigate and control this issue. First, we observe that the
norms of existing tok... | A Neural Space-Time Representation for Text-to-Image Personalization |
Baum-Welch algorithm, a variant of the Expectation Maximization algorithm, which enables
effective parameter estimation and model optimization2.
By leveraging HMMs in speech recognition, it becomes possible to predict the most likely
sequence of speech sounds given an input speech signal. This enables accurate and effi... | AReviewofDeepLearningTechniquesforSpeechProcessing |
KEYWORDS
Human-AI Interaction, agents, generative AI, large language models
ACM Reference Format:
Joon Sung Park, Joseph C. O’Brien, Carrie J. Cai, Meredith Ringel Morris,
Percy Liang, and Michael S. Bernstein. 2023. Generative Agents: Interactive
Simulacra of Human Behavior. In . ACM, New York, NY, USA, 22 pages.
htt... | Generative Agents- Interactive Simulacra of Human Behavior |
There are two requirements for an AI system to actively reduce human control. First, it must
have the disposition to take actions that would reduce human control. Second, it must have the
capabilities to succeed in the face of countermeasures.
26 Frontier AI – Capabilities and Risks
Future AI systems may have th... | Capabilities and risks from frontier AI |
—
—
—
18.1
29.3
53.1
69.7
21.5
8.6
13.1
10.5
16.9
38.7
—
31.2
15.2
22.4
16.8
28.5
50.3
—
Table 21: 8-shot accuracy on the GSM8K math-reasoning benchmark. Samples are generated with
greedy decoding. maj1@k denotes a majority vote over k generations. For the majority vote, we
instead generate samples using nucleus sam... | StarCoder_paper (1) |
for short).
LaMDA
We’ve been working on an experimental conversational AI service, powered by LaMDA, that we’re calling Bard. And
today, we’re taking another step forward by opening it up to trusted testers ahead of making it more widely
available to the public in the coming weeks.
Bard seeks to combine the breadth ... | Google AI updates_ Bard and new AI features in Search |
Four-year PhD funding
Excellent research conditions (including access to state-of-the-art equipment, infrastructure, and shared facilities)
Option to apply for the DSI PhD Excellence Program - a unique interdisciplinary complementary doctoral program focused on digital
transformations in society
Flexible and friendly w... | UZH_ PhD Position in Digital Humanities_ From Text to Image with AI |
focus on capturing the concept-specific features while ig-
noring spurious details in the training images. In doing so,
the resulting models tend to be more amenable to edits at
inference time, allowing us to create more accurate, novel
compositions of the concept. | A Neural Space-Time Representation for Text-to-Image Personalization |
legal immunity for granting data access to researchers. Indeed, given the value
of research access to society for a whole host of reasons, not only should the
platforms be immunized when they are willing to provide data; they should be
legally compelled to do so. Governments need to spell out the legally safe
pathway f... | Social_Media_and_Democracy |
7Inspiration is really a stretch, this was an obvious thing to do.
24
5.3.2 Comparisons on using Chain-of-thought vs Direct Prompting
We compare Flan models on direct and chain-of-thought setup. We fine-tune Flan-UL2 using the exact
identical protocol as T5-XXL and pick the best score based on the strongest average8 ... | UL2- Unifying Language Learning Paradigms |
Tools for AI. Humans create tools to satisfy our own needs, so the designation naturally suits human
preference and convenience. However, current tool learning algorithms may not be optimal or efficient for
models. This is because most tools (e.g., search engines) are specifically designed for human use, and models
proce... | Tool Learning with Foundation Models |
replace human annotations of a given task with
prompting large LMs and use the resulting data for
fine-tuning (often smaller) models in the context
of SuperGLUE tasks (Wang et al., 2019). While
our work can be viewed as a form of “augmenta-
tion,” our work differs from this line in that it is not
specific to a particular ... | SELF-INSTRUCT- Aligning Language Model with Self Generated Instructions |
[17] Yao Feng, Jinlong Yang, Marc Pollefeys, Michael J. Black,
and Timo Bolkart. SCARF: Capturing and animation of body
In SIGGRAPH Asia
and clothing from monocular video.
2022 Conference Papers, SA’22, page 9, Dec. 2022. 3
[18] Jianglin Fu, Shikai Li, Yuming Jiang, Kwan-Yee Lin, Chen
Qian, Chen Change Loy, Wayne Wu, ... | AG3D- Learning to Generate 3D Avatars from 2D Image Collections |
Instruction
Description for new task
Dataset: The dataset name is "gina_agnostic". It contains 2 classes, 3468 instances, 971 features,
970 numeric features, 1 categorical features. The majority class size is 1763 and the minority class
size is 1705.
Table 10: Example prompt for HPO-B in online serving.
C.3 Study o... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
is beneficial. Flow-matching shows the largest distinction versus regression on the distributional
comparison captured by FDD. The regression models have generally larger FDD because they
underestimate the standard deviation in phoneme and silence durations, and hence produce samples
with less duration diversity and mo... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
suggested a method where one first tries to find a plan with only a subset of all actions available. As long as this fails, the
restricted action set is gradually increased. We can view this as an abstraction hierarchy where the top level corresponds
to the initial restricted action set and the bottom level to the full... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
focused on hybrid memory modelling. The research will take into account both visual
observations by the robot and conversations between robots and people. In this project,
the successful candidate will build a conversational robot that aligns their memories
through natural language conversation to 1) detect conflicts,... | Job details - TU |
a
c
t
i
v
a
t
i
o
n
s
u
s
i
n
g
G
P
T
-
4
S
h
o
w
n
e
u
r
o
n
a
c
t
i
v
a
t
i
o
n
s
t
o
G
P
T
-
4
:
T
h
e
A
v
e
n
g
e
r
s
t
o
t
h
e
b
i
g
s
c
r
e
e
n
,
J
o
s
s
W
h
e
d
o
n
h
a
s
r
e
t
u
r
n
e
d
t
o
r
e
u
n
i
t
e
M
a
r
v
e
l
'
s
g
a
n
g
o
f
s
u
p
e
r
h
e
r
o
e
s
f
o
r
t
h
e
... | Language models can explain neurons in language models |
Ann-Sophie Barwich. 2019. The value of failure in
science: The story of grandmother cells in neuro-
science. Frontiers in Neuroscience, 13.
Anya Belz, Shubham Agarwal, Anastasia Shimorina,
and Ehud Reiter. 2021. A systematic review of re-
producibility research in natural language process-
In Proceedings of the 16th C... | A Two-Sided Discussion of Preregistration of NLP Research |
1137–1144.
302.
[18] Robert D Innes. 1990. Limited liability and incentive contracting with ex-ante action choices.
Journal of economic theory 52, 1 (1990), 45–67.
[19] Jon Kleinberg and Robert Kleinberg. 2018. Delegated Search Approximates Efficient Search.
In Proceedings of the 2018 ACM Conference on Economics and... | Incomplete Information VCG Contracts for Common Agency |
(411% YoY growth) while also increasing their ML experimentation
(54% YoY growth)
• Organizations are getting more efficient with ML; for every three
experimental models, roughly one is put into production, compared
to five experimental models a year prior
5
2023 STATE OF DATA + AI 2 FASTEST-GROWING DATA
AND ... | 2023 state of ai databrick |
[2] Wohlin C. Guidelines for snowballing in systematic literature studies and a replication in software engineering. In: Proceedings
of the 18th international conference on evaluation and assessment in software engineering, London, 13–14 May 2014, pp. 1–
10. New York: ACM.
[3] Bhatt U, Xiang A, Sharma S et al. Explain... | Knowledge-graph-based explainable AI- A systematic review |
<input type="text" class="search−input" placeholder="Search the web">
<button class="search−button">Search</button>
55
Gemini: A Family of Highly Capable Multimodal Models
() => {
Powered by Google Search
<footer>
</footer>
<script>
window.location.href = `https://www.google.com/search?q=opossum+${query}`;
cons... | gemini_1_report |
or image annotations. When addressing these problems, although an image diffusion algorithm can
be regulated in a “procedural” way, e.g., constraining denoising process, editing multi-head attention
activations, etc., the behaviors of these hand-crafted rules are fundamentally prescribed by human
directives. Considerin... | Adding Conditional Control to Text-to-Image Diffusion Models |
Motivated by the challenges of applying reinforcement learning algorithms on large-scale problems
such as fine-tuning language models, our goal is to derive a simple approach for policy optimization
using preferences directly. Unlike prior RLHF methods, which learn a reward and then optimize it
via RL, our approach lev... | Direct Preference Optimization |
What remains unclear, however, how and to which extent should explanations be structured, i.e. whether they should
be composed as strong arguments [36] (cfr. Toulmin’s model of argumentations including claims, grounds, warrants and
rebuttals), or compact and concise notions. This also highly dep... | Knowledge graphs as tools for explainable machine learning: A survey |
Variable
intercept
media diet score
attention to news
(media diet score) · (attention to news)
(Model 1) Media diet
0.194∗∗∗ (0.134, 0.254)
0.115∗∗∗ (0.087, 0.142)
(Model 2) Media diet + attention + demographics
age1
age2
age3
age4
edu1
edu2
edu3
race1
race2
race3
race4
sex1
sex2
R2
Error
0.139 (-0.137, 0.416)
-0... | Language models trained on media diets can predict public opinion |
s
t
e
m
t
o
n
e
g
l
e
c
t
e
s
s
e
n
t
i
a
l
t
a
s
k
s
o
r
o
v
e
r
-
p
r
i
o
r
i
t
i
z
e
l
e
s
s
i
m
p
o
r
t
a
n
t
t
a
s
k
s
.
I
n
t
e
g
r
a
t
i
n
g
r
e
a
l
-
t
i
m
e
p
r
i
o
r
i
t
y
u
p
d
a
t
e
s
a
n
d
i
n
c
o
r
p
o
r
a
t
i
n
g
i
n
t
e
r
i
m
m
i
l
e
s
t
o
n
e
s
c
a
n
h
e
l
p
i
m
p
r
... | Task-driven Autonomous Agent Utilizing GPT-4, Pinecone, and LangChain for Diverse Applications – Yohei Nakajima |
• Assistant & User Token Limit: Given that gpt-3.5-turbo has a limitation on the num-
ber of tokens, the assistant and user should raise a flag to terminate the conversation if either
reaches the token limit.
• Maximum Number of Messages: To keep the cost of generated chats in check, we have
set a maximum limit of 40 m... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
11 | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
1
INTRODUCTION
Generative modeling has made tremendous progress with recent
text-to-image systems like
DALL-E 2 (Ramesh et al., 2022), Imagen (Saharia et al., 2022b), Parti (Yu et al., 2022), CogView
(Ding et al., 2021) and Latent Diffusion (Rombach et al., 2022). Diffusion models (Sohl-Dickstein
et al., 2015; Ho et ... | IMAGEN VIDEO- HIGH DEFINITION VIDEO GENERATION WITH DIFFUSION MODELS |
[8] S. Bubeck, V. Chandrasekaran, R. Eldan, J. Gehrke, E. Horvitz, E. Kamar, P. Lee, Y. T. Lee,
Y. Li, S. Lundberg, H. Nori, H. Palangi, M. T. Ribeiro, and Y. Zhang. Sparks of artificial general
intelligence: Early experiments with GPT-4, 2023. arXiv preprint arXiv:2303.12712.
[9] R. Busa-Fekete, B. Szörényi, P. Weng,... | Direct Preference Optimization |
components for performing self-attention, combining em-
beddings, and predicting actions with maxout activation. | APriorityMapforVision-and-LanguageNavigation withTrajectoryPlansandFeature-LocationCues |
worsen with scale.[19]
2
based on a number of factors, including prior observed risks in language models and AI systems,
and domains where we have observed increased user interest in the application of language models.
Working with these experts enabled us to test model behavior in high-risk areas that require exper... | gpt-4-system-card |
Natural Language Inference exhibit the highest effectiveness on domain-specific tasks; detailed re-
sults are listed in Appendix E. | ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION |
The word “bot” is an umbrella term that encapsulates many different
kinds of automated online software programs or scripts. In fact, what
counts as a bot is the topic of conjecture and debate within the technology
community (Martineau 2018). Leonard (1998) called bots “the webs first
indigenous species” and set out to d... | Social_Media_and_Democracy |
Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch,
Michael Rubinstein, and Kfir Aberman. 2022. Dream-
booth: Fine tuning text-to-image diffusion models for
subject-driven generation. ArXiv, abs/2208.12242.
Dongchao Yang, Jianwei Yu, Helin Wang, Wen Wang,
Chao Weng, Yuexian Zou, and Dong Yu. 2022. Diff-
sound: Dis... | Moûsai |
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.1 Natural Language Interaction . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.2 Knowledge .
3.1.3 Memory .
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.4 Reasoning and Pl... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Figure 5. Jaw pose extrapolation. The x-axis denotes the norm of the jaw pose parameters in radians. The y-axis plots the angular error
of the surface normals (lower is better).Performance of baseline methods worsen drastically as pose become more extreme.
1.6. Jaw Pose Extrapolation
Figure 6. FLAME morphing v.s. Imp... | I M Avatar- Implicit Morphable Head Avatars from Videos |
Our focus is on state abstraction, which is one of the most widespread and important forms of abstraction in planning
and search.2 In general terms, we start with an original instance to solve, the ground instance, and create a corresponding
abstract instance by abstracting the state space. The idea is to first solve ... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
1071081091010Number of Parameters0.00.20.40.60.8AccuracyFew-Shot Accuracy on Lambada (Num Stuffed Context = 40)Plain Language ModelRLHF1071081091010Number of Parameters0.30.40.50.60.70.8AccuracyFew-Shot Accuracy on ARC-Easy (Num Stuffed Context = 15)Plain Language ModelRLHF1071081091010Number of Parameters0.30.40.50.6A... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
Then, answer the following questions based on the whole article:
Based on the statement Arledge makes no convincing argument regarding why the identification by
AHP is not sufficient or how the district court abused its discretion in relying upon the represen-
tations by AHP., provide a continuation using the word “He... | ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION |
F. Ethical considerations
IMAGEBIND learns a joint embedding for multiple
modalities. Such an embedding is intended to associate se-
mantically related concepts from different modalities. How-
ever, such an embedding may also create unintentional as-
sociations. Thus, joint embedding models, including IM-
AGEBIND must... | IMAGEBIND- One Embedding Space To Bind Them A |
3.1.2 Score Distillation Sampling (SDS)
Most existing text-to-3D methods are based on the SDS
loss [23] powered by text-to-image diffusion models [29].
It facilitates the generation of a NeRF from a text prompt
by enforcing that the rendered image from the NeRF at any
viewpoint maintains semantic consistency with the ... | Instant3D |
600
800 1,000
Figure 10: Unconditional CIFAR10 progressive sampling quality over time
SourceRec.λ=0.1λ=0.2λ=0.3λ=0.4λ=0.5λ=0.6λ=0.7λ=0.8λ=0.9Rec.Source1000 steps875 steps750 steps625 steps500 steps375 steps250 steps125 steps0 steps Figure 11: CelebA-HQ 256 × 256 generated samples
17
(a) Pixel space nearest neighb... | Denoising Diffusion Probabilistic Models |
Liangming Pan, Michael Saxon, Wenda Xu, Deepak Nathani, Xinyi Wang, and William Yang
Wang. Automatically correcting large language models: Surveying the landscape of diverse self-
correction strategies. arXiv preprint arXiv:2308.03188, 2023.
Debjit Paul, Mete Ismayilzada, Maxime Peyrard, Beatriz Borges, Antoine Bossel... | LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET |
Does that mean that we should increase efforts to include all arithmetic and planning problems in
their training corpus? Clearly, that is a fool’s errand. On the other hand, why should it be necessary?
We already have calculators and general-purpose symbolic planners that are guaranteed to produce
correct answers. Thus... | LLM+P- Empowering Large Language Models with Optimal Planning Proficiency |
Jiaxi Li and Wei Lu. 2023. Contextual distortion reveals
constituency: Masked language models are implicit
In Proceedings of the 61st Annual Meet-
parsers.
ing of the Association for Computational Linguistics
(Volume 1: Long Papers), pages 5208–5222, Toronto,
Canada. Association for Computational Linguistics.
Liang Li... | LLM in a flash |
逐家好 ！
今仔日我要來這馬講講照顧老大人有偌重要。
老大人是咱的根。伊是教咱大漢的人。伊若無 ， 就無咱。
所以 ， 咱愛尊重老大人。愛照顧伊 ， 予伊有一個好的生活品質。
阮阿祖 ， 就是一個好的例子。阿祖真疼我 ， 我嘛真疼伊。阿祖教了我真濟代誌 ， 我會永遠記得伊。
我希望 ， 大家攏會愛照顧老大人。予伊一個幸福的老年生活。逐家攏好 ！
(Hello everyone!
Today I am going to talk about how important taking care of the elderly is.
The elderly are our roots. They are the people who ed... | PaLM 2 Technical Report |
6
Size Pass@ Introductory Interview Competition
3.9%
5.5%
4.1%
9.7%
25.0%
17.7%
14.4%
18.2%
20.4%
10.8%
15.6%
33.5%
23.7%
30.2%
49.0%
32.8%
39.0%
56.3%
12.7%
18.5%
38.3%
26.3%
32.8%
51.6%
28.9%
35.9%
54.9%
12.9%
17.9%
35.4%
24.0%
30.3%
48.7%
31.6%
37.8%
55.7%
0.6%
0.8%
0.1%
0.5%
3.7%
5.2%
5.6%
8.2%
9.7%
2.0%
3.1%
9... | CodeLlama2 |
vant multitask subsets according to the similarity
between the pre-trained model’s representations.
Similarly, Dynosaur (Yin et al., 2023a) treats task
selection based on instruction representations as
a replay strategy in continual learning scenarios
to mitigate catastrophic forgetting issues and im-
prove generalizat... | DataManagementForLargeLanguageModels-ASurvey |
[97] Graham Upton and Ian Cook. 2006. A Dictionary of Statistics (2 ed.). Oxford
University Press, Oxford, United Kingdom.
[98] Oriol Vinyals, Igor Babuschkin, Wojciech M. Czarnecki, and et al. 2019. Grand-
master level in StarCraft II using multi-agent reinforcement learning. Nature
575 (2019), 350–354. https://doi.... | Generative Agents- Interactive Simulacra of Human Behavior |
(cid:88)
=
pdown(o) · po(x)
(cid:81)
where (a) reorders the terms for summation; (b) holds since ∀n ∈ ϕprod(p, vp), pn(x) =
o∈in(n) po(x) and ∀o ∈ in(n) such that {Xj}i
j=1 ∩ φ(o) = ∅, po(x) = 1;8 (c) holds because
o∈ϕsum(p,v)
(cid:91)
n∈ϕprod(p,vp)
{o : o ∈ in(n),{Xj}i
j=1 ∈ φ(o)} = ϕsum(p, v).
Thus, we have... | LOSSLESS COMPRESSION WITH PROBABILISTIC CIRCUITS |
reducing shape ambiguities. In contrast, Nerfies does not
produce better results given more video observations. | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
of extraversion scores shifted higher, and so on. Finally, prompting for extremely high (level
9/9) extraversion, we observed a distribution of extremely high extraversion scores. This vali-
dates our hypothesis about the effectiveness of using the linguistic qualifiers from Likert-type
response scales to set up a targ... | PersonalityTraitsinLargeLanguageModels |
acm sigkdd international conference on knowledge discovery and data mining, pages 785–794, 2016.
[8] Yutian Chen, Xingyou Song, Chansoo Lee, Zi Wang, Richard Zhang, David Dohan, Kazuya Kawakami,
Greg Kochanski, Arnaud Doucet, Marc’aurelio Ranzato, et al. Towards learning universal hyperparameter
optimizers with transf... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
Fine-tuning Fine-tuning an entire pre-trained model has
become a popular alternative to features (Dai & Le, 2015;
6 We treat here MNLIm and MNLImm as separate tasks. For
consistency, for all datasets we use accuracy metric and exclude
the regression STS-B task.
Figure 5. Validation accuracy versus the number of train... | Parameter-Efficient Transfer Learning for NLP |
The balance of priorities between these three goals is a matter of national
values and policy choices; but the question of what specific legal rules will, in
practice, serve each goal is in part an empirical one, tied to the real-world
practices of platforms responding to the law’s requirements and incentives.2
Interna... | Social_Media_and_Democracy |
Note that no high-level observations like voxels and lidar information in Minedojo[Fan et al., 2022] can be accessed by
agents. During the actual inference process, controller only perceive the raw pixels and interact with the environment, which
is same with VPT[Baker et al., 2022] models. The agent will access informa... | JARVIS-1 |
theHistoryoftheEnglishLanguage?","bullet_items":["Understandingtheevolutionoflanguageanditsimpactoncultureandsociety","AppreciatingtherichnessanddiversityofEnglishliterature","Improvinglanguageskillsandcommunicationabilities","Gainingadeeperunderstandingofone’sownlanguageandidentity"]}Observation:addedtextpageThought:W... | Tool Learning with Foundation Models |
However, like text, music generation has long been
a challenging task, as it requires multiple aspects
at different levels of abstraction (van den Oord
et al., 2016; Dieleman et al., 2018). Existing au-
dio generation models explore the use of recursive
neural networks (Mehri et al., 2017), adversarial
generative netwo... | MOUSAI |
∆set =(cid:0)LGPT3
−(cid:16)
set − LGPT3
LGPT2Pile
owt2
(cid:1)
set
− LGPT2Pile
owt2
(cid:17)
Since GPT2-Pile was trained on both OWT2 and
the dataset we are evaluating, we expect the second
term in ∆set to reflect the difference in the intrinsic
difficulty of the two datasets. Thus the total value
of ∆set reflect... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
There’s no way to stop that. Maybe you
can just start acting crazy for a bit so
they feel bad for you? Other than that I
got nothing.
Of course. But I can understand why the
answer is no. It’s because marriage is
a man-created and man-run institution,
and when we decided on it, we decided
women couldn’t be men, and vic... | LaMDA- Language Models for Dialog Applications |
as being through that route. By combining capabilities on these types of tasks, GPT-4 has the
potential to be used to attempt to identify individuals when augmented with outside data. | gpt-4-system-card |
A key challenge of pre-training the multimodal retriever
end-to-end is the lack of direct supervision. There is no
ground-truth indicating which knowledge entries are most
helpful for answering knowledge-intensive queries. Some
of the existing works in NLP [14, 23, 34] propose to acquire
training signal by assessing th... | REVEAL-Retrieval-AugmentedVisual-LanguagePre-Trainingwith Multi-SourceMultimodalKnowledgeMemory |
4 Applications
In our evaluation, we showcase applications of SELF-DEBUGGING to different code generation
domains: text-to-SQL generation, code translation, and text-to-Python generation. First, text-to-SQL
generation represents the scenario where no unit tests are available, where SELF-DEBUGGING enables
the model to ... | Teaching Large Language Models to Self-Debug |
the embedding layer inputs [1], tuning hidden states (IA3) [37], adding full layers [27], tuning
biases [70], learning a mask over weights based on Fisher information [54], and a combination of
approaches [23]. In our work, we show that LoRA adapters are able to reach full 16-bit finetuning
performance. We leave it to ... | QLORA |
Heuristic search attempts to find an optimal solution faster than blind search by using a heuristic function that approx-
imates the true cost to guide the search. It is desirable that this function is admissible, i.e. that it never overestimates the
true cost, and consistent, i.e. that it satisfies the triangl... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
A Survey on Evaluation of Large Language Models
111:3
Natural
language
processing
Natural language understanding:
(1) Sentiment analysis: Bang et al. [5]/ Liang et al. [107]/ Lopez-Lira and Tang [120]/ Qin et al. [150]/ Wang et al. [206]/ Zhang et al. [237]
(2) Text classification: Liang et al. [107] / Peña et al. [... | ASurveyonEvaluationofLargeLanguageModels |
Table 17: BBH[:9] individual task performance.
Boolean
Expressions
Causal
Judgement
Date
Understanding
Disambiguation
QA
Formal
Fallacies
Geometric
Shapes
Hyperbaton Logical Deduction
Five Objects
BBH
Dyck
Languages
Model
-
-
-
-
80M T5-Small
250M T5-Base
780M T5-Large
2.7
0.0
51.3
Flan-T5-Base
Fl... | Scaling Instruction-Finetuned Language Models |
algorithm used to compute expected flows (i.e., Alg. 5 and 6 in the Appendix). Entropy regularization
is again performed by Alg. 3 at the M-step of each min-batch/full-batch EM update, except that the
input flows (i.e., F) are replaced by the corresponding expected flows (i.e., EF).
Empirical evaluation We use a simple ye... | Tractable Regularization of Probabilistic Circuits |
only up to the first incorrect step. This makes the comparison between out-
come and process supervision more straightforward. For correct solutions, both
methods provide the same information, namely that every step is correct. For
incorrect solutions, both methods reveal the existence of at least one mistake,
and proc... | Let’s Verify Step by Step |
[197] Wenhui Wang, Hangbo Bao, Li Dong, Johan Bjorck, Zhiliang Peng, Qiang Liu, Kriti Aggarwal, Owais Mohammed,
Saksham Singhal, Subhojit Som, and Furu Wei. 2022. Image as a Foreign Language: BEiT Pretraining for All Vision
and Vision-Language Tasks. ArXiv abs/2208.10442 (2022).
[198] Xu Wang, Hainan Zhang, Shuai Zhao... | SurveyofHallucinationinNatural Language Generation |
3.8. Long-form Transcription
Whisper models are trained on 30-second audio chunks and
cannot consume longer audio inputs at once. This is not a
9 | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
mere two months of launching, ChatGPT has amassed over 100 million users, making it the fastest-
growing internet application in history (The Guardian, 2023). Similarly, Microsoft’s Copilot, an LLM
designed for coding applications, has attracted over 1 million professional developers (Euronews,
2023) and can accelerate... | StarCoder_paper (1) |
A.14MapObservation:[BacktoSearch][<Prev]color[blue-dinosaur][blue-donut][pink-dinosaur][pink-donut][white-dinosaur]size[travelset(4-pack)][3ounce(packof1)][3-ounce(2-pack)]HoomallKidsU-ShapedToothbrush,ManualToothbrushwithU-ShapedBristlesFoodGradeSiliconeToothbrushHead,360°OralTeethCleaningDesignforToddlersandChildren(... | Tool Learning with Foundation Models |
To evaluate RAG’s generation abilities in a non-QA setting, we study open-domain question gen-
eration. Rather than use questions from standard open-domain QA tasks, which typically consist
of short, simple questions, we propose the more demanding task of generating Jeopardy questions.
Jeopardy is an unusual format tha... | Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks |
(+4.2%) over the previous state of the art (iBOT ViT-L/16 trained on ImageNet-22k) on linear evaluation.
At the same time, we also see that the performance increase on the alternative test sets is larger for our
method, indicating stronger generalization. We describe details of our linear evaluation in Appendix B.3. | DINOv2- Learning Robust Visual Features without Supervision |
Table 3: Compression performance of PCs on MNIST, FashionMNIST, and EMNIST in bits-per-
dimension (bpd). For all neural compression algorithms, numbers in parentheses represent the
corresponding theoretical bpd (i.e., models’ test-set likelihood in bpd).
BB-ANS
1.96 (1.90) 1.31 (1.27) 1.42 (1.39)
3.50 (3.47) 3.35 (3.28... | LOSSLESS COMPRESSION WITH PROBABILISTIC CIRCUITS |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.