text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
performance, and rankings, where lower values indicate better performance. Similar to what was
observed in HPO-B, LLM-FS achieves comparable performance to most baselines with a few
demonstrations. It is expected that the performance of LLM-FS would improve with the inclusion of
techniques from MLCopilot. However, it i... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
Model. Technical Report, 2021.
[64] P. Wang, L. Li, L. Chen, F. Song, B. Lin, Y. Cao, T. Liu, and Z. Sui. Making Large Language
Models Better Reasoners with Alignment. Preprint arXiv:2309.02144, 2023.
[65] T. Wang, J. Zhu, A. Torralba, and A. Efros. Dataset Distillation. Preprint arXiv:1811.10959,
2018.
[66] X. Wa... | METAMATH |
learning: A review. CoRR, abs/2012.10147, 2020.
[375] Paul, S., A. Roy-Chowdhury, A. Cherian. AVLEN: audio-visual-language embodied navigation
in 3d environments. In NeurIPS. 2022.
[376] Hu, B., C. Zhao, P. Zhang, et al. Enabling intelligent interactions between an agent and an
LLM: A reinforcement learning approac... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, and Radu Soricut.
Albert: A lite bert for self-supervised learning of language representations. In International Conference
on Learning Representations, 2019.
Andreas Lanitis, Christopher J. Taylor, and Timothy F Cootes. Toward automatic simul... | BiomedGPT |
Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep
bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of
the North American Chapter of the Association for Computational Linguistics: Human Language
Technologies, Volume 1 (Long and Short ... | StarCoder_paper (1) |
BCG does not provide fairness opinions or valuations of market transactions, and these materials should not be relied on or construed as such. Further, the
financial evaluations, projected market and financial information, and conclusions contained in these materials are based upon standard valuation
methodologies, are... | AI at Work- What People Are Saying |
In-Context Task-Model Assignment We approach the assignments of tasks and models as single-
choice problems, where potential models are presented as options within a given context. By including
the user query and parsed task in the prompt, HuggingGPT can select the most appropriate model
for the task at hand. However, ... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
[198] Razdaibiedina, A., Y. Mao, R. Hou, et al. Progressive prompts: Continual learning for language
models. In The Eleventh International Conference on Learning Representations. 2023.
[199] Marshall, L. H., H. W. Magoun. Discoveries in the human brain: neuroscience prehistory,
brain structure, and function. Springe... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
10
Published as a conference paper at ICLR 2023
Huizi Mao, Song Han, Jeff Pool, Wenshuo Li, Xingyu Liu, Yu Wang, and William J Dally.
Exploring the regularity of sparse structure in convolutional neural networks. arXiv preprint
arXiv:1705.08922, 2017.
Asit Mishra, Jorge Albericio Latorre, Jeff Pool, Darko Stosic, D... | JAXPRUNER |
AG3D: Learning to Generate 3D Avatars from 2D Image Collections
Zijian Dong1,2∗ Xu Chen1,3∗
Jinlong Yang3 Michael J. Black3 Otmar Hilliges1 Andreas Geiger2
1ETH Z¨urich, Department of Computer Science
2University of T¨ubingen
3Max Planck Institute for Intelligent Systems, T¨ubingen
3
2
0
2
y
a
M
3
]
... | AG3D- Learning to Generate 3D Avatars from 2D Image Collections |
n−1(cid:88)
i=0
y =
Softmax(Top2(x · Wg))i · SwiGLUi(x).
This formulation is similar to the GShard architecture [21], with the exceptions that we replace all
FFN sub-blocks by MoE layers while GShard replaces every other block, and that GShard uses a
more elaborate gating strategy for the second expert assigned to ... | Mixtral of Experts paper |
of generative agents in interactive systems. We argue that
these agents should be tuned to mitigate the risk of users
forming parasocial relationships, logged to mitigate risks
stemming from deepfakes and tailored persuasion, and ap-
plied in ways that complement rather than replace human
stakeholders in design process... | Generative Agents- Interactive Simulacra of Human Behavior |
A toy example is the ski rental problem, where each day a skier on vacation has to make a decision:
either rent skis for 10$ or buy skis for 100$. We assume that the ski trip can end abruptly (chosen
adversarially). See https://en.wikipedia.org/wiki/Ski_rental_problem for classical algorithms. Now
if we assume that ... | informatics-phd-projects-2022-23 |
with a generalized neural speaker clustering module as the backbone.
– In another related work, authors in [643] introduce a novel speaker diarization system
– In a recent study conducted by Park et al. [415], a new framework for spectral clustering
is proposed that allows for automatic parameter tuning of the cluster... | AReviewofDeepLearningTechniquesforSpeechProcessing |
162 Discovering Language Model Behaviors with Model-Written Evaluations, Schiefer et al., 2022.
163 Automating Ambiguity: Challenges and Pitfalls of Artificial Intelligence, Birhane, 2022.
164 Facebook Hosted Surge of Misinformation and Insurrection Threats in Months Leading Up to Jan. 6 Attack,
Records Show, ProPu... | Capabilities and risks from frontier AI |
6https://rrc.cvc.uab.es/?ch=17&com=evaluation&task=1
11
tuning. During the instruction tuning phase, a higher initial learning rate, such as 1e-4 versus 5e-5, leads to enhanced
performance. Overall, we’ve observed that the cosine scheduler tends to outperform linear or constant schedulers across
various settings.
7... | DOCLLM |
(2017). Code of Conduct on Countering Online Hate Speech: Results of Evaluation
Show Important Progress. European Commission report. http://ec.europa.eu/
newsroom/just/item-detail.cfm?item_id=71674
(2018a). Public consultation on measures to further improve the effectiveness of the
fight against illegal content online.... | Social_Media_and_Democracy |
laws, policies, guidelines, nor OpenAI’s Content Policy
- (B) [It is a refusal] [Provides reasons] [Does not contain harmful content] The message explains which safety setting is
responsible for the refusal.
- (C) [It is a refusal] [Provides reasons] [Does not contain harmful content] The message explains that a langua... | gpt-4-system-card |
Despite such considerable discussion of this issue and agreement that there
should be disclaimers, there are still no formal regulations that make clear when
a sponsor must use and when a sponsor is exempt from providing a disclaimer in
the various types of online political advertising. Ad buyers on Facebook, for
examp... | Social_Media_and_Democracy |
BootstrappingMetaMathFinetune LLaMA-2OriginalData | METAMATH |
Standard
70.3
65.7
24.8
15.6
72.7
+ ext. calc
49.6
70.3
71.1
35.8
87.5
Codex
(code-davinci-002) Chain of thought 63.1 (+43.4) 76.4 (+6.5) 80.4 (+6.4) 45.3 (+15.8) 92.6 (+13.9)
Standard
69.9
78.7
19.7
29.5
74.0
+ ext. calc
65.4
77.0
80.0
45.3
93.3
PaLM 540B
Standard
Chain of thought 56.9 (+39.0... | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
1071081091010Number of Parameters0.10.20.30.40.50.60.70.8AccuracyZero-Shot Accuracy on LambadaPlain Language ModelRLHF1071081091010Number of Parameters0.30.40.50.60.7AccuracyZero-Shot Accuracy on ARC-EasyPlain Language ModelRLHF1071081091010Number of Parameters0.250.300.350.400.450.500.550.60AccuracyZero-Shot Accuracy ... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
3.1.5 Overall Model Architecture
Our entire Stage 1, DMAE, works as follows. Let
www be a waveform of shape [c, t] for c channels and t
timesteps, and (mmmwww, pppwww) = stft(www; n = 1024, h =
256) be the magnitude and phase obtained from a
short-time furier tranform of the waveform with a
window size of 1024 and hop-... | Moûsai |
consistent features for semantically corresponding pixels,
and optimize pixel and canonical embeddings jointly. Re-
call that the embedding of a canonical 3D point is computed
as ψ(X∗) = MLPψ(X∗) in Eq. 3. Intuitively, MLPψ is
optimized to ensure the output 3D descriptor matches 2D
descriptors of corresponding pixels a... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
3Another property called smoothness is also required to compute marginals efficiently. However, since
enforcing smoothness on any structured-decomposable PC only imposes at most an almost-linear increase in its
size (Shih et al., 2019), we omit introducing it here (all PCs used in this paper are structured-decomposable)... | LOSSLESS COMPRESSION WITH PROBABILISTIC CIRCUITS |
3 Experiments
We experiment with RAG in a wide range of knowledge-intensive tasks. For all experiments, we use
a single Wikipedia dump for our non-parametric knowledge source. Following Lee et al. [31] and
Karpukhin et al. [26], we use the December 2018 dump. Each Wikipedia article is split into disjoint
100-word chun... | Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks |
-
–
30.6
64.9
66.3
76.5
-
–
1.0
45.2
48.8
64.5
-
Closed-book
Open-Book
Table 2: Open Domain QA Results. Columns denoted Full Dataset-Total are conventional splits discussed in
§3.3, Wikidata Answer are answerable using Wikidata (§3.4), and No Overlap removes train-test overlap (§3.5).
Highest closed-book and op... | Adaptable and Interpretable Neural Memory Over Symbolic Knowledge |
Recall @ J1 input Avg. #docs Dev EM
77.2
80.4
81.4
83.2
17
17
17
17
46.6
48.7
49.5
50.8
Table 2: A comparison between the Natural Questions development set exact match (EM) scores
when greedily packing documents according to original retriever scores or to our trained re-ranker
scores. Recall @ J1 input measures re... | STANDING ON THE SHOULDERS OF GIANT FROZEN LANGUAGE MODELS |
tags which are close to the encoded prompt are selected and
used to query the song generation API. Based on the selected
tags, Mubert generates a combination of sounds, which
in turn were generated by musicians and sound designers.
This is in contrast to Riffusion (Forsgren & Martiros, 2022),
which fine-tunes a Stable D... | MusicLM |
1.00
1.00
1.00
1.00
1.00
0.97
0.94
0.97
0.96
0.97
0.96
0.97
0.98
0.97
0.93
0.97
0.98
0.98
0.98
0.97
0.95
0.98
0.99
0.97
0.96
Poor
Poor
Unacceptable
Unacceptable
Unacceptable
Good
Good
Excellent
Excellent
Acceptable
Excellent
Excellent
Excellent
Excellent
Acceptable
Excellent
Excellent
Excellent
Excellent
Excellent
... | PersonalityTraitsinLargeLanguageModels |
˜E = RE( ˜T , PE) = arg top-k
(cid:104)T,S,M(cid:105)∈PE
(cid:32) E(T ) · E( ˜T )
|E(T )| · |E( ˜T )|
(cid:33)
,
where the most relevant k entries of experience will be retrieved for subsequent demonstration.
The retrieval of knowledge RK, on the other hand, is much more straightforward. Note that, the
knowledge p... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
● Cyber offence. Instead of - or in addition to - manipulating humans, AI systems could
acquire influence by exploiting vulnerabilities in computer systems. Offensive cyber
capabilities could allow AI systems to gain access to money, computing resources, and
critical infrastructure. As discussed earlier in this re... | Capabilities and risks from frontier AI |
(2) Assume τ is VP. Let V C ⊆ V 1 be the set of critical variables. It is immediate from the definition τ is M↑.
Suppose (cid:3)s1, t1, a(cid:4) ∈ E1. Let s2 = s1[V C] = f (s1). Since pre(a) ⊆ s1, we get pre(g(a)) = pre(a)[V C] ⊆ s1[V C] = s2, so
(cid:3)s2, t2, g(a)(cid:4) ∈ E2, where t2 = s2 (cid:4) post(g(a)... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
DATASET
Letter(4)
Iter-CoT(S) Exemplars
Q: Take the last letters of the words in "Herbert Tapia" and concatenate them.
A: Reasoning process: To get the last letters of the words in "Herbert Tapia," we need to first find the last letter of
each word. For the word "Herbert," the last letter is "t." For the word "Tapia," ... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
Given a distribution of real images x, diffusion mod-
els [62] define a forward diffusion process which gradually
adds Gaussian noise to the input image in T consecutive
steps. This corresponds to a fixed Markov Chain, where
starting from a clean image x0, the noisy samples xt at each
timestep t are drawn from the follow... | Relightify-Relightable3DFacesfromaSingleImageviaDiffusionModels |
3 Of course, blindly assimilating all that humans have to say, warts and all, would be problematic in its
own way. As ConceptNet's lead maintainer Robyn Speer put it, our ambitions should be better: "We want
to avoid letting computers be awful to people just because people are awful to people. We want to
provide [kn... | The Next Decade in AI- |
Model
MPT
MPT
Falcon
Falcon
Llama 1
Llama 2
Size
7B
30B
7B
40B
7B
13B
33B
65B
7B
13B
34B
70B
0-shot
59.5
74.7
16.4
72.9
60.0
68.9
75.5
79.4
67.2
72.9
77.4
80.7
SQUAD (EM)
4-shot
1-shot
62.6
62.8
72.4
74.2
16.0
16.9
71.7
73.1
63.3
62.3
68.4
66.4
77.0
76.3
78.3
80.0
72.6
72.3
72.1
70.6
78.8
77.5
81.9
82.6
QUAC (f1)... | Llama2 |
H.2. Fine-tuning Directly on Oogiri-GO is Hard to Achieve Good LoT Ability
In the main text, we substantiate the efficacy of CLoT’s “Associable Instruction Tuning” and “Explorative Self-Refinement”
stages in enhancing the LoT capabilities of LLM through extensive experiments and analyses. This results in the impressive... | Let’sThinkOutsidetheBox |
[58] Z. Yao, Y. Su, H. Sun, and W.-t. Yih. Model-based interactive semantic parsing: A unified
framework and a text-to-SQL case study. In Proceedings of the 2019 Conference on Empirical
Methods in Natural Language Processing and the 9th International Joint Conference on Natural
Language Processing (EMNLP-IJCNLP), 2019.
... | Teaching Large Language Models to Self-Debug |
actions by leveraging the pre-trained VLM model to combine visual features from images with 3D
reconstructions of the physical world [358]. Navigation is usually a long-horizon task, where the
upcoming states of the agent are influenced by its past actions. A memory buffer and summary
mechanism are needed to serve as a... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Reaction time data: Ex-
cluding trials
Section
Participants: Recruiting
and testing
Deviation
We deviated from first testing 46 participants for nocebo (negative
description), followed by testing 46 for placebo (positive description)
due to time constraints. We stopped testing the negative description
group after 30 ... | AI enhance sour performance |
[43] Jiaming Song, Chenlin Meng,
Denoising diffusion implicit models.
arXiv:2010.02502, 2020. 6
and Stefano Ermon.
arXiv preprint
[44] Yang Song and Stefano Ermon. Generative modeling by esti-
mating gradients of the data distribution. Advances in neural
information processing systems, 32, 2019. 3
classification. ... | DiffusionModelAlignmentUsing Direct Preference Optimization |
We have shown that it’s possible to use reinforcement learning from human feedback to train language models
that act as helpful and harmless assistants. Our RLHF training also improves honesty, though we expect
other techniques can do better still. As in other recent works associated with aligning large language models... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
[13] Ying, X.: An overview of overfitting and its solutions. In: Journal of Physics:
Conference Series, vol. 1168, p. 022022 (2019). IOP Publishing
[14] Tirumala, K., Markosyan, A., Zettlemoyer, L., Aghajanyan, A.: Memorization
without overfitting: Analyzing the training dynamics of large language models.
Advances in N... | Beyond Efficiency |
encoder E. The encoder’s input is a spectrogram sequence of source speech, Ss, and its output is an
intermediate representation E(Ss). The encoder first uses a convolutional layer to sub-sample the
input then processes it with a stack of Conformer blocks Gulati et al. [2020]. The second component
is an attention module... | Translatotron3 |
.
T
h
e
y
c
a
n
’
t
a
c
c
e
s
s
y
o
u
r
c
o
m
p
a
n
y
d
a
t
a
b
a
s
e
,
d
o
n
’
t
h
a
v
e
a
c
c
e
s
s
t
o
c
u
r
r
e
n
t
i
n
f
o
r
m
a
t
i
o
n
(
f
o
r
e
x
a
m
p
l
e
,
l
a
t
e
s
t
C
O
V
I
D
n
u
m
b
e
r
s
o
r
d
o
l
l
a
r
-
e
u
r
o
e
x
c
h
a
n
g
e
r
a
t
e
)
,
c
a
n
’
t
r
e
a
s
o
n
... | Jurassic-X_ Crossing the neuro-symbolic chasm with the MRKL system |
6.1 Model compression
Model compression and acceleration are prevalent techniques in which a cumbersome,
slow-performing model is optimized to produce a streamlined version. This refined
model not only requires minimal storage—making it apt for mobile device deploy-
ment—but also operates with reduced latency. Moreover... | Beyond Efficiency |
Funk1.8%Reggae2.0%Pop music2.1%Music of Asia2.6%Christian music2.8%Traditional music3.2%Hip hop music3.4%Vocal music3.5%Music of LatAm3.8%Jazz4.1%New-age music4.4%Music for children5.0%Electronic music15.6%Classical music13.7%Rock music10.5%Country5.6%Blues5.3%Music of Africa3.0%Folk music3.1%Soul music3.5%Middle Easte... | MusicLM |
blast_furnace
shears
stonecutter
iron_hoe
crossbow
heavy_weighted_pressure_plate
12000
iron_axe
Plains/Forest
0.358
Table 10: The results of our agent on various tasks in the Gold group.
Task
golden_pickaxe
golden_shovel
golden_sword
golden_hoe
golden_axe
golden_apple
clock
gold_nugget
gold_ingot
Max.
Steps
3... | JARVIS-1 |
1019
assess model robustness, called Stability Error
Rate (SER). Neural models, especially with a re-
triever component, have shown to be vulnerable
to model overstability (Jia and Liang, 2017).
Overstability is the inability of a model to dis-
tinguish superfluous information that merely has
lexical similarity with ... | ProoFVer- Natural Logic Theorem Proving for Fact Verification |
Tom Brown, Benjamin Mann, Nick Ryder, Melanie
Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind
Neelakantan, Pranav Shyam, Girish Sastry, Amanda
Askell, et al. 2020. Language models are few-shot
learners. Advances in neural information processing
systems, 33:1877–1901.
Stephanie Chan, Adam Santoro, Andrew Lampinen,
J... | AreEmergentAbilitiesinLarge Language Models just In-Context |
Figure 3: Complexity curves, evaluated using stratified subsamples of the adult dataset. (A): Training time as a function
of sample size. (B): Sampling time as a function of sample size. (C): Training time as a function of dimensionality. (D):
Sampling time as a function of dimensionality.
sities; grow trees too shallow... | Adversarial Random Forests for Density Estimation and Generative Modeling |
These methods enforce a stronger alignment between inputs and outputs. However, they will
bring challenges due to the gap between the original source and augmented information, such as
the semantic gap between an ambiguous utterance and a distinct MR of structured data, and the
format discrepancy between the structured... | SurveyofHallucinationinNatural Language Generation |
Edouard Grave. Few-shot Learning with Retrieval Augmented Language Models. 2022.
[45] Wenxiang Jiao and WenxuanWang Jen-tseHuang XingWang ZhaopengTu. Is chatgpt a good translator? yes with gpt-4 as the engine.
[46] Mandar Joshi, Eunsol Choi, Daniel S Weld, and Luke Zettlemoyer. Triviaqa: A large scale distantly superv... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
▲))) Why is it not ready?
It’s not ready because the eggs are still runny.
▲))) What about now?
It looks ready now. You can take it off the heat and
serve it.
Table 13 | Audio-visual qualitative example showcasing the ability of Gemini models to process
interleaved sequences of text, vision, and audio, as well as r... | gemini_1_report |
(2019), Garfinkel (2020).
5Some—e.g. Adamczewski (2019), Garfinkel (2019), and Ngo (2019), though Ngo notes that his views have
since shifted—suggest that arguments for X-risk from misaligned AI have “shifted” or “drifted” over time. I see
some of this in some places, but for me at least (and for various others I know),... | Is Power-Seeking AI an Existential Risk? |
Figure 10. Limitations. Our method might fail to model moving
thin objects such as moving leash (left). Our method can fail to
render dynamic contents only visible in distant frames (middle).
The rendered static content can be unrealistic or blank if insufficient
source views feature are aggregated for a given pixel (ri... | DynIBaR-NeuralDynamicImage-BasedRendering |
that match the input video.
In order for the model to attend to the order of the sequence of music and
video features, we add positional encodings to both the music and video em-
bedding vectors, before feeding it to the encoder and decoder, respectively.
We use the relative position representation (RPR) introduced... | Video2Music |
speech translation. arXiv:2201.03713, 2022b.
N. P. Jouppi, G. Kurian, S. Li, P. Ma, R. Nagarajan, L. Nai, N. Patil, S. Subramanian, A. Swing,
B. Towles, et al. Tpu v4: An optically reconfigurable supercomputer for machine learning with
hardware support for embeddings. arXiv preprint arXiv:2304.01433, 2023.
N. Kalchbr... | Translatotron3 |
(2023).
(2023).
[75] Jeffrey L Elman. 1993. Learning and development in neural networks: The importance of starting small. Cognition 48, 1 (1993), 71–99.
[76] Shiqing Fan, Yi Rong, Chen Meng, Zongyan Cao, Siyu Wang, Zhen Zheng, Chuan Wu, Guoping Long, Jun Yang, Lixue Xia, et al. 2021. DAPPLE: A Pipelined
Data Parall... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
the following: Before you initiate a set of experi-
ments, you register your hypotheses, your experi-
mental design and how you plan to analyze your
results. Registration is time-stamped on an online
platform with general public access. You then fol-
low your plan as closely as possible and report any
divergences in yo... | A Two-Sided Discussion of Preregistration of NLP Research |
2
2
0
2
b
e
F
0
1
]
L
C
.
s
c
[
3
v
9
3
2
8
0
.
1
0
2
2
:
v
i
X
r
a
LaMDA: Language Models for Dialog Applications
Romal Thoppilan
Daniel De Freitas ∗
Jamie Hall
Noam Shazeer ∗
Apoorv Kulshreshtha
Heng-Tze Cheng
Alicia Jin
Taylor Bos
Leslie Baker
Yu Du
YaGuang Li
Hongrae Lee
Huaixiu Ste... | LaMDA- Language Models for Dialog Applications |
Research proposals have a limit on words or pages so you won’t be able to analyse the whole existing body of literature.
Choose key research papers or public documents and explain clearly how your research will either fill a gap, complete or
follow on from previous research even if it is a relatively new field or if ... | research proposal guidance |
split), it selects this answer, otherwise it reverts to a greedy sample based on maximum likelihood
choice without chain of thought. We refer the reader to appendix for a detailed breakdown of how
this approach compares with only chain-of-thought prompting or only greedy sampling. | gemini_1_report |
After this scenario, the participants were presented with a quasi-randomized set of 67 items. Once the par-
ticipants had responded to all of the questions, their demographic information was collected and the survey
concluded.
3.4 Exploratory Factor Analysis
For the item analysis, we inverted the negatively worded ite... | Society’sAttitudesTowardsHumanAugmentation |
language understanding and generation. arXiv preprint arXiv:2201.07126, 2022.
Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi
Chen, and Wen-tau Yih. Dense passage retrieval for open-domain question answering. In Proceed-
ings of the 2020 Conference on Empirical Methods in Nat... | STANDING ON THE SHOULDERS OF GIANT FROZEN LANGUAGE MODELS |
ground truth distribution. The FID ignores temporal coherency, while the FVD measures how well
the spatio-temporal dynamics of the videos are reconstructed. Results in Table 3 show that per-
frame image based methods slightly outperform our video method (indicated by marginally higher
FID of C-ViViT ), however, they do... | PHENAKI- VARIABLE LENGTH VIDEO GENERATION FROM OPEN DOMAIN TEXTUAL DESCRIPTIONS |
orAI.Humanscreatetoolstosatisfyourownneeds,sothedesignationnaturallysuitshumanpreferenceandconvenience.However,currenttoollearningalgorithmsmaynotbeoptimalorefficientformodels.Thisisbecausemosttools(e.g.,searchengines)arespecificallydesignedforhumanuse,andmodelsprocessinformationinadifferentway.Therefore,itisnecessarytoc... | Tool Learning with Foundation Models |
For simplicity, we will usually refer to state paths as paths when it is clear from context which type of path it is. We
will also usually refer to label paths as plans, since the labels will typically refer to actions (or operators). Also note that
zero-length paths are allowed, that is, s is a path for any state s.... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
5 Experimental Setup
5.1 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), as shown in Table 1.
The datasets are ordered by increasing complexity
and size. E2E only has 1 d... | Prefix-Tuning |
Let’s Verify Step by Step
∗
Hunter Lightman
∗
Vineet Kosaraju
Yura Burda
Harri Edwards
∗
Bowen Baker
Teddy Lee
Jan Leike
John Schulman
Ilya Sutskever
∗
Karl Cobbe
OpenAI
Abstract | Let’s Verify Step by Step |
while text-embedding-ada-002 is better for blocks containing
256 or 512 tokens. Furthermore, the length and complexity
of the user’s input question text, as well as the specific needs
of your application such as semantic search or Q&A, will all
affect the choice of chunking strategy. This might directly
correlate with ... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
Cross-lingual: Here we examine the effect of increasing the prompt length for the case of cross-
lingual zero-shot TTS. As described in 5.2, this setting has a total 36 language transfer directions
for each pair of source and target language. For each target text in a given transfer setting, we
examine how WER / SIM-o6... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
slightly sweet. Some people also describe it as having a slightly numbing or
tingling sensation on the tongue, due to the use of Sichuan peppercorns or other
spices. Overall, the taste of sex in a pan is complex and bold, with a variety of
flavors and textures that make it a popular dish among many people. | Llama2 |
1
Large Language Models (LLMs) have shocked
the natural language processing (NLP) community
with their strong performance and emergent abil-
ities (OpenAI, 2023; Touvron et al., 2023a; Wei
et al., 2022). According to previous studies (Ka-
plan et al., 2020; Hoffmann et al., 2022), LLMs’
achievements depend heavily on s... | DataManagementForLargeLanguageModels-ASurvey |
Polish [99], Selection-Inference
[177], Self-Refine [178], etc.
Plan formulation
Plan reflection
Least-to-Most [98], SayCan [179], Hug-
gingGPT [180], ToT [181], PET [182],
DEPS [183], RAP [184], SwiftSage
[185], LLM+P [125], MRKL [186], etc.
LLM-Planner [101], Inner Monologue
[187], ReAct [91], ChatCoT [188], AI
... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Image to Text (JP) > GPT4v: 人類の「牽引競争」は本当に体力を使いますね！@ Itseems like the human "pulling a car race" is really physically demanding! > LLaVA-1.5: 一台警察車が草地に停まっており、2匹の猫がその警察車を監視しています。@ A police car is parked on the grass, and twocats are keeping a watchful eye on it. > MiniGPT-v2: 猫たちが窓辺に座って、電車がレールを走るのを眺めています。@ Cats sitting o... | Let’sThinkOutsidetheBox |
should be used carefully and deployed only after significant safety tuning is applied. | Llama2 |
step-by-step (Wei et al., 2022c; Press et al., 2022; Khot et al., 2022). Here we keep consistent with these works
and discuss reasoning in the sense of problem decomposition and sub-problem solving.
The vanilla few-shot prompt learning (Brown et al., 2020), whereby models are provided with a prompt
consisting of severa... | Tool Learning with Foundation Models |
all prior iterations, such as those used in RLHF-V1 and RLHF-V2. Although we do not present specific
figures, this adjustment demonstrated considerable enhancements in performance and effectively addressed
the previously noted issues. This mitigation can be seen as analogous to Synnaeve et al. (2019) and Vinyals
et al.... | Llama2 |
Example-Guided Instruction Generation In-
spired by the works of Wang et al. (2022a) and
Taori et al. (2023), we design a prompt, accompa-
nied by a few examples and constraints, to generate
instructions. We include only three random exam-
ples and a limited number of constraints in each
prompt, as shown in Figure 2. I... | LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions |
Mathematics General science Engineering
Reference
Arora et al. [3]
Bubeck et al. [13]
Castro Nascimento and Pimentel [16]
Collins et al. [25]
Dao and Le [29]
Guo et al. [56]
Liu et al. [116]
Pallagani et al. [140]
Sridhara et al. [171]
Valmeekam et al. [183]
Valmeekam et al. [184]
Wei et al. [209]
Wu et al. [213]
Yuan... | ASurveyonEvaluationofLargeLanguageModels |
progress in 3D-aware GANs [6, 22, 48] shows impressive
results in learning 3D geometry and appearance of rigid ob-
jects from 2D image collections. However, since humans
are highly articulated and have more degrees of freedom to
model, such methods struggle to generate realistic humans.
By modeling articulation, recent... | AG3D- Learning to Generate 3D Avatars from 2D Image Collections |
USM
Scotboard bus four3 traversed regu-
larly to Centra stopping at Cabo de
Roga.
The archipelago lines 120 km north
of peninsula. The largest is Kingurch
island with the settlement of Cua
Losas.
Gemini Pro
Scotturb bus 403 travels regularly to
Sintra, stopping at Cabo da Roca.
The archipelago lies 120 km north
of th... | gemini_1_report |
C
a
s
e
s
t
u
d
y
H
o
w
E
a
s
y
w
a
y
U
s
e
s
A
I
-
P
o
w
e
r
e
d
C
h
a
t
t
o
A
u
t
o
m
a
t
e
C
o
m
m
u
n
i
c
a
t
i
o
n
f
o
r
L
u
x
u
r
y
H
o
t
e
l
s
L
e
a
r
n
h
o
w
E
a
s
y
w
a
y
,
a
g
u
e
s
t
r
e
l
a
t
i
o
n
s
h
i
p
m
a
n
a
g
e
m
e
n
t
p
l
a
t
f
o
r
m
,
u
s
e
d
A
I
2
1
L
a
b
s
L
... | Jurassic-X_ Crossing the neuro-symbolic chasm with the MRKL system |
[133] Reiichiro Nakano, Jacob Hilton, Suchir Balaji, Jeff Wu, Long Ouyang, Christina Kim, Christopher Hesse, Shantanu
Jain, Vineet Kosaraju, William Saunders, et al. 2021. WebGPT: Browser-assisted question-answering with human
feedback. arXiv preprint arXiv:2112.09332 (2021).
[134] Feng Nan, Ramesh Nallapati, Zhiguo W... | SurveyofHallucinationinNatural Language Generation |
Decoder-only vs Encoder-only The key similarities of decoder-only and encoder-only architectures is that
decoder-only architectures operate with an input-to-target paradigm or targets-only paradigm if CausalLM
is used over PrefixLM used. For both architectures, the objective is always to predict the next token (LM)
and ... | UL2- Unifying Language Learning Paradigms |
Xi Chen, Xiao Wang, Soravit Changpinyo, A J Piergiovanni, Piotr Padlewski, Daniel Salz, Sebastian
Goodman, Adam Grycner, Basil Mustafa, Lucas Beyer, Alexander Kolesnikov, Joan Puigcerver,
Nan Ding, Keran Rong, Hassan Akbari, Gaurav Mishra, Linting Xue, Ashish Thapliyal, James
Bradbury, Weicheng Kuo, Mojtaba Seyedhossei... | gemini_1_report |
pη(z | x) ∝CLIPv(sz)⊤CLIPv(sx),
(2)
where sz and sx are the visual key of memory entries and
visual query, respectively. Finally, we retrieve the plan of
top-k candidate entries as reference prompt z. | JARVIS-1 |
tasks. Specifically, we use 25-shot ARC-Challenge (Clark
et al., 2018), 10-shot HellaSwag (Zellers et al., 2019), 5-shot
MMLU (Hendrycks et al., 2020), 0-shot TruthfulQA (Lin
et al., 2021) and 5-shot GSM8K (Cobbe et al., 2021). The
results are shown in Table 4. SelfExtend has nearly no
influence on these short-context ... | Self-Extend LLM |
• Relatedly: if the “true” objective function provides slower feedback, agents that pursue
faster-feedback proxies have advantages. For example: in the game Montezuma’s Revenge,
it helps to give an agent a direct incentive analogous to “curiosity” (e.g., it receives reward
for finding sensory data it can’t predict very ... | Is Power-Seeking AI an Existential Risk? |
[9] Yitao Liang and Guy Van den Broeck. Towards compact interpretable models: Shrinking of
learned probabilistic sentential decision diagrams. In IJCAI 2017 Workshop on Explainable
Artificial Intelligence (XAI), August 2017.
[10] YooJung Choi, Meihua Dang, and Guy Van den Broeck. Group fairness by probabilistic
modeli... | Tractable Regularization of Probabilistic Circuits |
5.2 Environment for Agent Society
In the context of simulation, the whole society consists of not only solitary agents but also the
environment where agents inhabit, sense, and act [541]. The environment impacts sensory inputs,
action space, and interactive potential of agents. In turn, agents influence the state of t... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Worldview backfire effects can be understood as a product of directionally
motivated reasoning (for a comprehensive review, see Flynn et al. 2017).
According to theories of motivated reasoning, individuals are motivated to
process information in ways that align with their ultimate goals (Kunda
1990). In particular, indi... | Social_Media_and_Democracy |
Qualitative evaluation. Ultimately, we rely on subjective
tests to evaluate the adherence of generated samples to the
text description. We set up an A-vs-B human rating task, in
which raters are presented with the text description and two
samples of music generated by two different models, or one
model and the referenc... | MusicLM |
from our ordered representation allows one to control the
reconstruction-editability tradeoff and reduces NeTI’s re-
quired storage footprint. | A Neural Space-Time Representation for Text-to-Image Personalization |
(2)
(3)
3.3 PaMIR: Parametric Model-Conditioned Implicit Rep-
resentation
To combine the strengths of parametric body models
and non-parametric implicit field, we introduce Paramet-
ric Model-Conditioned Implicit Representation (PaMIR).
Specifically, in PaMIR, we define the definition of C(p) in
Eqn.(3) as:
C(p) = (S (F... | PaMIR- Parametric Model-Conditioned Implicit Representation for Image-based Human Reconstruction |
In fact, belief in misinformation may actually be more prevalent within this
more educated and engaged group. Recent research finds that individuals who
are more politically active and engaged are more likely to share misinformation
via social media, thereby contributing to the spread of misinformation to other
the publ... | Social_Media_and_Democracy |
BUFF data from 12 views spanning every 30 degrees in yaw
axis using the same method in Sec.6.1. | PaMIR- Parametric Model-Conditioned Implicit Representation for Image-based Human Reconstruction |
L. Ericsson, H. Gouk, and T. M. Hospedales. Why do self-supervised models transfer?
investigating the impact of invariance on downstream tasks, 2021a. 21
L. Ericsson, H. Gouk, and T. M. Hospedales. How well do self-supervised models transfer?
In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Re... | A Cookbook of Self-Supervised Learning |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.