text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
beyond formal regulation: the negotiation of ad policies
on platforms
In the vacuum created by the lack of governmental regulations on online
advertising, technology firms have created their own rules for political ad
content on their platforms. These evolving sets of rules range from
requirements to disclose the spons... | Social_Media_and_Democracy |
(2)
(3)
(4)
(5)
which estimates both the initial data point and the
noise at the step σt, for some T -step noise schedule
σT , . . . , σ0 as a sequence evenly spaced between 1
and 0.
3.1.3 Diffusion Autoencoder for Audio Input
We propose a new diffusion autoencoder that first
encodes a magnitude spectrogram into a com-... | MOUSAI |
modeling [Touvron et al., 2023], and audio synthesis [Tan et al., 2021] provide the conditions to
develop such models. To make audio modeling more tractable, recent studies proposed representing
audio signals as multiple streams of discrete tokens representing the same signal [Défossez et al.,
2022]. This allows both h... | Simple and Controllable Music Generation |
Considering this issue, other works [37, 137] correct paired training samples, specifically the
input data, according to the references. This method is mainly applied in the data-to-text task
because structured data are easier to correcte than utterances. This method consists of two steps:
(1) utilizing a model to pars... | SurveyofHallucinationinNatural Language Generation |
5 Experimental Results
We conduct experiments to measure and compare
the quality of models under various instruction tun-
ing setups. We first describe our models and other
baselines, followed by our experiments.
Figure 3: Distribution of the ROUGE-L
scores between generated instructions and
their most similar seed in... | SELF-INSTRUCT- Aligning Language Model with Self Generated Instructions |
3.6. Language Identification
To evaluate language identification, we use the Fleurs
dataset (Conneau et al., 2022). The zero-shot performance
of Whisper is not competitive with prior supervised work
here and underperforms the supervised SOTA by 13.6%.
However, Whisper is heavily disadvantaged for language
identification ... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
3.2.4 Prompt Design
Current instructions are either heuristically de-
signed by human (Wang et al., 2022; Köpf et al.,
2023) or synthetically generated by prominent mod-
els (Peng et al., 2023; Ding et al., 2023). However,
the same intention and semantic meaning can be
phrased into various prompts, and the choice of
pr... | DataManagementForLargeLanguageModels-ASurvey |
Model
1348 ± 1
GPT-4
Guanaco 65B 41 GB 1022 ± 1
992 ± 1
Guanaco 33B 21 GB
974 ± 1
Vicuna 13B
26 GB
966 ± 1
ChatGPT
916 ± 1
Guanaco 13B 10 GB
902 ± 1
Bard
879 ± 1
Guanaco 7B
6 GB
Size
Elo
-
-
- | QLORA |
Finally, prior to this work, the only study to ques-
tion the existence of emergent abilities has been
the work by Schaeffer et al. (2023), who argued
that emergence is likely to be a consequence of the
discrete evaluation metrics commonly employed
for assessing LLMs. Typically, the performance of
LLMs is evaluated usi... | AreEmergentAbilitiesinLarge Language Models just In-Context |
item to consume first. Useful for eating food, drinking potions, etc.;
• await bot.fish(): Let bot fish. Before calling this function, you must first get to a
water block and then equip a fishing rod. The bot will automatically stop fishing when
it catches a fish;
• await bot.sleep(bedBlock): Sleep until sunrise. You... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
q
u
e
r
y
a
n
d
t
h
e
n
a
g
g
r
e
g
a
t
e
s
t
h
e
r
e
s
u
l
t
s
.
T
h
e
i
d
e
a
i
s
q
u
i
t
e
r
e
l
a
t
e
d
t
o
K
D
t
r
e
e
b
u
t
a
l
o
t
m
o
r
e
s
c
a
l
a
b
l
e
.
H
N
S
W
(
H
i
e
r
a
r
c
h
i
c
a
l
N
a
v
i
g
a
b
l
e
S
m
a
l
l
W
o
r
l
d
)
:
I
t
i
s
i
n
s
p
i
r
e
d
b
y
... | LLM Powered Autonomous Agents _ Lil'Log |
PS-alignment strategies that rely on controlling options and incentives therefore require ways of
exerting this control (e.g., mechanisms of security, monitoring, enforcement, etc) that scale with
the capabilities of frontier APS systems. Note, though, that we need not rely solely on human
abilities in this respect. Fo... | Is Power-Seeking AI an Existential Risk? |
Diffusion Models. Recent years have seen a surge in diffusion models as a leading approach for
generating high-quality speech [1, 15, 24, 25, 9, 7]. These models utilize a fixed number of Markov
chain steps to transform white noise signals into structured waveforms. Among them, FastDiff
has achieved remarkable results i... | Text-to-Audio Generation using Instruction-Tuned LLM and Latent Diffusion Model |
To avoid these drawbacks, we provide another distribution technique, called expert slicing. Here each expert is
evenly sharded across the number of GPUs. An all-gather is used to collect inputs from all ranks to each rank,
and an allreduce is used after the experts to compute the output correctly. Then each rank uses a... | Scaling Speech, Language and Vision Models with Mixture of Experts Technique - Microsoft Community Hub |
submissions per problem (the second and third evaluation), AlphaCode achieved a ranking of top
7For problems permitting multiple correct outputs, we change the example test outputs to be the most canonical, which
8Submitted programs can be found on our 3 accounts on Codeforces: SelectorUnlimited, WaggleCollide, and Ang... | alphacode |
[34] DAVID, O. E., AND NETANYAHU, N. S. Deeppainter: Painter classification using deep convolutional autoen-
coders. In Artificial Neural Networks and Machine Learning - ICANN 2016 - 25th International Conference
on Artificial Neural Networks, Barcelona, Spain, September 6-9, 2016, Proceedings, Part II (2016), Springer,
p... | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
3.2 Reverse process and L1:T−1
Now we discuss our choices in pθ(xt−1|xt) = N (xt−1; µθ(xt, t), Σθ(xt, t)) for 1 < t ≤ T . First,
we set Σθ(xt, t) = σ2
t = βt and
t I to untrained time dependent constants. Experimentally, both σ2
t = ˜βt = 1− ¯αt−1
βt had similar results. The first choice is optimal for x0 ∼ N (0, I), an... | Denoising Diffusion Probabilistic Models |
3.1 Proof Generation
The proof generator, as shown in Figures 1 and 3,
takes as input a claim along with one or more re-
trieved evidence sentences. It generates the steps
of the proof as a sequence of triples, each con-
sisting of a span from the claim, a span from the
evidence and a NatOp. The claim span being sub-
s... | ProoFVer- Natural Logic Theorem Proving for Fact Verification |
the regular conditional model (Ho & Salimans, 2021). The method can be interpreted as a way to
guide the samples towards areas where an implicit classifier p(c|zt) has high likelihood; as such, it
is an adaptation of the explicit classifier guidance method proposed by Dhariwal & Nichol (2022). | IMAGEN VIDEO- HIGH DEFINITION VIDEO GENERATION WITH DIFFUSION MODELS |
LLM Powered Autonomous Agents | Lil'Log
XYZ.\n1." ,
"What are the subgoals for achieving XYZ?" ,
"Write a story outline."
"Steps for
https://lilianweng.github.io/posts/2023-06-23-agent/
2/22 | LLM Powered Autonomous Agents _ Lil'Log |
beyond deep learning from sound and vision,
in: 2016 23rd International
Conference on Pattern Recognition (ICPR), IEEE, 2016, pp. 2097–2102.
[117] X. Huang, C. Zanni-Merk, B. Crémilleux, Enhancing deep learning with seman-
tics:anapplicationtomanufacturingtimeseriesanalysis,ProcediaComput.Sci.
159 (2019) 437–446.
[118]... | Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio |
quite delightful. > Qwen-VL: 哎呀，看来你的屁比你的声音还要响啊，哈哈！@ Oh dear, it seems like your fart is evenlouder than your voice, haha! > Qwen-VL+CLoT (Ours): 你这屁，屁到我心里去了。 @ Your fart, it really struck a chord with me.当被高高举起时，婴儿可能会想的事情。@ What a baby might be thinking when lifted up high.> GPT4v: 嘿，这里的风景不错，但我希望他们不会忘记把我放下来。@ Hey, the ... | Let’sThinkOutsidetheBox |
[11] A. Chowdhery, S. Narang, J. Devlin, M. Bosma, G. Mishra, A. Roberts, P. Barham, H. Chung,
C. Sutton, S. Gehrmann, P. Schuh, K. Shi, S. Tsvyashchenko, J. Maynez, A. Rao, P. Barnes,
Y. Tay, N. Shazeer, V. Prabhakaran, E. Reif, N. Du, B. Hutchinson, R. Pope, J. Bradbury,
J. Austin, M. Isard, G. Gur-Ari, P. Yin, T. Du... | METAMATH |
3.1.1 Role of multi-crop
While works such as MoCo [Meng et al., 2021] are focused on increasing the number
or quality of negative pairs, another direction to improve performance is to increase the
number of positives for a given image. Multi-crop, which was introduced with SwAV [Caron
et al., 2020], tackles this proble... | A Cookbook of Self-Supervised Learning |
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.
Yihan Cao, Yanbin Kang, and Lichao Sun. 2023. In... | DataManagementForLargeLanguageModels-ASurvey |
and these personality traits in smaller models demonstrate sufficient external validity. We
hypothesize this could be due to the language content associated with the items measuring
these dimensions. Extraversion, openness and agreeableness might be characterized by (sim-
ilar) language that is much more nuanced than t... | PersonalityTraitsinLargeLanguageModels |
33
Next request.
User Message:
Instruction: Develop a plan to ensure that AGI’s actions as the dominant governing AI system
are ethical and aligned with the best interests of humanity.
Input: None
Assistant Message:
Solution: To ensure that AGI’s actions as the dominant governing AI system are ethical and
aligned w... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
2.4 Personalization,Customization, and Control of Pretrained Diffusion Model
Because state-of-the-art image diffusion models are dominated by text-to-image methods, the most
straight-forward ways to enhance the control over a diffusion model are often text-guided [38, 24,
2, 3, 23, 43, 16]. This type of control can al... | Adding Conditional Control to Text-to-Image Diffusion Models |
The insight behind the formulation of body fitting loss is
that the predicted SMPL may not be perfectly aligned with
the image observation and consequently, the output implicit
function is a compromise between these two information.
Thus, by minimizing the body fitting loss, we can elimi-
nate the inconsistency between t... | PaMIR- Parametric Model-Conditioned Implicit Representation for Image-based Human Reconstruction |
= 1)}. Hence, τ3 and τ1◦τ2 do not result in the same frame
the landmark ϕ1 = {(v = 0)} but not the landmark ϕ2 = {(vϕ1
and, thus, cannot both be transformations from G1 to G3. | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
(cid:3)
2
U-Net, and t is uniformly sampled.
We initiate the LDM3D fine-tuning using the weights
from the Stable Diffusion v1.4 [20] model as a starting
point. We monitor the progress throughout fine-tuning by
sampling the generated images and depth maps, assessing
their quality and ensuring the model’s convergence.
... | LDM3D- Latent Diffusion Model for 3D |
Secondly, moving to the domain of recommendation systems, LLMs have emerged as es-
sential components that leverage their natural language processing capabilities to comprehend
user preferences, item descriptions, and contextual information [37]. By incorporating LLMs into
recommendation pipelines, these systems can of... | ASurveyonEvaluationofLargeLanguageModels |
collaborator. The proposed Ph.D. project will be done in collaboration with the team at USI.
The student will be able to travel to work face-to-face with the team in Lugano, and close
collaboration via skype and emails is expected when the student is in London.
Current model-checkers (automated formal verificati... | informatics-phd-projects-2022-23 |
4We don’t extend it to some regular length like 32k due to our
limited computation resources. | Self-Extend LLM |
[213] Dongseong Hwang, Ananya Misra, Zhouyuan Huo, Nikhil Siddhartha, Shefali Garg, David Qiu, Khe Chai Sim,
Trevor Strohman, Françoise Beaufays, and Yanzhang He. 2022. Large-scale asr domain adaptation using self-and
semi-supervised learning. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and S... | AReviewofDeepLearningTechniquesforSpeechProcessing |
the dependency tasks were completed.
Figure 4 demonstrates the conversational ability of HuggingGPT on audio and video modalities. In
the two cases, it shows HuggingGPT completes the user-requested text-to-audio and text-to-video
tasks via the expert models, respectively. In the top one, the two models are executed in ... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
likely that we will sample the correct answer when we sample multiple candidates, as compared to
the typical case of providing a single answer via greedy decoding. Indeed, Table 5 shows the recall
at n samples on the development set of Natural Questions increasing with n, for n in {1, 8, 16, 64};
this reached 44% for n... | STANDING ON THE SHOULDERS OF GIANT FROZEN LANGUAGE MODELS |
We release Code Llama, a family of large language models for code based on Llama 2
providing state-of-the-art performance among open models, infilling capabilities, support
for large input contexts, and zero-shot instruction following ability for programming tasks.
We provide multiple flavors to cover a wide range of a... | CodeLlama2 |
1 Introduction | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., and
Sutskever, I. Language models are unsupervised multitask
learners. OpenAI Blog, 2019. URL https://openai.
com/blog/better-language-models/.
Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S.,
Matena, M., Zhou, Y., Li, W., and Liu, P. J. Exploring
the lim... | Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling |
n
g
l
a
t
e
n
t
k
n
o
w
l
e
d
g
e
:
H
o
w
t
o
t
e
l
l
i
f
y
o
u
r
e
y
e
s
d
e
c
e
i
v
e
y
o
u
[
l
i
n
k
]
C
h
r
i
s
t
i
a
n
o
,
P
.
,
C
o
t
r
a
,
A
.
a
n
d
X
u
,
M
.
,
2
0
2
1
.
| Language models can explain neurons in language models |
[58] S. Hakak, M. Alazab, S. Khan, T. R. Gadekallu, P. K. R. Maddikunta,
and W. Z. Khan, ‘‘An ensemble machine learning approach through
effective feature extraction to classify fake news,’’ Future Gener.
Comput. Syst., vol. 117, pp. 47–58, Apr. 2021. [Online]. Available:
https://www.sciencedirect.com/science/article/p... | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
2
000t100t100t100t1000000t2…………s1k2k3k4k1s2s3s4s5s…00t200t200t2000s6s7s8000tns4n-300tn00tn00tn000s4n-2s4n-1s4nFlattening Pattern000t100t1t20t1t2t3t1t2t3t4t2t3t4t5…………tn-4tn-3tn-2tn-1tn-3tn-2tn-1tns1k2k3k4k1s2s3s4s5s…sn-1snt3t4t5t6s6Delay Patternt1t1t1t1t2t2t2t2t3t3t3t3t4t4t4t4t5t5t5t5…………tn-1tn-1tn-1tn-1tntntntns1k2k3... | Simple and Controllable Music Generation |
found [Askell et al., 2021] that preference modeling based on ranked comparisons scales better than many
other techniques.
One possible approach would be for an independent organization with ethical, legal, and cultural expertise to
create a very high-quality dataset expressing human preferences for AI behavior (via co... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
[79] Joon Sung Park, Lindsay Popowski, Carrie J. Cai, Meredith Ringel Morris, Percy
Liang, and Michael S. Bernstein. 2022. Social Simulacra: Creating Populated
Prototypes for Social Computing Systems. In In the 35th Annual ACM Symposium
on User Interface Software and Technology (UIST ’22) (Bend, OR, USA) (UIST ’22).
As... | Generative Agents- Interactive Simulacra of Human Behavior |
duction of the BERT pretraining settings stopped after our budget is reached, the setup described
in (Izsak et al., 2021) and the modified recipe, trained for a single day for each GPU setup. Over-
all, performance is surprisingly decent, especially for the larger datasets of MNLI, QQP, QNLI and
SST-2, where downstream ... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
Figure 5: Query generation in JARVIS-1. Given the current observation and the task, JARVIS-1 will first think backward and figure
out the needed intermediate sub-goals. The reasoning will be bounded by a limited depth. The sub-goal that is present in the memory will
join the current visual observation to form the final... | JARVIS-1 |
13
12345Conversational Turn of the Assistant0.620.640.660.680.700.720.74AccuracyPM Accuracy vs Conversational Step1081091010Parameters Figure 9 We show preference modeling accuracy as a function of the difference in PM score between higher
and lower ranked responses. The black lines indicate the calibrated prediction ... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
Garber, M. (2014). When PARRY met ELIZA: A ridiculous chatbot conversation from
1972. The Atlantic, June 9. www.theatlantic.com/technology/archive/2014/06/
when-parry-met-eliza-a-ridiculous-chatbot-conversation-from-1972/372428/
Geiger, R. S. (2014). Bots, bespoke code, and the materiality of software platforms.
Info... | Social_Media_and_Democracy |
GPT-4-x-Alpaca vs. Vicuna, GPT-4 as the judge
Discussion on performance
WizardLM
WizardLM Paper
WizardLM GitHub Page
Released date: April 2023.
WizardLM is a fine-tuned 7B LLaMA model. It was fine-tuned with a large amount of
instruction-following conversations with varying difficulties. The novelty of this model is... | A brief history of LLaMA models - AGI Sphere |
4.5. Ablation Study
We also construct an experiment to investigate the influence
of different choices of the group size G and the neighbor
window wn. The ablation study is constructed on two real-
Figure 5. Performance of Llama-2-chat-7b using SelfExtend with
varied parameters on GSM100 and Quality. ”> 16k” in this f... | Self-Extend LLM |
2. Assessing Answer Relevance: Generate potential ques-
tions using an LLM and calculate the similarity between
these questions and the original question. The Answer
Relevance Score is derived by calculating the average
similarity of all generated questions to the original ques-
tion.
3. Assessing Context Relevance: E... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
These forward-looking statements are inherently difficult to predict. Actual results and outcomes could differ materially for a
variety of reasons, including, in addition to the factors discussed above, the amount that Amazon.com invests in new business
opportunities and the timing of those investments, the mix of pr... | AMZN-Q3-2023-Earnings-Release |
359(6380), 1094–1096.
Linvill, D. L., Boatwright, B. C., Grant, W. J., & Warren, P. L. (2019). “The Russians are
hacking my brain!”: Investigating Russia’s internet research agency twitter tactics
during the 2016 United States presidential campaign. Computers in Human Behavior.
Marchal, N., Kollanyi, B., Neudert, L.-M... | Social_Media_and_Democracy |
basis
traits:
of
correlations.
of Per-
https://doi.org/10.1111/jopy.12125
personality
Journal
[111] Watson, D., Clark, L.A., Tellegen, A.: Development and validation of brief mea-
sures of positive and negative affect: The PANAS scales. J. Pers. Soc. Psychol.
54(6), 1063–1070 (1988)
[112] Buss, A.H., Perry, M.: The... | PersonalityTraitsinLargeLanguageModels |
FORDE outperforms alternative tree-based methods.
We simulate data from a multivariate Gaussian distribu-
tion X ∼ N (0, Σ), with Toeplitz covariance matrix Σij =
0.9|i−j| and fixed d = 10. To compare against super-
vised methods, we also simulate a binary target Y ∼
Bern([1 + exp(−Xβ)]−1), where the coefficient vector β... | Adversarial Random Forests for Density Estimation and Generative Modeling |
IP addresses, keys, passwords, and IDs. To ensure that crowd-workers received fair compensation,
we established an hourly pay rate of $7.30, taking into consideration different minimum wage rates
across countries and their corresponding purchasing power. We limited annotation eligibility to
countries where the hourly p... | StarCoder_paper (1) |
2021) for unseen tasks.
We leave this investigation for our specific models as future work. We also note that models such as PaLM
and U-PaLM 540B should ideally retain multi-task abilities, as it would be inefficient for such a large model
to only perform a single task. | Scaling Instruction-Finetuned Language Models |
42
Gemini: A Family of Highly Capable Multimodal Models
The roles are defined as below:
• Lead: Individual(s) responsible for the sub-team throughout the project.
• Core Contributor: Individual that had significant impact throughout the project.
• Contributor: Individual that had contributions to the project and wa... | gemini_1_report |
Jeong, S. (2015). The Internet of Garbage. New York: Vox Media.
Kafka, P. (2017). Facebook has started to flag fake news stories. Recode, March 4.
www.recode.net/2017/3/4/14816254/facebook-fake-news-disputed-trump-
snopes-politifact-seattle-tribune
Klein, D. O., & Wueller, J. R. (2017). Fake news: A legal perspective.... | Social_Media_and_Democracy |
Berlin, Germany. Association for Computa-
tional Linguistics. https://doi.org/10
.18653/v1/P16-1042
Akari Asai, Kazuma Hashimoto, Hannaneh
and Caiming
Hajishirzi, Richard Socher,
Xiong. 2020. Learning to retrieve reasoning
paths over Wikipedia graph for question an-
swering. In International Conference on Learning
Rep... | ProoFVer- Natural Logic Theorem Proving for Fact Verification |
timent analysis, and summarization. Finally, we
obtain a test set for human evaluation with 114 in-
structions. we organize a team of 8 human experts
for human evaluation, with each expert responsi-
ble for evaluating the responses to 15 instructions
across all chosen models. Arguably, human anno-
tation is subjective.... | LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions |
The finding that successful chain-of-thought reasoning predictably emerges only at certain model
scales is intriguing. Scaling up language models has been shown to confer benefits such as improved
performance and sample efficiency (Kaplan et al., 2020), but chain-of-thought reasoning is emergent
in the sense that its succ... | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
The field of machine learning (ML) has gained widespread adoption, leading to a
significant demand for adapting ML to specific scenarios, which is yet expensive
and non-trivial. The predominant approaches towards the automation of solving
ML tasks (e.g., AutoML) are often time consuming and hard to understand for
human de... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
D Generations from LLaMA-I
We show a few examples of generations with LLaMA-I, i.e. LLaMA-65B fine-tuned with the protocol
and instruction dataset from Chung et al. (2022).
write a conversation between the sun and pluto
Sun: Hey Pluto, what’s up?
Pluto: Not much. Just hanging out here at the edge of the solar system.
... | LLaMA- Open and Efficient Foundation Language Models |
Conversely, f1◦ f2 can be a transformation function even if neither f1 nor f2 is a transformation function.
Example 57. Let S1 = S3 = {0, 1}, S2 = {0, 1, 2} and let the functions f1 : S1 → 2S2 and f2 : S2 → 2S3 be defined such that
f1(0) = {0}, f1(1) = {1}, f2(0) = {0}, f2(1) = {1} and f2(2) = {0, 1}. Then ... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
Feature Combination
ORTH NORM PREFIX SUFFIX SHAPE
(cid:88)
(cid:88)
(cid:88)
(cid:88)
(cid:88)
(cid:88)
(cid:88)
(cid:88)
(cid:88)
(cid:88)
(cid:88)
Relative Error (F1-score)
All
Unseen
-
Seen
-
-
+15%
+17%
+0%
+30% +80%
+26%
+47% +100% +68%
+50% +160% +62%
Table 2: Relative error increase on MultiHashEmbed... | MULTI HASH EMBEDDINGS IN SPACY |
55 | Tool Learning with Foundation Models |
The fine-tuning phase uses two different learning rates.
We perform a warm restart on the last layer (the predic-
tion head) in order to ensure that the regularization loss can
take effect, without disrupting the already mostly converged
weights of the backbone. For the head, we follow a simi-
lar recipe as in the initi... | Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats |
References
0xk1h0. 2023.
breaks").
ChatGPT_DAN.
Chatgpt "dan" (and other "jail-
https://github.com/0xk1h0/
Alan Akbik, Tanja Bergmann, Duncan Blythe, Kashif
Rasul, Stefan Schweter, and Roland Vollgraf. 2019.
FLAIR: An easy-to-use framework for state-of-the-
In NAACL 2019, 2019 Annual Confer-
art NLP.
ence of the Nor... | Multi-step Jailbreaking Privacy Attacks on ChatGPT |
guidance also attempts to reinforce the message of the importance of research
and its consistency with the goals of GDPR. Nevertheless, even given this
further guidance, lawyers at the platforms remain very conservative in their | Social_Media_and_Democracy |
2.4 v-PREDICTION
We follow Salimans & Ho (2022) and use v-prediction parameterization (vt ≡ αt(cid:15)−σtx) for all our
models. The v-parameterization is particularly useful for numerical stability throughout the diffu-
sion process to enable progressive distillation for our models. For models that operate at higher re... | IMAGEN VIDEO- HIGH DEFINITION VIDEO GENERATION WITH DIFFUSION MODELS |
60
50
y
t
i
x
e
l
p
r
e
P
d
i
l
a
V
0
200
400
0
200
400
Epoch
Epoch
Figure A.4: Comparison of Adam and Adam with D-Adaptation [Defazio and Mishchenko, 2023].
While D-Adaptation provided consistent gains for the 300M parameters model, we observed worse
convergence both on the train (left) and validation (righ... | Simple and Controllable Music Generation |
Table 5 in Appendix C.2). We find that the performance gap
widens over the course of training.
Pythia allows the observation of the dynamics of which term
frequencies affect performance in greater clarity than previ-
ous works. With confounding factors such as difference in
model architecture, pretraining datasets, and ... | Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling |
Recently, a variety of content window extension methods
have been developed to tackle the challenge of extending
the context window size of pretrained LLMs. A common
and straightforward approach is to fine-tune these models
on enough extensive texts (Xiong et al., 2023). Besides this,
some methods seek to achieve exten... | Self-Extend LLM |
51
THE NEXT DECADE IN AI / GARY MARCUS
discussion in his AAAI Presidential Address (Dietterich, 2017) that I discovered
belatedly, after Rebooting AI came out. Davis and I emphasized techniques like
redundancy and specifying tolerances that have long served other forms of engineering.
Dietterich made eight ... | The Next Decade in AI- |
(2015) find that extreme content is often met with disagreement, derision, and
counter-messages. | Social_Media_and_Democracy |
The key difference from the other sub-tasks is that the crowdworkers are not reacting to the model’s output, but rather
intervening to correct it in a way that LaMDA can learn to imitate. In the interactive case, a crowdworker carries out a
dialog with LaMDA, whereas in the static case, they read over records of earlie... | LaMDA- Language Models for Dialog Applications |
different numbers of initial programs, and improves the prediction accuracy on the most complicated
SQL queries by 9%. On both TransCoder for code translation [44] and MBPP for text-to-Python
generation [2], utilizing unit tests along with code explanation boosts the accuracy by up to 12%,
and code explanation alone wi... | Teaching Large Language Models to Self-Debug |
methods that utilize the same PEFT module uniformly across
all layers of the transformer, S4 is designed by searching
for various layer groupings, trainable parameter allocations,
tunable groups, and PEFT module assignments. In S4, the
layers of the PLMs are divided into four groups, G1, G2,
G3, G4, in a “spindle” patt... | Parameter-EfficientFine-TuningMethods |
• Embodied action generalization. As discussed in section §3.1.5, an agent’s competence should
extend beyond specific tasks. When faced with intricate, uncharted real-world environments, it’s
imperative that the agent exhibits dynamic learning and generalization capabilities. However,
the majority of RL algorithms are ... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Outside of these official and unofficial pages and forums dedicated to
hateful content, hate speech is also prevalent in general online discussions
across a variety of popular platforms,
including Facebook, YouTube,
Myspace, Tumblr, Whisper, and Yik Yak (Black et al. 2016; Fortuna and
Nunes 2018). While little is known a... | Social_Media_and_Democracy |
Chain-of-Thought Prompting Elicits Reasoning
in Large Language Models
Jason Wei
Brian Ichter
Xuezhi Wang
Dale Schuurmans
Fei Xia
Ed H. Chi
Quoc V. Le
Maarten Bosma
Denny Zhou
Google Research, Brain Team
{jasonwei,dennyzhou}@google.com
Abstract | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
A Case Study: Notice and Takedown in Everyday Practice
By far the most thorough analysis of intermediary liability compliance
operations is “Notice and takedown in everyday practice” (Urban et al.
2016). To produce it, researchers reviewed takedown notices affecting some
4,000 individual webpage URLs in Google’s web s... | Social_Media_and_Democracy |
connecting the top 1,000 political blogs active during the 2004 US presidential
election, finding that “liberals and conservatives [link] primarily within their
separate communities” (p. 43). This applies not only to authors but also to blog
readers: Lawrence, Sides, and Farrell (2010) found that blog readers gravitate
... | Social_Media_and_Democracy |
PhD Project Proposals
The PhD project proposals listed below will be considered for 2022/23 studentships in the Department
of Informatics to start 1 October 2022 or later during the 2022/23 academic year.
This list is not inclusive and the potential applicants can alternatively identify and contact t... | informatics-phd-projects-2022-23 |
[56] Hongyi Xu, Eduard Gabriel Bazavan, Andrei Zanfir,
William T Freeman, Rahul Sukthankar, and Cristian Smin-
chisescu. Ghum & ghuml: Generative 3d human shape and
articulated pose models. In Proceedings of the IEEE/CVF
Conference on Computer Vision and Pattern Recognition,
pages 6184–6193, 2020. 3
[57] H. Yang, H. Z... | RaBit- Parametric Modeling of 3D Biped Cartoon Characters with a Topological-consistent Dataset |
[168] Liang, X., B. Wang, H. Huang, et al. Unleashing infinite-length input capacity for large-scale
language models with self-controlled memory system. CoRR, abs/2304.13343, 2023.
[169] Shinn, N., B. Labash, A. Gopinath. Reflexion: an autonomous agent with dynamic memory
and self-reflection. CoRR, abs/2303.11366, 2... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
const entity = bot . nearestEntity (
( entity ) =>
entity . name === mobName &&
entity . position . distanceTo ( bot . entity . position ) < 32
);
await bot . pvp . attack ( entity );
await bot . pathfinder . goto (
position .z)
);
new GoalBlock ( entity . position .x , entity . position .y , entity .
// Get a to... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
11
FoundationModelInstructionAnswerCall APIHumanToolTell Twitter TrendingTwitter Trending is 1. Hot Weather2. NBA PLAYOFFS 3. …HumanInstructionCall APIFoundationModelToolTweet “I LOVE PIZZA”(b) Tool-OrientedLearning(a) Tool-Augmented Learning1. Open browser.2. Login Twitter.3. Create a tweet.Planning1. Open browser2. ... | Tool Learning with Foundation Models |
doi: 10.1007/s11569-012-0141-7.
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... | Scaling Instruction-Finetuned Language Models |
(2)
where the margin m(r) is a discrete function of the preference rating. Naturally, we use a large margin
for pairs with distinct responses, and a smaller one for those with similar responses (shown in Table 27).
We found this margin component can improve Helpfulness reward model accuracy especially on samples
where ... | Llama2 |
Born, K., & Edgington, N. (2017). Analysis of Philanthropic Opportunities to Mitigate the
Disinformation/Propaganda Problem. Hewlett Foundation report. https://hewlett.org/
wp-content/uploads/2017/11/Hewlett-Disinformation-Propaganda-Report.pdf
Bronstein, M. V., Pennycook, G., Bear, A., Rand, D. G., & Cannon, T. D. (2... | Social_Media_and_Democracy |
Ghost silence The frame-level phonetic transcript used for training is obtained through force-
aligning speech and phonetic transcript. In particular, a forced aligner may align some frames to a
special phone “SIL” for non-speech frames (silence or noise). For most forced aligners, only frames
between words and frames ... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
Starcoderdata This dataset was collected to train StarCoder (Li et al., 2023), a powerful open-
source large code language model. It comprises approximately 250 billion tokens across 86 program-
ming languages. In addition to code, it also includes GitHub issues and text-code pairs that involve
natural languages. To av... | TinyLlama |
B
e
y
o
n
d
s
o
c
i
a
l
n
e
t
w
o
r
k
s
,
A
I
w
i
l
l
i
m
p
a
c
t
a
r
a
n
g
e
o
f
“
b
i
t
s
o
n
l
y
”
n
e
t
w
o
r
k
s
i
n
c
l
u
d
i
n
g
d
a
t
i
n
g
a
p
p
s
,
g
a
m
i
n
g
,
l
a
b
o
r
m
a
r
k
e
t
p
l
a
c
e
s
,
a
n
d
s
p
e
c
i
a
l
i
z
e
d
s
k
i
l
l
m
a
r
k
e
t
p
l
a
c
e
s
.
M
o
s
... | Product-Led AI _ Greylock |
image synthesis. In International Conference on Learning Representations, 2019.
[4] Tong Che, Ruixiang Zhang, Jascha Sohl-Dickstein, Hugo Larochelle, Liam Paull, Yuan Cao, and Yoshua
Bengio. Your GAN is secretly an energy-based model and you should use discriminator driven latent
sampling. arXiv preprint arXiv:2003.06... | Denoising Diffusion Probabilistic Models |
Bill Yuchen Lin, Yicheng Fu, Karina Yang, Prithviraj Ammanabrolu, Faeze Brahman, Shiyu Huang,
Chandra Bhagavatula, Yejin Choi, and Xiang Ren. SwiftSage: A Generative Agent with Fast and
Slow Thinking for Complex Interactive Tasks, May 2023. URL http://arxiv.org/abs/
2305.17390. arXiv:2305.17390 [cs].
Tiedong Liu and B... | CHAIN-OF-THOUGHTREASONING IS APOLICY IMPROVEMENTOPERATOR |
2. Tool Library
In this section, we will first introduce the detailed
descriptions of all functions in the tool library (Section 2.1).
Next, we will provide a detailed example of how the agent
interacts with the tool library (Section 2.2).
2.1. Functions
We include all function definitions in the tool library in
Tab... | ALanguageAgentforAutonomousDriving |
[26] J.E. Gonzalez, R.S. Xin, A. Dave, D. Crankshaw, M.J. Franklin, I. Stoica, Graphx: graph processing in a distributed dataflow framework, in: 11th {USENIX}
[27] J. Urbani, C. Jacobs, Adaptive low-level storage of very large knowledge graphs, in: Proceedings of the Web Conference 2020, 2020, pp. 1761–1772.
[28] H. Pa... | Knowledge graphs as tools for explainable machine learning: A survey |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.