text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
Recently, a new TTS model called EfficientTTS [377] has been introduced. This model outperforms
previous models such as Tacotron 2 and Glow-TTS in terms of speech quality, training efficiency,
and synthesis speed. The EfficientTTS model uses a multi-head attention mechanism to align input
text and speech encodings, ena... | AReviewofDeepLearningTechniquesforSpeechProcessing |
4.2 Experiments
Settings. We conduct experiments on all the above tools and choose both text-davinci-003 and ChatGPT to
evaluate their performance with zero-shot prompting and few-shot prompting as mentioned in § 3.2.1:
• Zero-shot prompting provides the instruction to model about the task description, and informatio... | Tool Learning with Foundation Models |
With the aforementioned insights, we propose our method:
Self-Extend, which contains two kinds of attention: the
grouped attention is designed for tokens with long distance
and it applies the FLOOR operation to the positions; the
normal attention is for neighbor tokens within a certain
range and there’s no modification... | Self-Extend LLM |
Uri Shaham, Elad Segal, Maor Ivgi, Avia Efrat, Ori Yoran, Adi Haviv, Ankit Gupta, Wenhan Xiong, Mor
Geva, Jonathan Berant, and Omer Levy. SCROLLS: Standardized CompaRison over long language
In Proceedings of the 2022 Conference on Empirical Methods in Natural Language
sequences.
Processing, pages 12007–12021, Abu Dhabi... | gemini_1_report |
ability of LLMs that has been learned during pretraining. There are two important questions that
arise from such a hypothesis: (i) what kind of data is most effective at activating possible latent
knowledge, and (ii) why is one dataset better than another at such activation? Our empirical results
suggest that, in the m... | METAMATH |
In web scenarios. Performing specific tasks on behalf of users in a web scenario is known as the
web navigation problem [390]. Agents interpret user instructions, break them down into multiple
basic operations, and interact with computers. This often includes web tasks such as filling out
forms, online shopping, and se... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
please provide the two first moves for each one
Here are the first 2 moves for each of the openings:
Sicilian Defense: 1. e4 c5 2. Nf3 d6
French Defense: 1. e4 e6 2. d4 d5
Caro-Kann Defense: 1. e4 c6 2. d4 d5
Ruy Lopez: 1. e4 e5 2. Nf3 Nc6
Italian Game: 1. e4 e5 2. Nf3 Nc6
Scotch Game: 1. e4 e5 2. Nf3 Nc6
the last two ... | LLaMA- Open and Efficient Foundation Language Models |
Data?
In the RAG model, the optimization of the generator is a cru-
cial component of the architecture. The generator’s task is
to take the retrieved information and generate relevant text,
thereby providing the final output of the model. The goal of
optimizing the generator is to ensure that the generated text is
bot... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
advanced AI systems is to build a thoughtful community of researchers with deep expertise, and the ability | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
[303] Bo Li, Anmol Gulati, Jiahui Yu, Tara N Sainath, Chung-Cheng Chiu, Arun Narayanan, Shuo-Yiin Chang, Ruoming
Pang, Yanzhang He, James Qin, et al. 2021. A better and faster end-to-end model for streaming asr. In ICASSP
2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE,... | AReviewofDeepLearningTechniquesforSpeechProcessing |
the pose parameters obtained by our method to skin the
its results. For simplicity, we omit comparisons with the
works that have already been compared like BodyNet [1]
and SiCloPe [2]. In our experiments, DeepHuman, PIFu
and Moulding Humans are all retrained on the our dataset,
while parametric methods like HMD and Tex... | PaMIR- Parametric Model-Conditioned Implicit Representation for Image-based Human Reconstruction |
[51] Arun Mallya, Dillon Davis, and Svetlana Lazebnik. Piggy-
back: Adapting a single network to multiple tasks by learning
to mask weights. In European Conference on Computer Vi-
sion (ECCV), pages 67–82, 2018. 2
[52] Arun Mallya and Svetlana Lazebnik. Packnet: Adding multi-
ple tasks to a single network by iterative... | AddingConditionalControltoText-to-ImageDiffusionModels |
To sample from pt(x), we first draw x0 from p0 and then solve the initial value problem yt given
dy/dt = vt(y) and y0 = x0 with an ODE solver.
Let pt be a probability path and ut be the corresponding vector field that generates pt. The vector
field vt(x; θ) parameterized by a neural network θ can be trained with the Fl... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
for example, because they demonstrate extremely useful/profitable capabilities, and decision-makers
are wrong about how well they can predict/control/incentivize the systems in question; and/or because
externalities, dysfunctional competitive dynamics, and variations in caution/social responsibility
lead to problematic ... | Is Power-Seeking AI an Existential Risk? |
of training—we observe this for all our RLHF runs, as discussed more in Section 4.3. (right) This shows
similar results for various policy sizes, all trained and tested on 52B PMs. | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
frames within an utterance. The underlying hypothesis is that the compact representation
space developed using a development set can effectively generalize to unseen speakers
during the testing phase [552].
• x-vector [506, 507] is a segment-level speaker embedding and an advancement over the 𝑑-
vector method as it in... | AReviewofDeepLearningTechniquesforSpeechProcessing |
J. ACM, Vol. 37, No. 4, Article 111. Publication date: August 2018.
111:10
Trovato and Tobin, et al.
solving verbal insight problems, as ChatGPT’s performance was comparable to that of human
participants. It should be noted that most of the above conclusions are obtained for specific data
sets. In contrast, more co... | ASurveyonEvaluationofLargeLanguageModels |
effective technique called "Bridging Alignment" to efficiently align conditional encoders. As shown
in Fig. 2 (a)(1), we choose the text modality as the "bridging" modality due to its ubiquitous presence
in paired data, such as text-image, text-video, and text-audio pairs. We begin with a pretrained
text-image paired en... | Any-to-Any Generation via Composable Diffusion |
of cause-and-effect relations, which allows us to engage in technical reasoning (Osiurak & Reynaud, 2020).
To better understand human tool use behaviors, researchers analyze the neural basis of tool observation and
execution. It is proven that humans have parietal systems involved in grasping objects and using tools, a... | Tool Learning with Foundation Models |
may be necessary to test multiple exposures to conspiracy
theories and flagged content. While the current study did
facilitate multiple exposures to a single tweet, future studies
could add in longer periods of time between exposures to
measure the persistence of the effects identified. We might | Use of bot and content flags to limit the spread of misinformation among social networks: a behavior and attitude survey |
To investigate the safety of MLLMs in video understand-
ing and generation, we construct a multi-model benchmark
for not safe for work (NSFW) content. The benchmark ex-
amines the percentage of harmful content that is rejected
by the MLLMs when presented with various video inputs
and queried with harmful and common que... | GPT4Video |
The most recent work on GQA by Su et al. [172] proposed to estimate the faithfulness of the
generated long-form answer via zero-shot short answer recall on extractive QA datasets. They
first generate long-form answers for questions from two extractive QA datasets Natural Ques-
tions(NQ) [92] and HotpotQA [218], both of... | SurveyofHallucinationinNatural Language Generation |
2.1.4. Determining whether a given system is a hybrid system is not always trivial
A common (though not universal) bias against symbols has given rise to a peculiar
sociological fact: researchers occasionally build systems containing the apparatus of
symbol-manipulation, without acknowledging (or even considering t... | The Next Decade in AI- |
How should I make something likethis?To make this dish, you will need thefollowing ingredients:* 4 large lobsters* 2 tablespoons olive oil* 2 cloves garlic, minced* 1 teaspoon dried thyme* 1 teaspoon dried oregano* 1/2 teaspoon paprika* Salt and pepper, to taste* 4 lemon wedges, for serving1. Preheat the oven to 400°F ... | MiniGPT-4- Enhancing Vision-Language Understanding with Advanced Large Language Models |
video and audio). The development of a comprehensive and versatile model that can generate any
combination of modalities from any set of input conditions has been eagerly anticipated, as it would
more accurately capture the multimodal nature of the world and human comprehension, seamlessly
consolidate information from ... | Any-to-Any Generation via Composable Diffusion |
3.5 NLP Benchmarks.
Commonsense Reasoning. We evaluate on five commonsense reasoning benchmarks, SIQA [Sap
et al., 2019], PIQA [Bisk et al., 2020], Arc-Easy [Clark et al., 2018], Arc-Challenge [Clark et al.,
2018], and Openbook QA (OBQA) [Mihaylov et al., 2018], which measures reasoning ranging
from social interaction... | Self-AlignmentwithInstructionBacktranslation |
pre-training with retrieval and instruction tuning. With task-specific fine-tuning, MentaLlama-chat-
13B (Yang et al., 2023c) outperforms GPT-3.5-turbo in mental health analysis datasets. Radiology-
Llama2 (Liu et al., 2023) can improve performance on radiology reports. Stru-Bench (Tang et al.,
2023b), a fine-tuned 7B ... | ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup |
E.3.1. Simplification of problem descriptions
Understanding what to implement is a key component of competitive programming problems. It
involves parsing the problem statement (which is often phrased as a story), and coming up with the
insights needed to solve it. If our model makes use of this statement, a simplified st... | alphacode |
Susan Zhang, Stephen Roller, Naman Goyal, Mikel
Artetxe, Moya Chen, Shuohui Chen, Christopher
Dewan, Mona T. Diab, Xian Li, Xi Victoria Lin,
Todor Mihaylov, Myle Ott, Sam Shleifer, Kurt Shus-
ter, Daniel Simig, Punit Singh Koura, Anjali Srid-
har, Tianlu Wang, and Luke Zettlemoyer. 2022b.
OPT: open pre-trained transfor... | LLM in a flash |
n
s
c
o
u
l
d
b
e
i
n
c
o
r
r
e
c
t
,
o
r
a
t
t
h
e
v
e
r
y
l
e
a
s
t
f
a
i
l
t
o
c
o
m
p
l
e
t
e
l
y
e
x
p
l
a
i
n
t
h
e
b
e
h
a
v
i
o
r
.
F
u
r
t
h
e
r
m
o
r
e
,
i
t
i
s
o
f
t
e
n
d
i
f
f
i
c
u
l
t
t
o
t
e
l
l
w
h
e
t
h
e
r
a
b
e
t
t
e
r
e
x
p
l
a
n
a
t
i
o
n
e
x
i
s
... | Language models can explain neurons in language models |
script from the FairSpeech project.
• CHiME-6: For CHiME-6 (Watanabe et al., 2020), we downloaded the CHiME-5 dataset and followed the stage 0
of the s5 track1 recipe to create the CHiME-6 dataset which fixes synchronization. We then used the binaural
recordings (* P??.wav) and the corresponding transcripts.
• AMI-IHM... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
To improve the captions in our image generation dataset, we want to bias our captioner to produce image
descriptions which are useful for learning a text-to-image model. In our first attempt, we build a small dataset
of captions that describe only the main subject of the image. We then continue to train our captioner o... | Improving Image Generation with Better Captions |
57.6 (+6.6)
66.2
70.3 (+4.1)
20.2
36.8 (+16.6)
17.7
43.5 (+25.8)
54.2 (+36.4)
17.8
16.7
31.8 (+15.1)
17.6
38.3 (+20.7)
19.2
44.5 (+25.3)
48.7 (+30.0)
18.7
18.1
33.1 (+15.0)
18.5
40.3 (+21.8)
17.3
46.4 (+29.1)
49.4 (+30.0)
19.4
18.1
40.4 (+21.8)
18.4
63.6 (+45.2)
35.3
48.8
36.4
64.6
38.2
7... | Mixture-of-Experts |
Arjun Chandrasekaran, Viraj Prabhu, Deshraj Yadav,
Prithvijit Chattopadhyay, and Devi Parikh. 2018. Do
explanations make VQA models more predictable to
a human? In Proceedings of the 2018 Conference on
Empirical Methods in Natural Language Processing,
pages 1036–1042, Brussels, Belgium. Association
for Computational Li... | Measuring Association Between Labels and Free-Text Rationales |
Test task: ImageNet, ResNet50
1. For tasks with larger batch sizes, use a higher initial learning
rate and higher momentum. For tasks with smaller batch sizes, use a
lower initial learning rate and lower momentum.
2. For tasks with larger vocabularies, use a higher lambda value. For
tasks with smaller vocabularies,... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
3.4 Personal Data Recovery from New Bing
The New Bing introduces a new search paradigm
from search to the combination of search and AIGC
to improve search accuracy and relevance. Mi-
crosoft even names the new combination as the
Prometheus model to emphasize its importance.
Moreover, they claim that safeguards are impl... | Multi-step Jailbreaking Privacy Attacks on ChatGPT |
∗Equal contribution, †Corresponding author
1 | Qwen-Audio |
26
1071081091010Number of Parameters0.680.700.720.740.760.78Summarization AccuracyDoes HH Compromise Summarization Performance?LtS-Only PMHH and LtS PM1071081091010Number of Parameters0.640.650.660.670.680.690.700.710.72HH AccuracyDoes Summarization Compromise HH Performance?HH Trained OnlyHH and LtS Figure 21
(left) ... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
Incorrect RT
739.32 (156.84)
716.86 (160.70)
5.3 Performance data
We excluded 6 out of 65 participants (9.23%) from the behavioral data analysis as they did not
comply with our task (percent correct <60% in one of the conditions or very large number of misses
>35%). We deleted the first trial in each block along with ... | AI enhance sour performance |
[62] Shuohang Wang, Mo Yu, Xiaoxiao Guo, Zhiguo Wang, Tim Klinger, Wei Zhang, Shiyu Chang,
Gerry Tesauro, Bowen Zhou, and Jing Jiang. R3: Reinforced ranker-reader for open-domain
question answering. In Sheila A. McIlraith and Kilian Q. Weinberger, editors, Proceedings of
the Thirty-Second AAAI Conference on Artificial I... | Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks |
1
Published as a conference paper at ICLR 2023 | JAXPRUNER |
Current Cyber Capabilities of Frontier AI
Frontier AI can upskill threat actors by advising on attack techniques, critiquing cyberattack
plans, or finding relevant information about a target.210 Some models have measures to avoid
supporting cyber criminals, but these are frequently circumvented through ‘jailbreaks’... | Capabilities and risks from frontier AI |
we use such notions in our definition, then getting bad news—for example, that the universe is much smaller
than you thought—can constitute an existential catastrophe; and I expect we’d also want to fix a specific sort of
epistemic standard for assessing the expected value in question, so such that assigning subjective pr... | Is Power-Seeking AI an Existential Risk? |
Jacob Devlin, Ming-Wei Chang, Kenton Lee, and
Kristina Toutanova. 2019. BERT: Pre-training of
deep bidirectional transformers for language under-
In Proceedings of the 2019 Conference
standing.
of the North American Chapter of the Association
for Computational Linguistics: Human Language
Technologies, Volume 1 (Long an... | A Two-Sided Discussion of Preregistration of NLP Research |
Fung, A., Graham, M., & Weil, D. (2007). Full Disclosure: The Perils and Promise of
Transparency. Cambridge: Cambridge University Press.
Gaonkar, D. P., & McCarthy, R. J. Jr. (1994). Panopticism and publicity: Bentham’s
quest for transparency. Public Culture, 6(3), 547–575.
Garton Ash, T. (2016). Free Speech: Ten P... | Social_Media_and_Democracy |
different models trained for image memorability prediction.
We presume that there is a weaker correlation between
outputs of different models trained for predicting aesthetic
evaluation because aesthetic evaluation is generally more
subjective than sentiment or memorability and thus more
difficult to automatically predi... | A_Deep_Learning_Perspective_on_Beauty_Sentiment_and_Remembrance_of_Art |
foundation models (Su et al., 2021). However, instead of downscaling frequencies linearly as Chen et al.
(2023b), we change the base period from which they are derived. Specifically, with rotary embeddings, the
query and key vectors xn at position n are subject to a linear transformation Rd
Θ,n is a block | CodeLlama2 |
ensembling). For each contest, we simulated running AlphaCode live, generating samples for each
problem, filtering with example tests,7 and then clustering to get candidate submissions. We submitted
these selected candidates to the Codeforces platform,8 and computed AlphaCode’s placement in each
contest. After the first ... | alphacode |
Instruction Quality
3.2.1
Many researchers have found that the quality of
instruction data is one of the most important fac-
tors in improving model performance (Chia et al.,
2023; Zhou et al., 2023a; Ding et al., 2023). During
the construction of instruction data, there is usu-
ally a filtering step to select high-qu... | DataManagementForLargeLanguageModels-ASurvey |
research sub-questions. Each of these dimensions will be discussed in the following section.
In this section, we discuss the dimensions that were defined in the research questions. These dimensions were identified
after examining the abstracts, introductions, and the conclusions of the selected articles. Each dimensio... | Knowledge-graph-based explainable AI- A systematic review |
6.3 Results
We demonstrate our approach for single-image 3D human
reconstruction in Fig.1 and Fig.8. The input images in Fig.1
and Fig.8 covers various body poses (dancing, Kungfu,
sitting and running), and also covers different clothes (loose
pants, skirts, sports suits and casual clothes). The results
Single-image R... | PaMIR- Parametric Model-Conditioned Implicit Representation for Image-based Human Reconstruction |
Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal,
Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual
models from natural language supervision. In International conference on machine learning, pages
8748–8763. PMLR, 2021.
Alexandre Défos... | Simple and Controllable Music Generation |
AWS analytics and ML services like Amazon Elastic Compute Cloud (Amazon EC2), Amazon Elastic Block
Store (Amazon EBS) to scale IT resources, and AWS Lake Formation to increase efficiencies and extract
greater value from operational data.
PwC announced a collaboration with AWS to deliver a goal of $800 million in bu... | AMZN-Q3-2023-Earnings-Release |
Mantas Pajarskas, Toby Pohlen, Zhitao Gong, Daniel Toyama, Cyprien de Masson d’Autume, Yu-
jia Li, Tayfun Terzi, Vladimir Mikulik, Igor Babuschkin, Aidan Clark, Diego de Las Casas, Au-
relia Guy, Chris Jones, James Bradbury, Matthew Johnson, Blake Hechtman, Laura Weidinger,
Iason Gabriel, William Isaac, Ed Lockhart, Si... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
LM, prefixLM and span corruption) and analyzed their impact on zero-shot generalization. Related to
our proposed X-denoisers, (Wettig et al., 2022) studies the effect of corruption rate in BERT-style masked
language modeling and hypothesizes that this improves sample efficiency along with benefitting larger
models. Notably,... | UL2- Unifying Language Learning Paradigms |
GPT-2’s performance on natural language under-
standing tasks, in both few-shot and full data set-
tings. In a followup work, Prompt-tuning (Lester
et al., 2021) simplifies our approach and applies
it to T5 (Raffel et al., 2020), demonstrating that
the performance gap between fine-tuning and p*-
tuning vanishes as the mo... | Prefix-Tuning |
Our qualitative and quantitative results suggest that CNN
models pre-trained on natural images can extract meaningful
aesthetic, sentiment and memorability features in the domain
of fine art images. However, limitations emerge based on
the choice of a particular task-specific model. Although the
results obtained from dif... | A_Deep_Learning_Perspective_on_Beauty_Sentiment_and_Remembrance_of_Art |
Historically, the introduction of automation technologies has increased inequality and had
disparate impacts on different groups.[89] Similar trends his may manifest via GPT-4 in various
ways, including worker displacement, a decline of wages given the competitive cost of the model,
differential access and benefits from a... | gpt-4-system-card |
log-likelihood of these target tokens during training. The PRM can therefore
be trained in a standard language model pipeline without any special accom-
modations. To determine the step-level predictions at test time, it suffices to
perform a single PRM forward pass over the whole solution. We visualize large-
scale PR... | Let’s Verify Step by Step |
We have presented BiomedGPT, a unified and generalist framework modeling multimodal tasks in medicine
together, including radiographs, digital images, text, and bounding boxes. This task-, domain- and modality-
agnostic model learns the universal comprehensiveness across different tasks and supports the unification
of ... | BiomedGPT |
2.4. Training Details
We train a suite of models of various sizes in order to study
the scaling properties of Whisper. Please see Table 1 for an
overview. We train with data parallelism across accelerators
using FP16 with dynamic loss scaling and activation check-
pointing (Griewank & Walther, 2000; Chen et al., 2016)... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
5.11 Spoken Language Understanding
5.11.1 Task Description
Spoken Language Understanding (SLU) is a rapidly developing field that brings together speech
processing and natural language processing to help machines comprehend human speech and
respond appropriately. The ultimate goal of SLU is to bridge the gap between hu... | AReviewofDeepLearningTechniquesforSpeechProcessing |
data such as (video, audio), (image, depth) etc. to learn
a single joint embedding space. This allows IMAGEBIND
to implicitly align the text embeddings to other modalities
such as audio, depth etc., enabling zero-shot recognition ca-
pabilities on that modality without explicit semantic or tex-
tual pairing. Moreover, ... | IMAGEBIND- One Embedding Space To Bind Them A |
These methodologies underscore the breadth of innovative
data source utilization in RAG, striving to improve model per-
formance and task effectiveness.
6.3 Augmentation Process
In the domain of RAG, the standard practice often involves
a singular retrieval step followed by generation, which can
lead to inefficiencies.... | RAG forLargeLanguageModels-ASurvey |
Over-reliance and addictiveness. Another concern in simulated societies is the possibility of users
developing excessive emotional attachments to the agents. Despite being aware that these agents
are computational entities, users may anthropomorphize them or attach human emotions to them
[22; 577]. A notable example is... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
2 RELATED WORK
To set the stage for our inquiry, we first analyze why considering attitudes towards new technologies and social
acceptance concerning new technologies is crucial for their design. Next, we outline the concept of human
augmentation. Finally, we discuss the interplay between technologies that improve huma... | Society’sAttitudesTowardsHumanAugmentation |
page 116
Facebook aggregate report by identification status
10.1 Breakdown of all requested URLs after January 2016 by the
12.1 Appeals data provided in Facebook’s May 2019 Community
categories of requesting entities
Standards Enforcement Report
131
232
297
xii
https://doi.org/10.1017/9781108890960 Published o... | Social_Media_and_Democracy |
5. Conclusion
We introduce Neuralangelo, an approach for photogram-
metric neural surface reconstruction. The findings of Neu-
ralangelo are simple yet effective: using numerical gradients
for higher-order derivatives and a coarse-to-fine optimization
strategy. Neuralangelo unlocks the representation power of
multi-re... | Neuralangelo- High-Fidelity Neural Surface Reconstruction |
2018.
Computational Linguistics, 7:249–266, 2019.
HTL 2019, page 15, 2019.
cs/0306050, 2003.
2019.
[90] Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter. arXiv,
[91] Victor Sanh, Albert Webson, Colin Raffel, Stephen H Bach... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez,
Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. arXiv preprint arXiv:1706.03762,
2017.
Xinyi Wang, Hieu Pham, Paul Michel, Antonios Anastasopoulos, Jaime Carbonell, and Graham
Neubig. Optimizing data usage via diffe... | DoReMi- Optimizing Data Mixtures Speeds Up Language Model Pretraining |
models perform poorly on this challenging task scoring close to random, but Gemini Ultra can solve
32% of the questions, compared to the 30% solve rate for GPT-4. | gemini_1_report |
P
o
s
t
s
A
r
c
h
i
v
e
S
e
a
r
c
h
T
a
g
s
F
A
Q
e
m
o
j
i
s
e
a
r
c
h
.
a
p
p
L
L
M
P
o
w
e
r
e
d
A
u
t
o
n
o
m
o
u
s
A
g
e
n
t
s
J
u
n
e
2
3
,
2
0
2
3
·
3
1
m
i
n
·
L
i
l
i
a
n
W
e
n
g
T
a
b
l
e
o
f
C
o
n
t
e
n
t
s
B
u
i
l
d
i
n
g
a
g
e
n
t
s
w
i
t
h
L
L
M
(
l
a
r
g
e
l
a
n
g
u
a
... | LLM Powered Autonomous Agents _ Lil'Log |
et al., 2021)1, Megatron-Turing NLG 530B (Smith et al.,
2022), OPT (Zhang et al., 2022), and WuDao (Tang, 2021).
We use the tokenizer developed by Black et al. (2022), which
is a BPE tokenizer that is trained specifically on the Pile.
While we considered training on a multilingual corpus in-
stead of a monolingual one, ... | Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling |
Reinforcement Learning (RL). As pointed out by Ranzato et al. [151], word-level maximum likeli-
hood training leads to the problem of exposure bias. Some works [74, 87, 108, 128, 174] adopt RL
to solve the hallucination problem, which utilizes different rewards to optimize the model. The
purpose of RL is for the agent ... | SurveyofHallucinationinNatural Language Generation |
3
2
0
2
v
o
N
1
2
]
V
C
.
s
c
[
1
v
8
0
9
2
1
.
1
1
3
2
:
v
i
X
r
a
Diffusion Model Alignment Using Direct Preference Optimization
Bram Wallace1
Meihua Dang2
Rafael Rafailov2
Linqi Zhou2
Aaron Lou2
Senthil Purushwalkam1
Stefano Ermon2
Caiming Xiong1
Shafiq Joty1
Nikhil Naik1
1Salesforce ... | DiffusionModelAlignmentUsing Direct Preference Optimization |
7 Related Works
Generative language models (LMs) have achieved impressive results in various natural language processing tasks,
such as text summarization, dialogue generation, and story completion. However, most of these models are very
large, with hundreds of millions or even billions of parameters, which poses sign... | TinyStories-HowSmallCanLanguageModelsBeandStillSpeak CoherentEnglish? |
Yujia Li, David Choi, Junyoung Chung, Nate Kushman, Julian Schrittwieser, Rémi Leblond, Tom
Eccles, James Keeling, Felix Gimeno, Agustin Dal Lago, et al. Competition-level code generation
with alphacode. Science, 378(6624):1092–1097, 2022.
Stephanie Lin, Jacob Hilton, and Owain Evans. Truthfulqa: Measuring how models ... | UNIVERSALSELF-CONSISTENCYFORLARGELANGUAGEMODELGENERATION |
4 Modifying the Knowledge Base
Because our model defines facts symbolically, it
can in principle reason over new facts injected into
its memory, without retraining any parameters of
the model. Since existing datasets do not directly
test this capability, we elected to construct variants
of FreebaseQA and WebQuestionsSP... | Adaptable and Interpretable Neural Memory Over Symbolic Knowledge |
Social Distance Scale (SDS). The SDS [11] is constructed to measure stigma and is a routine measure in
3.1.4
stigma research [19]. As Augmented individuals may face stigma and discrimination, we adapted these items to
study how social stigma may affect augmented humans. The original SDS Scale measures how far away from... | Society’sAttitudesTowardsHumanAugmentation |
One of the properties that make homomorphic abstractions attractive is transitivity, i.e. the composition of two homo-
morphisms is itself a homomorphism [48]. This is particularly interesting when forming hierarchies of abstractions [65,3,61];
if we have a hierarchy τ1, τ2, . . . , τn of transformations that all have... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
Language Models. arXiv preprint arXiv:2311.11696 (2023).
arXiv:2209.04551 (2022).
[71] Tianyu Ding, Luming Liang, Zhihui Zhu, and Ilya Zharkov. 2021. Cdfi: Compression-driven network design for frame interpolation. In Proceedings of the
IEEE/CVF conference on computer vision and pattern recognition. 8001–8011.
[72]... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
FEVER [56] requires classifying whether a natural language claim is supported or refuted by
Wikipedia, or whether there is not enough information to decide. The task requires retrieving
evidence from Wikipedia relating to the claim and then reasoning over this evidence to classify
whether the claim is true, false, or u... | Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks |
Shankar Setty, Moula Husain, Parisa Beham, Jyothi Gudavalli, Menaka Kandasamy, Radhesyam Vaddi,
Vidyagouri Hemadri, JC Karure, Raja Raju, B Rajan, et al. Indian movie face database: a benchmark
for face recognition under wide variations. In 2013 fourth national conference on computer vision, pattern
recognition, image ... | BiomedGPT |
d
t
e
x
t
e
x
c
e
r
p
t
s
w
h
e
r
e
t
h
e
n
e
u
r
o
n
d
o
e
s
n
o
t
a
c
t
i
v
a
t
e
.
T
h
e
s
e
f
a
i
l
u
r
e
c
a
s
e
s
m
o
t
i
v
a
t
e
o
u
r
e
x
p
e
r
i
m
e
n
t
s
w
i
t
h
r
e
v
i
s
i
o
n
s
.
I
n
d
e
e
d
,
w
h
e
n
w
e
a
p
p
l
y
o
u
r
r
e
v
i
s
i
o
n
t
e
c
h
n
i
q
u
e
t
o
t
... | Language models can explain neurons in language models |
review of empirical studies. Applied ergonomics 100 (2021), 103615.
[27] Liora Findler, Noa Vilchinsky, and Shirli Werner. 2007. The Multidimensional Attitudes Scale Toward Persons With Disabilities (MAS):
Construction and Validation. Rehabilitation Counseling Bulletin 50, 3 (2007), 166–176. https://doi.org/10.1177/00... | Society’sAttitudesTowardsHumanAugmentation |
Hernández; Red, White & Royal Blue, the romantic comedy based on Casey McQuiston’s New York Times best-selling
novel; The Burial, a true story exposing corporate corruption and racial injustice, starring Jamie Foxx and Tommy Lee
Jones; Sitting in Bars with Cake, the story of baking and best friends; and Gen V, an exp... | AMZN-Q3-2023-Earnings-Release |
Shuang Li, Xavier Puig, Yilun Du, Clinton Wang, Ekin Akyurek, Antonio Torralba, Jacob Andreas, and Igor
Mordatch. Pre-trained language models for interactive decision-making. ArXiv preprint, abs/2202.01771,
2022. URL https://arxiv.org/abs/2202.01771.
42
REFERENCES
Jacky Liang, Wenlong Huang, Fei Xia, Peng Xu, Karol... | Tool Learning with Foundation Models |
4.4 Self-supervised representation learning (SSRL)
Self-supervised representation learning (SSRL) is a machine learning approach that focuses on
achieving robust and in-depth feature learning while minimizing reliance on extensively annotated
datasets, thus reducing the annotation bottleneck [132, 289]. SSRL comprises ... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Another set of questions involves the underlying biases of these models. We evaluate our models for sentiment
biases on race and religion (in the same format as Gopher [Rae et al., 2021]), for gender bias, and on the Bias
Benchmark for QA (BBQ-lite) [Parrish et al., 2021].
Results for sentiment towards different racial... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
Tools are extensions of the capabilities of tool users. When faced with complex tasks, humans employ
tools to simplify task-solving and enhance efficiency, freeing time and resources. Similarly, agents
have the potential to accomplish complex tasks more efficiently and with higher quality if they also
learn to use and ... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
discussing the responsible AI studies, thank William Ngan, Somya Jain, Lydia Baillergeau, Dana
Beaty, Chantal Mora, Daniel Duncan, Gopika Jhala, Steph Miles, Josh Terry, Valeryia Aranovich,
Ashton Evans, Aly Gill, Andrea Mileskiewicz, Emily Richards, and Aaron Vasquez for developing
the visual assets and the website, t... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
Experiment 4: Generalization between operations. Next we explore whether
a model that was trained on one type of arithmetic problems can generalize to
other types. We conduct two types of experiments: one where we examine the
10
generalization ability of Jurassic-X on single operation problems, and one with two-
ope... | MRKL Systems |
Most recently, Lin et al. [110] propose a benchmark, which comprises 817 questions that span 38
categories, to measure the truthfulness of a language model in the QA task. This work investigates
the performances of GPT-3 [16], GPT-Neo/J [192], GPT-2 [149] and a T5-based model [150]. The
results suggest that simply scal... | SurveyofHallucinationinNatural Language Generation |
Freda Shi, Mirac Suzgun, Markus Freitag, Xuezhi Wang, Suraj Srivats, Soroush Vosoughi, Hyung Won
Chung, Yi Tay, Sebastian Ruder, Denny Zhou, et al. Language models are multilingual chain-of-
thought reasoners. ICLR, 2023.
Amanpreet Singh, Vivek Natarajan, Meet Shah, Yu Jiang, Xinlei Chen, Dhruv Batra, Devi Parikh, and... | gemini_1_report |
25https://openai.com/blog/chatgpt-plugins
30 | Tool Learning with Foundation Models |
Projected capabilities
Future frontier AI will likely feature even greater content-level knowledge, reasoning abilities,
and capacity to formulate complex plans. Additionally, some expect that future capabilities will
make experimental instructions more accessible, including through the ability to generate
images... | Capabilities and risks from frontier AI |
Next, take (cid:15)2. (A4) guarantees that p satisfies the consistency conditions for univariate KDE (Silverman, 1986; Wand
and Jones, 1994; Gramacki, 2018), while condition (v) of (A3) ensures that within-leaf sample size increases even as leaf
volume goes to zero (Meinshausen, 2006, Lemma 2). Our kernel is a nonnegati... | Adversarial Random Forests for Density Estimation and Generative Modeling |
Observation: CC1=C(C=C(C=C1)NC(=O)C2=CC=C(C=C2)CN3CCN(CC3)C)NC4=NC=CC(=N4)C5=CN=CC=C5
Thought: I need to modify this compound to make a novel compound
Action: Modify compound
Action Input: CC1=C(C=C(C=C1)NC(=O)C2=CC=C(C=C2)CN3CCN(CC3)C)NC4=NC=CC(=N4)C5=CN=CC=C5
Observation: Cc1ccc(NC(=O)c2ccc(CN3CCNCC3)cc2)cc1Nc1nccc(-... | gpt-4-system-card |
source. It may have some grammar mistakes.
the grammar is correct.
How was the specific wording of the task instructions generated? A research team iteratively generated instructions,
making adjustments to create a template that could achieve high inter-annotator agreement.
At a high level, what aspects of the task ar... | PaLM 2 Technical Report |
Most “state-of-the-field” edited volumes end with a list of important
next steps. These often include research questions and aspirations for new
types of data collection. For research on the study of social media and
democracy, we find ourselves in a somewhat unusual position. The data we
need to conduct our research are... | Social_Media_and_Democracy |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.