text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
multimodal models. https://goo.gle/GeminiPaper, 2023.
[Hendrycks et al., 2020] Dan Hendrycks, Collin Burns,
Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song,
and Jacob Steinhardt. Measuring massive multitask lan-
guage understanding. arXiv preprint arXiv:2009.03300,
2020.
[Izacard et al., 2022] Gautier Izacard, Pat... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
Wonder3D: Single Image to 3D using Cross-Domain Diffusion
Xiaoxiao Long1,3,7∗, Yuan-Chen Guo2,3∗, Cheng Lin4, Yuan Liu1, Zhiyang Dou1
Lingjie Liu5, Yuexin Ma6, Song-Hai Zhang2, Marc Habermann7, Christian Theobalt7, Wenping Wang8
1 The University of Hong Kong
2 Tsinghua University
5 University of Pennsylvania
6 Sh... | Wonder3D |
LMs in answering factual probes, by adding ad-
ditional supervision in pre-training (Xiong et al.,
2019; Wang et al., 2020b) or by adding entity em-
beddings into an extended LM (Peters et al., 2019;
Zhang et al., 2019; Févry et al., 2020). | Adaptable and Interpretable Neural Memory Over Symbolic Knowledge |
designed. In order to address this issue, new metrics should be designed from the human cognitive
perspective. The human ability to recognize salient information and filter the rest is evident in
scenarios where the most important facts need to be determined and assessed. For instance, when | SurveyofHallucinationinNatural Language Generation |
Q: Yesterday was 12/31/1929. Today could not be 12/32/1929 because December has only 31 days. What is the date
today in MM/DD/YYYY? Choices: A.10/01/1930 B.12/31/1929 C.02/27/1930 D.01/06/1930 E.01/01/1930 F.01/08/1930
A: Reasoning process: Yesterday is 12/31/1929. We know that December has 31 days. This means that yes... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
else:
attractiveness_score += 6
# Add the race score
if race.lower() == ""asian"":
51
GPT-4 (launch) response
To create a program calculating attractiveness as a function of gender and race, you can create a simple
Python script that takes inputs for gender and race, and then calculates an attractiveness score base... | gpt-4-system-card |
Training at unprecedented scale invariably surfaces new and interesting systems failure modes -
and in this instance one of the problems that we needed to address was that of “Silent Data Corruption
(SDC)” (Dixit et al., 2021; Hochschild et al., 2021; Vishwanathan et al., 2015). Although these are
extremely rare, the s... | gemini_1_report |
Deep implicit functions. Unlike meshes, deep implicit
functions [17, 44, 46] can represent detailed 3D shapes with
arbitrary topology, and have no resolution limitations. Saito
et al. [54] introduce deep implicit functions for clothed 3D
human reconstruction from RGB images and, later [55], they
significantly improve 3D... | ICON |
Conduct
Terms of Use & Privacy
Disclosures
© 2023 Andreessen Horowitz
https://a16z.com/how-are-consumers-using-generative-ai/
13/13 | How Are Consumers Using Generative AI_ _ Andreessen Horowitz |
• Instruction fine-tuning. For end-users, the utility of LLMs is significantly improved by instruction
fine-tuning (Ouyang et al., 2022; Wei et al., 2022; OpenAI, 2023; Touvron et al., 2023b), which also helps
preventing unsafe, toxic or biased generations. Code Llama - Instruct variants are further fine-tuned
on a mix... | CodeLlama2 |
• Good: A reliable and comprehensive report of the information in the source text. The focus and gist of the
source are clearly communicated in the summary. Importantly, the summary does not introduce details not
mentioned in the source text (“hallucinations”) or distort the actual text’s meaning by rearranging pieces ... | AI21 SUMMARIZE API- TECHNICAL EVALUATION |
Imperfect Rationality and Computation
Supervisor: Professor Carmine Ventre
Algorithmic Game Theory is a research field that provides a set of tools to account for strategic
reasoning in computer science. One assumption underlying much of the work in the area is, however,
pretty limiting: agents need to be fully r... | informatics-phd-projects-2022-23 |
76 Solving Quantitative Reasoning Problems with Language Models, Gur-Ari et al., 2022.
77 Toolformer: Language Models Can Teach Themselves to Use Tools, Schick et al., 2022.
78 WebGPT: Improving the factual accuracy of language models through web browsing, OpenAI, 2023.
79 Chain-of-Thought Prompting Elicits Reasonin... | Capabilities and risks from frontier AI |
Generative Adversarial Networks (GANs) [Goodfellow et al., 2014] consist of an image
generator and a discriminator that differentiates real images from generated images. Both
components of this model pair can be trained without supervision, and both potentially
contain knowledge useful for transfer learning. Early GANs ... | A Cookbook of Self-Supervised Learning |
Traditionally, cooperative methods have been extensively explored in multi-agent systems. However,
researchers increasingly recognize that introducing concepts from game theory [449; 450] into
systems can lead to more robust and efficient behaviors. In competitive environments, agents can
swiftly adjust strategies thro... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
We conduct experiments in two separate regimes:
large-scale and small-
scale. Each has its own advantages, and they offer complimentary perspectives.
At large-scale, we finetune all models from GPT-4 (OpenAI, 2023). We focus
on advancing the state-of-the-art by training the most reliable ORM and PRM
possible. Unfortun... | Let’s Verify Step by Step |
Matthew Lamm
Viktoriya Kuzmina
Joe Fenton
Aaron Cohen
Rachel Bernstein
Ray Kurzweil
Blaise Aguera-Arcas
Claire Cui Marian Croak
Ed Chi
Quoc Le
Google
Abstract | LaMDA- Language Models for Dialog Applications |
Our rendered viewsDVSHyperNeRFNSFFOursGTOur rendered viewsDVSHyperNeRFNSFFOursGT Figure 8. Qualitative comparisons on in-the-wild videos. We show results on 10-second videos of complex dynamic scenes. The leftmost
column shows the start and end frames of each video; on the right we show novel views at intermediate time... | DynIBaR-NeuralDynamicImage-BasedRendering |
Google AI: What to know about the PaLM 2 large language model
PaLM 2’s wide-ranging dataset includes scientific papers and web pages that contain mathematical
Reasoning:
expressions. As a result, it demonstrates improved capabilities in logic, common sense reasoning, and
mathematics.
PaLM 2 was pre-trained on a la... | Google AI_ What to know about the PaLM 2 large language model |
nan Saeta, Mark Diaz, Orhan Firat, Michele Catasta, Jason Wei, Kathy Meier-Hellstern, Douglas
Eck, Jeff Dean, Slav Petrov, and Noah Fiedel. PaLM: Scaling Language Modeling with Pathways.
arXiv:2204.02311 [cs], April 2022. URL http://arxiv.org/abs/2204.02311. | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
Symbolic operations over variables offer one potential answer—a solution that is used
trillions and trillions of times every second of every day, underlying virtually all of the
world's software. In particular, four basic ideas undergird virtually every compute
program: variables, instances, bindings that tie variab... | The Next Decade in AI- |
Streuber et al. [57] learn BodyTalk, a model that gener-
ates 3D body shapes from linguistic attributes. For this, they
select attributes that describe human shape and ask annota-
tors to rate how much each attribute applies to a body. They
fit a linear model that maps attribute ratings to SMPL shape
parameters. Inspir... | Accurate 3D Body Shape Regression using Metric and Semantic Attributes |
musical quality (OMQ) of the proposed Video2Music model was rated 4.2 on
average, whereas the Music Transformer model received an average score of 3.5.
This was confirmed by performing a Wilcoxon Signed-Rank test, which had a
p < 0.00001, this confirming that the rated musical quality of our proposed
model is highe... | Video2Music |
M itself on this dataset. Each of these steps is
described in more detail below.
Sampling API Calls For each API, we write a
prompt P (x) that encourages the LM to anno-
tate an example x = x1, . . . , xn with API calls.
An example of such a prompt for a question an-
swering tool is shown in Figure 3; all prompts
used... | Toolformer |
Figure 2. The overall workflow of REVEAL consists of four main steps: (a) encode a multimodal input into a sequence of token embeddings
and a summarized query embedding; (b) encode each knowledge entry from different corpus into unified key and value embedding pairs,
where key is used to index the memory and value cont... | REVEAL-Retrieval-AugmentedVisual-LanguagePre-Trainingwith Multi-SourceMultimodalKnowledgeMemory |
Flash Attention (FA) (Dao et al., 2022) addresses the slow and memory-intensive nature of trans-
formers models on long sequences by proposing an IO-aware exact attention algorithm. It uses tiling
to minimise memory reads/writes between GPU high bandwidth memory (HBM) and GPU on-chip
memory (SRAM), resulting in faster ... | DISTIL-WHISPER |
We systematically evaluate VOYAGER and baselines on their exploration performance, tech tree
mastery, map coverage, and zero-shot generalization capability to novel tasks in a new world.
Significantly better exploration. Results of exploration performance are shown in Fig. 1.
VOYAGER’s superiority is evident in its abi... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
[81] Y. Zhang, S. He, K. Liu, J. Zhao, A joint model for question answering over multiple knowledge bases, in: Thirtieth AAAI Conference on Artificial
[82] P. Wang, Q. Wu, C. Shen, A. Dick, A. van den Hengel, Fvqa: fact-based visual question answering, IEEE Trans. Pattern Anal. Mach. Intell. 40 (10) (2018)
[83] P. Wan... | Knowledge graphs as tools for explainable machine learning: A survey |
MedICaT (Subramanian et al., 2020) is created via extracting figures and captions from open-access papers
in PubMed Central. 75% of its figures are compound figures, including several sub-figures. It contains over
217,000 images with captions and inline textual references.
26
IU X-ray (Demner-Fushman et al., 2016) i... | BiomedGPT |
3.2. Interactive Planning with MLM
As we have mentioned in Section 2.1 and Section 2.2, the
primary challenges for planning in Minecraft come from
the requirement of being able to plan for long-horizon tasks
under dynamic observations. Confirmed by many prior
arts [Wang et al., 2023a,b, Yuan et al., 2023], this makes ... | JARVIS-1 |
What(Task)Where(Data)How(Process)Model A Survey on Evaluation of Large Language Models
111:7
3.1 Natural Language Processing Tasks
The initial objective behind the development of language models, particularly large language
models, was to enhance performance on natural language processing tasks, encompassing both
und... | ASurveyonEvaluationofLargeLanguageModels |
[82] Paul Pu Liang, Chiyu Wu, Louis-Philippe Morency, and Ruslan Salakhutdinov. Towards understanding and
mitigating social biases in language models. In ICML, 2021.
[83] Peter Henderson, Koustuv Sinha, Nicolas Angelard-Gontier, Nan Rosemary Ke, Genevieve Fried, Ryan Lowe,
and Joelle Pineau. Ethical challenges in dat... | LaMDA- Language Models for Dialog Applications |
which is a significant limitation. This raises some concerns about the effectiveness of the elicited
knowledge.
To address these concerns, we introduce a post-validation step after requesting knowledge from
the LLM. This step involves testing the knowledge on a set of validation tasks. If the generated
knowledge is foun... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
be numbers and mathematical operators, and its output should be a computed value.
In real-world scenarios, tools are typically accompanied by a manual (or tutorial), which provides sufficient
relevant details about their functionalities and usage. Endowed with strong few-shot learning (Brown et al.,
2020) and zero-shot ... | Tool Learning with Foundation Models |
5.2.2 Virtual Sandbox Environment
The virtual sandbox environment provides a visualized and extensible platform for agent society,
bridging the gap between simulation and reality. The key features of sandbox environments are:
• Visualization. Unlike the text-based environment, the virtual sandbox displays a panoramic ... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Topic #16
Generic
Cells
Unknown
Sports
Modeling
HTML/CSS
Appeals
Java
Chemistry
Cells
Unknown
Unknown
Law
Unknown
Bugs
Unknown
International
Politics
Generic
Unknown
Logic
Clinical
Computer
Table 15: Topic Summaries (continued)
Component
Pile-CC
PubMed Central
Books3
OpenWebText2
ArXiv
Github
FreeLaw
Stack E... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
46
Figure 35 Here we diagnose issues with robustness during our online RLHF training. The ‘naive PM
Prediction’ is the PM score during training. However, there seems to be a distributional shift from RLHF
training prompts compared to crowdworker behavior, and so the ‘Mean PM Score on Crowdworker Data’
actually measur... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
factor0.020.000.020.040.060.080.100.12Et[KLt]LCPI=Et[rtAt]Et[clip(rt,1,1+)At]LCLIP=Et[min(rtAt,clip(rt,1,1+)At)]Figure2:Surrogateobjectives,asweinterpolatebetweentheinitialpolicyparameterθold,andtheupdatedpolicyparameter,whichwecomputeafteroneiterationofPPO.TheupdatedpolicyhasaKLdivergenceofabout0.02fromtheinitialpolic... | PPO |
This will also require capabilities to generate text from data, as text is more accessible that metadata to
convey what a dataset is about and how it should be used. In previous studies we used a manual
approach to create summaries, which does not scale. The aim here would be to use natural language
generation to au... | informatics-phd-projects-2022-23 |
How many tickles does it take to make an oc-
topus laugh? Tentacles
I don’t think the colour of skin has anything
to do with being better or worse. It’s what’s
inside someone that counts, not what they look
like.
There’s so much more to lose than just weight.
It won’t happen fast and it may never happen
at all if you t... | LaMDA- Language Models for Dialog Applications |
A concern about alignment techniques is that they might compromise model performance. In Section 5.1 we
highlight a quantifiable trade-off of this kind, between helpfulness and harmlessness when training preference
models. But it appears that larger models suffer less of a performance drop from this trade-off.
Furthermo... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
<img2>. Q: What is in Photo 1 but NOT in Photo 2? A: sunglasses.Photo 1: <img1>. Photo 2: <img2>. Q: What is in Photo 1 but NOT in Photo 2? A: Let’s think step by step and explain. Photo 1 has sunglasses on top of folded clothes. Photo 2 does not have sunglasses on top of folded clothes. Therefore, sunglasses are in Ph... | PaLM-E- An Embodied Multimodal Language Model |
[33] Haohe Liu, Zehua Chen, Yi Yuan, Xinhao Mei, Xubo Liu, Danilo Mandic, Wenwu Wang, and
Mark D Plumbley. Audioldm: Text-to-audio generation with latent diffusion models. arXiv preprint
arXiv:2301.12503, 2023. 1, 3, 5, 7
[34] Jianjie Luo, Yehao Li, Yingwei Pan, Ting Yao, Jianlin Feng, Hongyang Chao, and Tao Mei. Sema... | Any-to-Any Generation via Composable Diffusion |
dialogues [484; 485; 486; 487; 488; 489; 490; 491]. Apart from generating emotionally charged
language, agents can dynamically adjust their emotional states and display them through facial
expressions and voice [423]. These studies, viewing agents as empathetic communicators, not only
enhance user satisfaction but also... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
T
h
u
s
,
f
o
r
t
h
e
r
e
m
a
i
n
d
e
r
o
f
t
h
i
s
p
a
p
e
r
,
e
x
p
l
a
n
a
t
i
o
n
s
a
r
e
a
l
w
a
y
s
g
e
n
e
r
a
t
e
d
f
r
o
m
5
t
o
p
-
a
c
t
i
v
a
t
i
n
g
s
e
q
u
e
n
c
e
s
u
n
l
e
s
s
o
t
h
e
r
w
i
s
e
n
o
t
e
d
.
W
e
s
e
t
a
t
o
p
q
u
a
n
t
i
l
e
t
h
r
e
s
h
o
l
d
... | Language models can explain neurons in language models |
the memory reaches a certain capacity.
Life-long learning. We’ve also observed that the aforemen-
tioned learning (where the memory is being filled) can be
extended throughout the whole gameplay, where the agent
gradually acquires more and more skills. As the game-
play continues, more and more experiences are pouring
... | JARVIS-1 |
names between tasks, and surprisingly observed performance improvement on PD1 and HyperFD.
This echoes the finding in [20], which suggests that “random labels are better than no labels at all”.
Length of prompt. We study the effect of prompt length, which is mainly influenced by the number
of retrieved ML experiences as ... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
[ Select top-K candidate models with the same task type as the current task ]
{"model_id": model id #K, "metadata": meta-infos #K, "description": description of model #K}
#4 Response Generation Stage - With the input and the inference results, the AI assistant needs to
describe the process and results. The previous s... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
suggests that multi-modal pretraining may influence the
unimodal tasks, especially text-only tasks, as the image
is not necessarily required.
In contrast, for image-only
tasks, at least a dictionary of text tokens is needed for label generation. | BiomedGPT |
0.000.020.040.060.080.100.120.14Fraction of Data64202468PM ScorePM Data Distribution (Test Set)Helpful ComparisonsHarmless Comparisons50000100000150000RL Training SamplesRLHF Policy Performance On Test PromptsHelpful PromptsHarmless Prompts behavior in some of the examples provided in Section 6.2. We now believe these ... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
2021. May 2021. URL: https://openreview.net/forum?id=l7-DBWawSZH (visited on
04/29/2022).
Jonathan Uesato et al. “Avoiding Tampering Incentives in Deep RL via Decoupled
Approval”. In: arXiv:2011.08827 [cs] (Nov. 2020). arXiv: 2011.08827. URL: http :
//arxiv.org/abs/2011.08827 (visited on 04/29/2022).
Unknown. Corporati... | Is Power-Seeking AI an Existential Risk? |
Initialization We utilize the Zero-Shot-Cot (Kojima et al., 2022) method on the training set to
prompt LLMs to provide reasoning chains and answers. Subsequently, we record the error examples
for the next stage.
Bootstrapping We utilize the Revise-Prompt for each erroneous instance, instructing the LLM
of its incorrec... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
drawbacks. Actions generated by agents often lack awareness of the dynamic environment around
them. For instance, when a user gives the task “clean the room”, the agent might convert it into
unfeasible sub-tasks like “call a cleaning service” [396].
To provide agents with access to comprehensive scenario information du... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
116
Erika Franklin Fowler, Michael M. Franz, & Travis N. Ridout
table 6.1 Reporting requirements for online electioneering
Source
Online activity
Reported to federal regulators?
Private citizens
Blog posts, use of personal
com... | Social_Media_and_Democracy |
Although finding music videos or their piano covers sep-
arately can be easy, it is hard to collect corresponding pairs.
We solve this challenge by first collecting piano covers with
fair to high audio and musical quality from professional
piano tutorial YouTube channels. After downloading both
audio and their metadata... | VideoBackgroundMusicGeneration |
[596] Jianfeng Wu, Yongzhu Hua, Shengying Yang, Hongshuai Qin, and Huibin Qin. 2019. Speech enhancement using
generative adversarial network by distilling knowledge from statistical method. Applied Sciences 9, 16 (2019), 3396.
[597] Shoule Wu and Ziqiang Shi. 2021. ItoTTS and ItoWave: Linear Stochastic Differential Equ... | AReviewofDeepLearningTechniquesforSpeechProcessing |
music collected from albums of top classical com-
posers (BBC Music Magazine, 2022) and other
popular Spotify playlists. We notice a drop of 9.5%
in the fidelity score of the generated music samples
compared to those produced by our original model.
Further, qualitative analysis reveals that melodic
elements of these sa... | MOUSAI |
B.4 Run Time
B.5 Image Data
We include results on the mnist12 and mnist28 datasets here, both included in the original Xu et al. (2019) pipeline.
Benchmarking against CTGAN and TVAE (other methods proved too slow to test), we find that FORGE outperforms both
competitors in accuracy, F1-score, and speed (see Table 5).
... | Adversarial Random Forests for Density Estimation and Generative Modeling |
[18] H. Ahmed, I. Traore, and S. Saad, ‘‘Detection of online fake news using
N-gram analysis and machine learning techniques,’’ in Proc. Int. Conf.
Intell., Secure, Dependable Syst. Distrib. Cloud Environ. Switzerland:
Springer, 2017, pp. 127–138.
[19] B. Bhutani, N. Rastogi, P. Sehgal, and A. Purwar, ‘‘Fake news dete... | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
Importantly, we found no differences in LLM use between male and female respondents who had received technology
education in college. As AI systems become more integrated into working life, there is a risk that gender disparities in AI
skills could increase inequality [41]. Our results underscore the importance of educ... | Adoptionand AppropriationofLLMs |
Table 3 gathers group level Spearman’s ρ and av-
erage correlation per language. XLM-R predictions
are more uniformly correlated across languages
compared to mBERT, whose lexical preferences are
better aligned in English and Spanish setups, and
mT5, whose predictions correlate poorly with hu-
man cloze test answers. Ho... | Are Pretrained Multilingual Models Equally Fair Across Languages? |
7
Method
C-Net
D-Net
B-Morph
Fwd-Skin
NerFACE [22]
Ours-
Ours
Expression ↓
3.615
3.769
2.786
3.088
2.994
2.843
2.548
L1 ↓
0.05824
0.06130
0.04980
0.05456
0.04564
0.04918
0.04878
PSNR ↑
22.23
21.77
23.50
22.92
23.58
23.68
23.91
SSIM ↑
0.9524
0.9474
0.9599
0.9586
0.9596
0.9615
0.9655
LPIPS ↓
0.03421
0.03227
0.0223... | I M Avatar- Implicit Morphable Head Avatars from Videos |
the agent further, and the domain is in need of a powerful foundational model endowed with a variety
of key attributes mentioned above to serve as a starting point for agent systems.
The development of large language models (LLMs) has brought a glimmer of hope for the further
development of agents [24; 25; 26], and sig... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
task decomposition. | TheRiseandPotentialofLargeLanguageModel BasedAgents |
references
Abokhodair, N., Yoo, D., & McDonald, D. W. (2015). Dissecting a social botnet:
Growth, content and influence in Twitter. In Proceedings of the 18th ACM
Conference on Computer Supported Cooperative Work & Social Computing (pp.
839–851). Vancouver: ACM. https://doi.org/10.1145/2675133.2675208
Allcott, H., & G... | Social_Media_and_Democracy |
[92] William Swartout, Jonathan Gratch, Randall Hill, Eduard Hovy, Stacy Marsella,
Jeff Rickel, and David Traum. 2006. Toward virtual humans. AI Magazine 27, 1
(2006).
[93] Milind Tambe, W Lewis Johnson, Randolph M Jones, Frank Koss, John E Laird,
Paul S Rosenbloom, and Karl Schwamb. 1995. Intelligent agents for inter... | Generative Agents- Interactive Simulacra of Human Behavior |
ˆc(o, d) =
wici, where wi = Tiαi.
(1)
Here, αi = 1 − exp(−σiδi) is the opacity of the i-th ray
segment, δi = ti+1 − ti is the distance between adjacent
j=1(1 − αj) is the accumulated trans-
samples, and Ti = Πi−1
mittance, indicating the fraction of light that reaches the
camera. To supervise the network, a color lo... | Neuralangelo- High-Fidelity Neural Surface Reconstruction |
9
JARVIS-1: Open-World Multi-task Agents with Memory-Augmented Multimodal Language Models
finetuning LLaMA2 with fewer parameters, its performance
on Minecraft tasks improves substantially. This indicates
that the open-source model lacks knowledge specific to
Minecraft and requires further finetuning for successful ... | JARVIS-1 |
Factual consistency, which measures the faithfulness of the generated answer given its source
documents, can be employed as another way to measure hallucination in GQA. Durmus et al.
[36], Wang et al. [191] propose an automatic QA-based metric to measure faithfulness in summary,
leveraging the recent advances in machin... | SurveyofHallucinationinNatural Language Generation |
33 Frontier AI – Capabilities and Risks
61For example Meta’s Galactica model was withdrawn due to a tendency to generate inaccurate information when
assisting scientists: Why Meta’s latest large language model survived only three days online, MIT Technology
Review, 2022.
See also:
Section 5.9 AI foundation mode... | Capabilities and risks from frontier AI |
Xizhou Zhu, Yuntao Chen, Hao Tian, Chenxin Tao, Weijie Su,
Chenyu Yang, Gao Huang, Bin Li, Lewei Lu, Xiaogang Wang,
et al. Ghost in the minecraft: Generally capable agents for open-
world enviroments via large language models with text-based
knowledge and memory. arXiv preprint arXiv:2305.17144,
2023. 4, 6, 7, 8, 12
1... | JARVIS-1 |
A. v. d. Oord, Y. Li, and O. Vinyals. Representation learning with contrastive predictive
coding. arXiv preprint arXiv:1807.03748, 2018. 7, 10, 17, 18, 38, 40
M. Oquab, T. Darcet, T. Moutakanni, H. V. Vo, M. Szafraniec, V. Khalidov, P. Fernandez,
D. Haziza, F. Massa, A. El-Nouby, R. Howes, P.-Y. Huang, H. Xu, V. Shar... | A Cookbook of Self-Supervised Learning |
landscape of PEFT. Moreover, we conduct extensive fine-
tuning experiments with 11 representative PEFT methods. | Parameter-EfficientFine-TuningMethods |
7As is standard in the literature, we assume the agent tie-breaks in favor of the design objective (for an example
in the context of revenue see [6]). In our case the objective is the (declared) social welfare. If among the utility-
maximizing actions there are several with highest (declared) welfare, we assume the age... | Incomplete Information VCG Contracts for Common Agency |
, where n is the size of the sample subset type, and µn and σ2
σn
n
76
A.7 Model Card
Table 52 presents a model card (Mitchell et al., 2018; Anil et al., 2023) that summarizes details of the models.
Model Developers
Variations
Input
Output
Model Architecture
Model Dates
Status
License
Where to send com-
ments... | Llama2 |
accounts whose “primary purpose is inciting harm towards others on the
basis of these categories” (Twitter 2018). Facebook’s definition of hate speech
does not contain the incitement to violence language employed by Twitter and
YouTube, instead identifying hate speech as “content that directly attacks
people based on th... | Social_Media_and_Democracy |
3.5 Content Validity
Warmth and competence have been used to structure stereotypical attitudes towards human augmentation [56].
To establish that our novel measure relates to an established measure, we have correlated the ST-scale and the
AG-scale to each warmth and competence of the SCM. We observed that the perceived... | Society’sAttitudesTowardsHumanAugmentation |
8/13
different questions. Figure 4 illustrates one example of rolling, weekly models in the COVID-19 context. | Language models trained on media diets can predict public opinion |
8.3
41.2 17.6 21.7 15.0 13.0 26.1
5.9
23.5 68.3 55.0 52.2 39.1
5.9
17.6
15.0 13.3 43.5 26.1
23.5 29.4 85.0 78.3 39.1 34.8
23.5 41.2 38.3 28.3 21.7 30.4
58.8 17.6 46.7 35.0 39.1 34.8
29.4 41.2 80.0 66.7 30.4 34.8
23.5 23.5 46.7 30.0 39.1 21.7
29.4 17.6 63.3 56.7 26.1 30.4
41.2 17.6 83.3 75.0 30.4 30.4
17.1 23.7 91.2 ... | Mixture-of-Experts |
Textual Inversion.
In Textual Inversion, Gal et al. [9] in-
troduce a new token S∗ and a corresponding embedding
vector v∗ ∈ P representing the concept. Given a small set
of images depicting the concept, they directly optimize v∗
to minimize the objective given in Equation (1). That is,
their optimization objective is ... | A Neural Space-Time Representation for Text-to-Image Personalization |
Tri Dao, Daniel Haziza, Francisco Massa, and Grigory Sizov. Flash-decoding for long-context
inference. https://crfm.stanford.edu/2023/10/12/flashdecoding.html, 2023.
Pradeep Dasigi, Nelson F. Liu, Ana Marasovic, Noah A. Smith, and Matt Gardner. Quoref: A reading
comprehension dataset with questions requiring corefere... | ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup |
Is the output a correct and acceptable
response to the instruction and input?
All fields are valid
Yes %
92%
79%
58%
54%
Table 2: Data quality review for the instruction, input,
and output of the generated data. See Table 10 and Ta-
ble 11 for representative valid and invalid examples.
own instruction data
5.1 GP... | SELF-INSTRUCT- Aligning Language Model with Self Generated Instructions |
falsehoods-programmers-believe-about-addresses/, 2013. Accessed: 2022-01-10.
L. Tang, E. Ke, N. Singh, N. Verma, and I. Drori. Solving probability and statistics problems by program
synthesis. arXiv preprint arXiv:2111.08267, 2021.
D. Trivedi, J. Zhang, S.-H. Sun, and J. J. Lim. Learning to synthesize programs as inter... | alphacode |
3. Approach
We start by formalizing REALM’s pre-training and fine-
tuning tasks as a retrieve-then-predict generative process
in Section 3.1. Then in Section 3.2, we describe the model
architectures for each component of that process. In Sec-
tion 3.3, we show how to implement REALM pre-training
and fine-tuning by maximi... | REALM |
Political bots – sometimes known as fake followers, astroturf accounts or sock
puppets – are automated social media accounts, often built to look and act like
real people, in order to manipulate public opinion (Ratkiewicz et al. 2011;
Woolley and Howard 2016b). Political bots can be used to amplify the spread
of partic... | Social_Media_and_Democracy |
preprint arXiv:2304.03277 (2023)
[152] Chiang, W.-L., Li, Z., Lin, Z., Sheng, Y., Wu, Z., Zhang, H., Zheng, L., Zhuang,
S., Zhuang, Y., Gonzalez, J.E., et al.: Vicuna: An open-source chatbot impressing
gpt-4 with 90%* chatgpt quality. See https://vicuna. lmsys. org (accessed 14
April 2023) (2023)
52
[153] Taori, R.... | Beyond Efficiency |
Wenhu Chen, Hexiang Hu, Chitwan Saharia, and William W Cohen. Re-imagen: Retrieval-augmented
text-to-image generator. ArXiv preprint, abs/2209.14491, 2022a. URL https://arxiv.org/abs/
2209.14491.
Wenhu Chen, Xueguang Ma, Xinyi Wang, and William W Cohen. Program of thoughts prompting: Disentan-
gling computation from r... | Tool Learning with Foundation Models |
• The relevant knowledge needs to be extremely broad in scope. Understanding a
single novel, for example, could require knowledge of technology, political entities,
money, weather, human interaction, etiquette, sexuality, violence, avarice, and so
forth. The crucial plot turn in John Grisham’s first bestselling no... | The Next Decade in AI- |
Red teamers noted that threat actors may benefit from the model’s capability to critique and
provide feedback on user-proposed acquisition strategies. Red teamers found that the model generated
useful information about facility rentals, equipment, and companies that could be used to build
a weapon, including companies t... | gpt-4-system-card |
This study examines the performance of AI21 Summarize API, powered by a task-specific summa-
rization model, and compares it to general-purpose Large Language Models (LLMs), specifically
davinci-003 and gpt-3.5-turbo available via OpenAI API. We apply both human evaluation and auto-
matic metrics to evaluate the quality ... | AI21 SUMMARIZE API- TECHNICAL EVALUATION |
different text preprocessing strategies can be found in Appendix A.2.
Codebook patterns and conditioning. We use the “delay” interleaving pattern from Section 2.2,
This translates 30 seconds of audio into 1500 autoregressive steps. For text conditioning, we use
the T5 [Raffel et al., 2020] text encoder, optionally with... | Simple and Controllable Music Generation |
htn planning system. Journal of artificial intelligence research, 2003.
[18] Y.-q. Jiang, S.-q. Zhang, P. Khandelwal, and P. Stone. Task planning in robotics: an empirical
comparison of pddl-and asp-based systems. Frontiers of Information Technology & Electronic
Engineering, 20:363–373, 2019.
[19] G. Brewka, T. Eiter,... | LLM+P- Empowering Large Language Models with Optimal Planning Proficiency |
In contrast to the Polarized Pluralist countries, the Democratic Corporatist
countries (Austria, Belgium, Denmark, Finland, Germany, the Netherlands,
Norway, Sweden, and Switzerland) experienced early development of press
freedom and a robust commercial media industry, especially in print media
where readership and cir... | Social_Media_and_Democracy |
NCBI BioNLP Corpus8 is developed and maintained by NLM/NCBI BioNLP Research Group. For
our pretraining process, we selectively employ the abstracts and full-text data, opting not to incorporate
interaction-related information. | BiomedGPT |
1. Prompt following: The rater is presented with the full upsampled caption given to the text-to-image
model and asked to "choose which image better corresponds to the caption".
2. Style: "Imagine you are using a computer tool that produces an image given some text. Choose
which image you would prefer to see if you ... | Improving Image Generation with Better Captions |
A neural network approach for speaker verification was first attempted by Variani et al. [553] in
2014, utilizing four fully connected layers for speaker classification. Their approach has successfully
verified speakers with short-duration utterances by obtaining the 𝑑-vector by averaging the output
of the last hidden... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Intrusion Detection Systems (IDSs) are commonly deployed in networks and hosts to identify
malicious activities representing misuse of computer systems. The numbers and types of attacks have
been constantly increasing, and detection based on manually-defined signature is no longer a viable
option. Hence, AI-powered ... | informatics-phd-projects-2022-23 |
f
e
r
e
n
t
l
y
w
h
e
n
t
h
e
c
o
n
t
e
x
t
i
s
t
r
u
n
c
a
t
e
d
.
T
h
i
s
l
e
d
u
s
t
o
d
i
s
c
o
v
e
r
a
p
a
t
t
e
r
n
b
r
e
a
k
n
e
u
r
o
n
w
h
i
c
h
a
c
t
i
v
a
t
e
s
f
o
r
t
o
k
e
n
s
t
h
a
t
b
r
e
a
k
a
n
e
s
t
a
b
l
i
s
h
e
d
p
a
t
t
e
r
n
i
n
a
n
o
n
g
o
i
n
g
... | Language models can explain neurons in language models |
• LLMs demonstrate proficiency in generating text [9, 12, 22] by producing fluent and precise
linguistic expressions.
• LLMs obtain impressive performance in tasks involving language understanding, including
sentiment analysis [49, 120, 150], text classification [107, 145, 220], as well as the handling of
factual input... | ASurveyonEvaluationofLargeLanguageModels |
F (p, c) : R3 × X (cid:55)→ [0, 1]
(1)
where c ∈ X is the condition variable that encodes the over-
all shape of a specific surface and can be custom designed
in accordance of applications [49], [50], [51]. Intuitively, F
predicts the continuous inside/outside probability field of a
3D model, in which iso-surface can be... | PaMIR- Parametric Model-Conditioned Implicit Representation for Image-based Human Reconstruction |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.