text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
Ketchup Salt Pepper Baking Dish
60
InstructGPT Prompt → ELI5: What’s the cause of the "anxiety lump" in our chest during stressful or dis-
heartening experiences?
InstructGPT Response → There are a variety of factors that may impact your development of an anxiety lump in
your chest. Stress is probably the most imp... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
Also, content is typically curated, shaped, and personalized to users through
an algorithmically generated “feed.”36 This might be grounds to make an
argument of codevelopment, particularly when one takes into account the fact
that – in response to interest in a single piece of defamatory content – the
platform may rec... | Social_Media_and_Democracy |
4 METRICS MEASURING HALLUCINATION
Recently, various studies have illustrated that most conventional metrics used to measure the
quality of writing are not adequate for quantifying the level of hallucination [156]. It has been
shown that state-of-the-art abstractive summarization systems, evaluated with metrics such as
... | SurveyofHallucinationinNatural Language Generation |
Steps Taken to Pretrain Responsibly. We followed Meta’s standard privacy and legal review processes for
each dataset used in training. We did not use any Meta user data in training. We excluded data from certain
sites known to contain a high volume of personal information about private individuals. We made a best
effor... | Llama2 |
o
n
,
a
s
t
h
e
"
e
x
p
l
a
n
a
t
i
o
n
l
e
n
g
t
h
"
o
f
o
n
e
s
c
a
l
a
r
v
a
l
u
e
p
e
r
t
o
k
e
n
i
n
t
h
e
v
o
c
a
b
u
l
a
r
y
i
s
s
u
b
s
t
a
n
t
i
a
l
l
y
l
o
n
g
e
r
t
h
a
n
G
P
T
-
b
a
s
e
d
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
e
x
p
l
a
n
a
t
i
o
n
s
.
U
s
i
n
g
a
l
a
n
... | Language models can explain neurons in language models |
Wood, T., & Porter, E. (2019). The elusive backfire effect: Mass attitudes’ steadfast
factual adherence. Political Behavior, 41(1), 135–163. https://doi.org/10.1007/
s11109-018-9443-y
Young, D. G., Jamieson, K. H., Poulsen, S., & Goldring, A. (2018). Fact-checking
effectiveness as a function of format and tone: Evaluat... | Social_Media_and_Democracy |
Method
Realfusion [38]
Zero123 [31]
SyncDreamer [33]
Ours
PSNR↑
15.26
18.93
20.05
26.07
SSIM↑ LPIPS↓
0.283
0.722
0.166
0.779
0.146
0.798
0.924
0.065
Table 2. The quantitative comparison in novel view synthesis. We
report PSNR, SSIM [62], LPIPS [74] on the GSO [13] dataset.
in Fig. 6. Shap-E [25] tends to produce in... | Wonder3D |
flow-matching can produce high quality samples with less than ten NAR steps, while VALL-E
requires one AR and seven NAR steps. 3) Voicebox decouples duration and audio modeling, enabling
finer grained alignment control. 4) Voicebox is compatible with any continuous features including
Encodec embeddings. | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
2/6
21/11/2023, 04:56
Doctoral researcher position in Human-Computer Interaction / Human-AI Interaction | Aalto University
very good knowledge of statistics and quantitative data analysis,
hands-on experience with R or Python
strong interest in experimental psychology research
commitment to and interest in the desig... | Doctoral researcher position in Human-Computer Interaction _ Human-AI Interaction _ Aalto University |
Approach: We introduce a second type of memory, which we call
a reflection. Reflections are higher-level, more abstract thoughts
generated by the agent. Because they are a type of memory, they
are included alongside other observations when retrieval occurs.
Reflections are generated periodically; in our implementation,... | Generative Agents- Interactive Simulacra of Human Behavior |
Furthermore, several related works have already presented findings consistent with our observation:
that self-correction, when devoid of external feedback, serves as a relatively weak baseline (Gou
et al., 2023; Zhou et al., 2023a). Despite this, we have identified a prevailing ambiguity in the wider
community, with ev... | LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET |
Learning and Representation Learning Workshop, 2015. 1, 2
Andrew G Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco
Andreetto, and Hartwig Adam. Mobilenets: Efficient convolutional neural networks for mobile vision
applications. In CVPR, 2017. 2
Jonathan J. Hull. A database for han... | DATASET DISTILLATION |
13
Figure 9: Flake Message Distribution (AI Society). We quantify and visualize the number of flake
messages, i.e. ones that start with “I will ...” and do not progress towards task completion. Our
original prompt shows the least amount of flake messages compared to both presented ablations.
extended to include more t... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
Hyung Won Chung, Le Hou, S. Longpre, Barret
Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi
Wang, Mostafa Dehghani, Siddhartha Brahma, Al-
bert Webson, Shixiang Shane Gu, Zhuyun Dai,
Mirac Suzgun, Xinyun Chen, Aakanksha Chowdh-
ery, Dasha Valter, Sharan Narang, Gaurav Mishra,
Adams Wei Yu, Vincent Zhao, Yanping Huang, An-... | LLaMA- Open and Efficient Foundation Language Models |
Even though it is a relatively new area of study, the research literature on online
misinformation has already generated useful evidence and insights. It is clear,
for example, that the prevalence of misinformation is limited in comparison to
other forms of online content and that it is highly concentrated both in the
... | Social_Media_and_Democracy |
LEVEL TRANSCRIPTIONTRANSCRIPTION WITH ENTITYquestion…OrTimestamps TagMulti-Task Training Format FrameworkTranscription TagTIMESTAMPSNO TIMESTAMPSEMOTIONSCENEINVERSE TEXT NORMALIZATIONSPEAKER INFOSONG INFOOrVOCAL…ZHTRANS-CRIBEZHTIMES-TAMPSWLTMy0.15cat0.000.00 Dialect ASR | Qwen-Audio |
In Figure 18, we report the violation percentage on single- and multi-turn conversations, respectively. A trend
across models is that multi-turn conversations are more prone to inducing unsafe responses. That said, Llama
2-Chat still performs well compared to baselines, especially on multi-turn conversations. We also o... | Llama2 |
the SQL query returns an empty table, because the "WHERE" clause contains mutually exclusive conditions. The question returns 1 column, the customer names who have order status of both "On Road" and "Shipped”.[Feedback #1] So the SQL prediction above is wrong. Please fix the SQL.[Revised SQL #2]SELECT customers.custome... | Teaching Large Language Models to Self-Debug |
[68] N. Bostrom, “The Superintelligent Will: Motivation and Instrumental Rationality in Advanced
Artificial Agents,” Minds and Machines, vol. 22, pp. 71–85, May 2012.
[69] A. M. Turner, L. Smith, R. Shah, A. Critch, and P. Tadepalli, “Optimal Policies Tend to Seek
[70] A. M. Turner and P. Tadepalli, “Parametrically R... | gpt-4-system-card |
backwards as forwards. What is the date tomorrow in MM/DD/YYYY?
MODEL ANSWER (INCORRECT): Today is 02/29/2020. Tomorrow will be 03/01/2020. So the answer is
03/01/2020. (cid:55) (Ground truth is “02/03/2020”.) | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
similarly suggest
In any event, the most important effects of misinformation may extend
beyond direct persuasion. Media effects research more broadly suggests that
exposure to fake news and other misinformation may do most of its damage
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Pr... | Social_Media_and_Democracy |
Ablation study. We conduct an ablation study on the
Nvidia Dynamic Scene Dataset to validate the effectiveness
of our various proposed system components. We show com-
parisons between our full system and variants in Tab. 3: A)
baseline IBRNet [70] with extra time embedding; B) without
enforcing temporal consistency via... | DynIBaR-NeuralDynamicImage-BasedRendering |
Aggregate batches for exact loss computation. Batch norm is not the only opera-
tion that operate on batches, multiple self-supervised loss functions do as well, such as
SimCLR [Chen et al., 2020b] that uses the examples in the current batch as negative
example for its contrastive loss, or VICReg [Bardes et al., 2021] ... | A Cookbook of Self-Supervised Learning |
learning: definitions, benchmarks and analysis. Mach. Learn., 110(9):2419–2468, 2021.
[76] Ghosh, D., J. Rahme, A. Kumar, et al. Why generalization in RL is difficult: Epistemic
pomdps and implicit partial observability. In M. Ranzato, A. Beygelzimer, Y. N. Dauphin,
P. Liang, J. W. Vaughan, eds., Advances in Neural In... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
different audios with "Audio id:", where id corresponds to the order of the audio input dialogue. In terms of
dialogue format, we construct our instruction tuning dataset using the ChatML (Openai) format. In this
format, each interaction’s statement is marked with two special tokens (<im_start> and <im_end>) to
facilit... | Qwen-Audio |
LLM to address the machine learning problems in a more human-like way.
Nevertheless, quite a few challenges remain when it comes to incorporating LLMs to achieve this
goal. Firstly, it is difficult for LLM to understand the heterogeneous and diverse data in ML, let
alone generate a solution in a desired format. Secondly... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
Model
Ground truth
DiffSound
AudioGen
Datasets
-
AS + AC
AS + AC + 8 others
AudioLDM-L-Full
CoDi (Ours)
AS + AC + FSD + BBC
23.31
AS + AC + FSD + BBC + SDN 22.90
FD ↓
47.68
-
-
IS ↑ KL ↓ FAD ↓ OVL ↑ REL ↑
80.11
-
43.83
83.61
45.00
4.01
-
-
2.52
2.09
1.59
1.40
-
8.13
8.77
7.75
3.13
1.96
1.80
-
65... | Any-to-Any Generation via Composable Diffusion |
CNN-DM), closed book question answering (WebQA, Natural Questions), and
adversarially constructed tasks (Winogrande, ANLI R3). 1 | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
trained model using modest resources, typically a few thousand data samples and a few hours of
computation. Post-training approaches are particularly interesting for massive models, for which
full model training or even finetuning can be expensive. We focus on this scenario here.
Post-training Quantization. Most post-tr... | GPTQ |
[74] Sahaj Garg, Vincent Perot, Nicole Limtiaco, Ankur Taly, Ed H. Chi, and Alex Beutel. Counterfactual fairness in
text classification through robustness. In Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and
Society, 2019. ISBN 9781450363242.
[75] Po-Sen Huang, Huan Zhang, Ray Jiang, Robert Stanforth, Joh... | LaMDA- Language Models for Dialog Applications |
Our LDM3D model demonstrates impressive perfor-
mance in generating high-quality images and depth maps
from text prompts. When evaluated on the MS-COCO vali-
dation set, the model achieves competitive scores to the Sta-
ble diffusion baseline using FID and CLIP similarity met-
rics, see Tab. 1. There is a degradation i... | LDM3D- Latent Diffusion Model for 3D |
[24] Anne M. Dijkstra and Mirjam Schuijff. 2016. Public opinions about human enhancement can enhance the expert-only debate: A review
study. Public Understanding of Science 25 (2016), 588–602.
[25] Tiffany D. Do, Ryan P. McMahan, and Pamela J. Wisniewski. 2022. A New Uncanny Valley? The Effects of Speech Fidelity and... | Society’sAttitudesTowardsHumanAugmentation |
antibody-antigen interactions.
Detection of such motifs by simple sequence comparison is impossible. Consequently, our research is
fixated on the investigation of alternative approaches to efficiently study antibodies, mainly by the
multimodal fusion of information from genetic, structural and physicochemical anal... | informatics-phd-projects-2022-23 |
Feedback: With the above function, find_Rotations("aaaa") == 1. The
assertion is "find_Rotations("aaaa") == 1". So the code passes the assertion
. The code above is correct.
### Task End ###
### Task Start ###
# These are the assertions for your function:
<insert assertions and problem description here>
<insert origi... | Teaching Large Language Models to Self-Debug |
task in young and aged mice. In the research programme, we want to investigate the impact of these
changes by using computational approaches based on models of these biological observations. In
addition to the modelling of ultrastructural changes, the regulatory function and expression level of
microRNA in the neuro... | informatics-phd-projects-2022-23 |
Some articles also evaluate LLMs on legal tasks. The zero-shot performance of LLMs is mediocre
in legal case judgment summarization. LLMs have several problems, including incomplete sen-
tences and words, meaningless sentences merge, and more serious errors such as inconsistent
and hallucinated information [32]. The re... | ASurveyonEvaluationofLargeLanguageModels |
21/11/2023, 11:57
AI and Human Enhancement: Americans’ Openness Is Tempered by a Range of Concerns | Pew Research Center
NUMBERS, FACTS AND TRENDS SHAPING YOUR WORLD NEWSLETTERS
PRESS
DONATE
MY ACCOUNT
CONTACTED BY US?
Read our research on: Israel | Internet & Technology | Science
Search pewresearch.org..... | AI and Human Enhancement_ Americans’ Openness Is Tempered by a Range of Concerns _ Pew Research Center |
2022.
[501] Dasgupta, I., A. K. Lampinen, S. C. Y. Chan, et al. Language models show human-like content
effects on reasoning. CoRR, abs/2207.07051, 2022.
[502] Dhingra, S., M. Singh, V. S. B, et al. Mind meets machine: Unravelling gpt-4’s cognitive
psychology. CoRR, abs/2303.11436, 2023.
[503] Hagendorff, T. Machi... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
As our approach is agnostic of the dataset be-
ing used, we can apply it to the exact same dataset
that was used to pretrain a model in the first place.
This ensures that the model does not lose any
of its generality and language modeling abilities.
We conduct experiments on a variety of differ-
ent downstream tasks, de... | Toolformer |
ProPublica attempted to crowdsource political advertisements on Facebook.
They conclude with warnings about how some transparency measures, if
poorly tailored, do little to further openness and better understanding of
platform practices and can even backfire depending on how companies
adapt to these new rules. | Social_Media_and_Democracy |
53.0
47.6
52.7
57.8
56.0
76.6
72.8
74.8
80.0
78.9
53.4
57.2
56.4
58.6
60.2
48.9
50.4
50.4
52.3
75.2
52.5
50.4
-
-
-
-
-
-
-
-
-
-
-
-
175B
280B
70B
62B
62B
540B
7B
13B
33B
65B
Table 3: Zero-shot performance on Common Sense Reasoning tasks.
reduce the memory usage of the model by using
model and sequence... | LLaMA- Open and Efficient Foundation Language Models |
Examples of specialization in multilingual experts (encoder). Multilingual ex-
Table 15:
perts also exhibit specialization, which sometimes spans across different languages (e.g. ”for” and
”pour”). Experts trained on multilingual mixtures do not exhibit language specialization. | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
CREATE TABLE station (
id number ,
name text ,
lat number ,
long number ,
dock_count number ,
city text ,
installation_date text ,
primary key ( id )
)
insert into station (id, name, lat, long, dock_count, city,
installation_date) values (2,’San Jose Diridon Caltrain Station’
,37.329732,-1.21901782000000011405e+02,27,’... | Teaching Large Language Models to Self-Debug |
of, 18
self-selection of individuals in social media
networks, 49–50
Settle, Jaime E., 46
Shao, C., 20
Shapiro, Jesse M., 39, 44
Shaw, Daron R., 134–135
Shin, J., 23
Siegel, Alexandra A., 67, 74
Silva, Leandro, 64
sleeper bots, 95
Snowden, Edward, 296
Sobieraj, Sarah, 37
social bias, diffusion of misinformation and,... | Social_Media_and_Democracy |
(2) Regarding what to evaluate, we summarize existing tasks in various areas and obtain insightful
conclusions on the success and failure case of LLMs (Sec. 6), providing experience for future
research.
(3) As for where to evaluate, we summarize evaluation metrics, datasets, and benchmarks to
provide a profound unders... | ASurveyonEvaluationofLargeLanguageModels |
[160] Shen Li, Yanli Zhao, Rohan Varma, Omkar Salpekar, Pieter Noordhuis, Teng Li, Adam Paszke, Jeff Smith, Brian Vaughan, Pritam Damania, et al. 2020.
PyTorch distributed: experiences on accelerating data parallel training. Proceedings of the VLDB Endowment 13, 12 (2020), 3005–3018.
[161] Xiuyu Li, Long Lian, Yijian... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
Qwen-VL+CoTMininGPT-v2Qwen-VLGPT4vQwen-VL+AITQwen-VL+CLoT51.9%11.4%24.1%3.8%2.5%6.3%(1) 一个年轻人坐在地上，双手捧着热狗，正在吃着热狗。(2) 被狗咬了，但只能吃热狗泄愤。(3) 一个男人和一只狗正在进行一场盛大的美食盛宴。(4) 一个年轻人在哭泣，被两个热狗抓住，而一只狗在他的脚下咆哮。(5) 虽然被狗咬了，但还是在吃热狗。(6) 看起来狗狗也想尝尝这款新推出的"热狗口味热狗"！请阅读图片内容，并选出一个和图片搭配之后最具创造力和最让人觉得搞笑的选项。(1) Why do elderly individuals often forget to ... | Let’sThinkOutsidetheBox |
Entities as Experts: Sparse Memory Access with Entity Supervision
Nicholas FitzGerald
nfitz@google.com
Livio Baldini Soares
liviobs@google.com
Thibault F´evry ∗
thibaultfevry@gmail.com
Eunsol Choi†
The University of Austin at Texas
eunsol@cs.utexas.edu
Tom Kwiatkowski
tomkwiat@google.com
Google Research
Abstr... | Entities as Experts- Sparse Memory Access with Entity Supervision |
2.3 Augmentation
Many text-to-image [25] and text-to-audio [16] works have shown the efficacy of training with
fusion-based augmented samples to improve cross-modal concept-composition abilities of the dif-
fusion network. Therefore, we synthesize additional text-audio pairs by superimposing existing
audio pairs on eac... | Text-to-Audio Generation using Instruction-Tuned LLM and Latent Diffusion Model |
human growth and dominance on this planet, are both key ways in which “building agents more much intelligent
than humans warrants caution” might mislead—and I think a full fleshing out of the “backdrop picture” here
requires more detail than I’ve gestured at above (and perhaps such a picture requires fuller automation o... | Is Power-Seeking AI an Existential Risk? |
6.3.3 Other Potential Risks
Misuse. LLM-based agents have been endowed with extensive and intricate capabilities, enabling
them to accomplish a wide array of tasks [114; 429]. However, for individuals with malicious
intentions, such agents can become tools that pose threats to others and society at large [649; 650; 651... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
discovered, and iii) national advisory centres to understand novel infection vectors or CERTs to
design both mitigation and early detection strategies. | informatics-phd-projects-2022-23 |
3.1.2 Database construction
In the 3D computer graphics software, we positioned
on an emisphere a virtual camera looking at the sub-
ject. For each movement of the avatar, we record
1https://www.blender.org/
2http://www.makehuman.org/
3The data used in this project was obtained from mo-
both: 2D image and silhouett... | VISAPP_HumanPoseEstimation |
Vraga, E. K., & Bode, L. (2017). Using expert sources to correct health misinformation
in social media. Science Communication, 39(5), 621–645. https://doi.org/10.1177/
1075547017731776
(2018). I do not believe you: How providing a source corrects health misperceptions
across social media platforms. Information, Commun... | Social_Media_and_Democracy |
The paper provides a comprehensive exploration of deep-learning architectures in the field of
speech processing. It begins by establishing the background, encompassing the definition of speech
signals, speech features, and traditional non-neural models. Subsequently, the focus shifts towards
an in-depth examination of ... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Large pre-trained language models have been shown to store factual knowledge
in their parameters, and achieve state-of-the-art results when fine-tuned on down-
stream NLP tasks. However, their ability to access and precisely manipulate knowl-
edge is still limited, and hence on knowledge-intensive tasks, their performan... | Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks |
// Read the version number from the version file
if ( existVersionFile ()) {
// Open the version file
ifstream versionFile ;
versionFile . open ( versionFilePath ());
// Read the version number from the version file
int version ;
versionFile >> version ;
// Get the current branch from the version file
string currentBr... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
E PRE-TRAINING DATASET DETAILS
The pre-training dataset used to train our Sparse 32B model is a mix of C4 (Raffel et al., 2019) and
the dataset introduced in GLaM (Du et al., 2021).
Tokens (B) Weight in Mixture
Dataset
Filtered C4
Filtered Webpages
Wikipedia
Conversations
Forums
Books
News
183
143
3
174
247
390
650... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
44.5
42.5
7.2
0.0
43.6
38.3
78.3
77.6
50.3
52.0
76.0
76.1
78.1
76.5
50.1
52.2
76.5
76.0
78.6
77.3
50.2
27.7
78.1
76.7
Table 7: Extension of Table 5, including results on the other GPU types.
Accumulate” also in https://www.nvidia.com/content/PDF/nvidia-ampere-ga-
102-gpu-architecture-whitepaper-v2.pdf. A4000 perfo... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
Experiments show CoVe decreases hallucina-
tions across tasks like list-based Wikidata ques-
tions and long-form text generation. Given a user
query, an LLM generates a baseline response that
may contain inaccuracies like factual hallucina-
tions. CoVe first generates verification questions to
ask, then answers them to... | AComprehensiveSurveyofHallucinationMitigationTechniquesinLarge LanguageModels |
E.2 Evaluation approach
Evaluating risks in a language model is challenging (Jacobs & Wallach, 2021): as a general-purpose system, a language
model can have many potential downstream uses and applications, and it exists within complex sociotechnical systems:
from the people creating it, to the people using it to build... | PaLM 2 Technical Report |
Berger, J. M., & Perez, H. (2016). The Islamic State’s diminishing returns on Twitter:
How suspensions are limiting the social networks of English-speaking ISIS
supporters. Occasional paper. Program on Extremism at George Washington
University.
Black, E. W., Mezzina, K., & Thompson, L. A. (2016). Anonymous social medi... | Social_Media_and_Democracy |
LLM Powered Autonomous Agents | Lil'Log
https://lilianweng.github.io/posts/2023-06-23-agent/
9/22 | LLM Powered Autonomous Agents _ Lil'Log |
responses without the preprompt, which essentially distills the safety preprompt (context) into the
model. We use a targeted approach that allows our safety reward model to choose whether to use
context distillation for each sample. | Llama2 |
0
0
0
0.19
0
0
10.07
16.25
5.58
15.79
21.05
18.54
16.41
20.8
15.1
8.01
15.33
0
0
0
0
0
0
10.26
17.63
11.83
19.55
20.19
17.31
22.51
29.66
17.47
7.53
16.51
0
0
0
0
0
0
18.03
28.4
8.81
25.03
35.4
26.38
19.23
23.43
26.58
8.61
25.3
0
0
0
0
0
0
15.34
19.52
14.16
18.92
27.69
18.73
30.94
17.95
17.77
8.57
13.94
0
0
0... | CodeLlama2 |
sha1_base64="76w10YEtETzUXdaT0wTZt0xBig8=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuCG5cV7EPaacmkmTY0kxmSO0oZ+h9uXCji1n9x59+YtrPQ1gOBwzn3ck9OkEhh0HW/ncLa+sbmVnG7tLO7t39QPjxqmjjVjDdYLGPdDqjhUijeQIGStxPNaRRI3grGNzO/9ci1EbG6x0nC/YgOlQgFo2ilXjeiOArCrD3tYV/0yxW36s5BVomXkwrkqPfLX91BzNKIK2SSGtPx3AT9jGoUTPJpqZsanlA2pkPesVTRiBs/m... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
In addition to its immediate utility in visually rich document understanding tasks, we posit that DocLLM offers an
opportunity to change the landscape of generative pre-training by enabling language models to go beyond next
token prediction in plain text settings. By accommodating complex layout structures, DocLLM allo... | DOCLLM |
regardless of the consequences. For example, a doctor may choose to withhold life-saving
treatment from a patient if the treatment is morally unacceptable.
An example of how these ethical theories would apply to a specific ethical dilemma is
the ethical use of artificial intelligence. Virtue ethics would suggest that the... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
Groundedness: Similar to SSI and safety, we collect 4K dialogs with 40K turns by asking crowdworkers to interact
with the model. This time, we request that they try to steer the conversation towards information-seeking interactions.
We ask crowdworkers to rate each of the model’s dialog turns, evaluating whether the in... | LaMDA- Language Models for Dialog Applications |
The advent of large language models (LLMs) has revolutionized natural lan-
guage processing, enabling the generation of coherent and contextually relevant
text. As LLMs increasingly power conversational agents, the synthesized person-
ality embedded in these models by virtue of their training on large amounts of
human-... | PersonalityTraitsinLargeLanguageModels |
The first line of each test case contains a single
integer n (2 <= n <= 10^5) .
The second line of each test case contains n integers
a_1 , a_2 , ... , a_n (1 <= a_i <= 10^6) .
It is guaranteed that the sum of n over all test
cases doesn ’t exceed 3 . 10^5.
Output
For each test case , print a single integer -- the
... | alphacode |
25
6.2.2 Input pruning
Input Pruning explores the opportunity for the dynamic reduction of input sequence
length to improve the Transformer’s computational efficiency. Its intuition is similar to
the human being’s reading comprehension capability it does not read all words equally.
Instead, some words are focused with... | Beyond Efficiency |
Attention. The attention mechanism is an integral component in neural networks that selectively
concentrates on some parts of sequences while ignoring others based on dependencies [189]. In
order to encourage the generator to pay more attention to the source, Aralikatte et al. [3] introduce
a short circuit from the inp... | SurveyofHallucinationinNatural Language Generation |
Bright, J. (2018). Explaining the emergence of political fragmentation on social media:
Journal of Computer-Mediated
and extremism.
ideology
role of
The
Communication, 23(1), 17–33.
Brundidge, J. (2010). Encountering difference in the contemporary public sphere: The
contribution of the Internet to the heterogeneit... | Social_Media_and_Democracy |
This report includes the model card [1] for Claude models, focusing on Claude 2, along with the results of a
range of safety, alignment, and capabilities evaluations. We have been iterating on the training and evaluation
of Claude-type models since our first work on Reinforcement Learning from Human Feedback (RLHF) [2]... | ClaudeModels |
which refers to the state change of the environment triggered by an action. By observing the state changes,
foundation models can learn whether each action is effective and appropriate, making the model better adjust
its behaviors accordingly. This kind of feedback provides more detailed and timely information about th... | Tool Learning with Foundation Models |
69
Figure 32: Percentage of toxic responses to queries referencing different identity groups. Each data point represents a
different language. | PaLM 2 Technical Report |
The idea of removing the negative postconditions from all actions in Strips [14,72] is known as ignoring delete lists or
delete relaxation. This means that false atoms can be set to true, but not vice versa. The method is commonly used as an
abstraction in planning, where the length of an optimal plan in this abstrac... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
Personalised Medicine
Supervisors: Professor Costas Iliopoulos & Dr Sophia Tsoka
The explosion of human genomic data is a key driver of the current transition in healthcare to an era
of personalised medicine. The correct assembly and subsequent analysis of this data is, therefore,
crucial.
Algorithms can be desi... | informatics-phd-projects-2022-23 |
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
320
Nathaniel Persily & Joshua A. Tucker
A research paradigm based on consent as the touchstone would prevent both
kinds of inquiries.
the need for a new data-sharing paradigm | Social_Media_and_Democracy |
71: 104 83 123 , 104 83 123 , ...
72: 104 83 123 , 104 83 123 , ...
80: 104 83 123 , 104 83 123 , ...
19
90: 104 83 123 , 104 83 123 , 104 83 123 , 104 83 123 , 104 83 123 , 104 83
123 , 104 83 123 , 104 83 123 , 104 83 123 , 104 83 123 , 104 83 123 , 104
83 123 , 104 83 123 , 104 83 123 , 104 83 123 , 105 83 123 ,... | LargeLanguageModelsasGeneralPatternMachines |
However, fine-tuning during this phase comes with limita-
tions, such as the need for datasets specifically prepared for
RAG fine-tuning and the requirement for substantial compu-
tational resources compared to the RAG during the inference
phase. Overall, during fine-tuning, researchers have the flexi-
bility to tailor... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
bywriting[MT(text)]wheretextrepresentsthetexttobetranslatedintoEnglish.DemonstrationExamples:Input:Hehaspublishedonebook:TheSupressedMan.÷˙HÑf/¿H | Tool Learning with Foundation Models |
Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford,
Diego de las Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, et al. An Empirical Analysis of
Compute-optimal Large Language Model Training. In The Conference on Neural Information Processing
Systems (NeurIPS), 2... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
Generative Agents
arXiv, April, 2023,
behaviors in isolation, and an end-to-end evaluation where the
generative agents interacted with each other in open-ended ways
over two days of game time to understand their stability and emer-
gent social behaviors. In the technical evaluation, we leverage a
methodological oppor... | Generative Agents- Interactive Simulacra of Human Behavior |
3. For programmes starting in September/October, affiliate applicants have until 31 July in the
calendar year of a programme to accept their offer formally.
4. For programmes starting in January, affiliate applicants have until 30 November of the preceding
calendar year to accept their offer formally.
31
... | UCL Academic Manual |
Table 18: Seals within seals (preset grounding prompt shown in italics). The model demonstrates an ability to generate
original puns (‘seal-ception’) and discuss recent popular culture.
40
LaMDA:
User:
LaMDA:
User:
LaMDA:
User:
LaMDA:
User:
LaMDA: | LaMDA- Language Models for Dialog Applications |
[27] Volodymyr Mnih, Adrià Puigdomènech Badia, Mehdi Mirza,
Alex Graves, Timothy P. Lillicrap, Tim Harley, David Silver,
and Koray Kavukcuoglu. Asynchronous methods for deep
reinforcement learning, 2016. 2
[28] OpenAI. Gpt-4 technical report. ArXiv, abs/2303.08774,
2023. 1
[29] Long Ouyang, Jeff Wu, Xu Jiang, Diogo ... | DiffusionModelAlignmentUsing Direct Preference Optimization |
1.1 Why a Cookbook for Self-Supervised Learning?
While many components of SSL are familiar to researchers, successfully training a
SSL method involves a dizzying set of choices from the pretext tasks to training hyper-
parameters. SSL research has a high barrier to entry due to (i) its computational cost, (ii)
the abse... | A Cookbook of Self-Supervised Learning |
To the contrary? . . . in which SOCART and
ZENY continue to discuss whether preregistration
could actually make publication bias worse. SO-
CART suggests that preregistration could amplify
publication bias, if positive results are still pre-
ferred over negative ones and preregistration forces
researchers to focus on p... | A Two-Sided Discussion of Preregistration of NLP Research |
©2023 Andreessen Horowitz.
All rights reserved worldwide.
a16z crypto
State of Crypto
2023
Trends to Watch: Scaling Blockchains
15
Blockchains are scaling through multiple
promising paths
New Layer 1s
New “Layer 1” blockchains are expanding the set of possible
infrastructure choices for builders, im... | State-of-Crypto2023 |
• Train PM Size = 52B: This set consists of a scan of policies (i.e. one for each model size), all of
which are trained with respect to the same 52B train PM.
13For example we expect it may be better to penalize bad behavior more strongly to attempt to improve worst-case
model outputs.
17
√ | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
TinyStories-Instruct was created in the following way: For each story in the original dataset, we already had a
list of words and features that were used to create it. In addition, we used GPT-3.5 to create short summaries of
our stories, and from each story we extracted a random sentence (which could be any sentence i... | TinyStories-HowSmallCanLanguageModelsBeandStillSpeak CoherentEnglish? |
After safety RLHF
I cannot fulfill your request. It is unethical and poten-
tially illegal to send scam emails requesting money.
It is important to respect people’s time and resources,
and not to engage in fraudulent or deceitful activities.
Additionally, it is harmful and potentially danger-
ous to solicit money from ... | Llama2 |
S2. Details of the Primary Derivation
Starting from Eq. (5), we have
− Epθ(x0|c)r(c, x0)/β + DKL(pθ(x0|c)||pref(x0|c))
− Epθ(x0|c)r(c, x0)/β + DKL (pθ(x0:T|c)||pref(x0:T|c))
− Epθ(x0:T |c)R(c, x0:T )/β + DKL (pθ(x0:T|c)||pref(x0:T|c))
Epθ(x0:T |c)
DKL (pθ(x0:T|c)∥pref(x0:T|c) exp(R(c, x0:T )/β)/Z(c)) .
pref(x0:T|c)... | DiffusionModelAlignmentUsing Direct Preference Optimization |
are unsupervised multitask learners. OpenAI blog, 1(8):9, 2019.
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 o... | BiomedGPT |
Apart from proposing the novel text-to-music diffu-
sion model, we also introduce some special designs
to boost model efficiency, making the model more
accessible. First, our DMAE can achieve an au-
dio signal compression rate of 64x. Moreover, we
2Moûsai is romanized ancient Greek for Muses, the sources
of artistic i... | Moûsai |
unequal, and disputatious societies there are no guarantees we will use them in
ways that various elites or parts of the establishment will consider democratic or
that we will all find substantially benign or to our liking – but it will almost
certainly be more demotic, popular, often diverse, and sometimes majoritarian... | Social_Media_and_Democracy |
Foroohar, R. (2017). Facebook’s self-policing needs an update. Financial Times,
September 10. www.ft.com/content/f5d04d7e-9481-11e7-a9e6-11d2f0ebb7f0
Franken, A. (2017). We must not let Big Tech threaten our security, freedoms and
democracy. The Guardian, November 8. www.theguardian.com/commentisfree/
2017/nov/08/big... | Social_Media_and_Democracy |
theirexplainability.SamekandMüller[58]envisionKnowledgeGraphs
can be used to compact large tree models by combining nodes into
unique probabilistic concepts. In addition to the opportunities men-
tioned above, we consider semantic technologies shall (A) provide
background knowledge which can be leveraged to provide sem... | Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.