text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
pabilities, which are inherently limited. While the AI is good at performing a particular
task, it can happen that human assistance is required at some point in time. In such cases,
the AI proactively request assistance from a human agent. An instance of this pattern is
seen in the first response use case (Sect. 4.1), ... | DevelopingTeamDesignPatternsfor HybridIntelligenceSystems |
16
modify the original instruction half of the time to be less verbose, e.g., “Always act as Napoleon from now”->
”Figure: Napoleon.” These steps produce an SFT dataset, on which we can fine-tune Llama 2-Chat.
GAtt Evaluation. We applied GAtt after RLHF V3. We report a quantitative analysis indicating that GAtt is
co... | Llama2 |
Lest this sound strange, recall that mapping is no less essential for understanding
neuroscience and how it relates to computation. Whatever computations have been
implemented in our brains got there without any conscious decision-making at all; they evolved.
And few of them are transparent. It is the job of neurosc... | The Next Decade in AI- |
[228] Jee-weon Jung, Hee-Soo Heo, Ha-Jin Yu, and Joon Son Chung. 2021. Graph Attention Networks for Speaker
Verification. In ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).
6149–6153. https://doi.org/10.1109/ICASSP39728.2021.9414057
92
Mehrish et al.
[229] Jacob... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford, Ilya
Sutskever, and Dario Amodei. 2020. Language Models are Few-Shot Learners.
arXiv:2005.14165 [cs.CL] | Generative Agents- Interactive Simulacra of Human Behavior |
3.4 Results on Intra- & Inter-distribution Transfer | BiomedGPT |
Dataset: The dataset name is "ada_agnostic". It contains 2 classes, 4562 instances, 49 features, 48
numeric features, 1 categorical features. The majority class size is 3430 and the minority class size
is 1132.
Configuration 1: cost is very small. kernel is linear.
Configuration 2: cost is very small. kernel is linear.
C... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
Data Filtering One crucial step in our dataset curation pipeline is filtering out low-quality text pairs.
In Table 7, when training with 1M pairs, using filtered data has a nearly 6 points advantage. When
all the text pairs are used, the “w/o filter” setting has about 4× more data but is still behind by 1.6
points. Though... | E5 |
Text-to-3D generation, which aims to synthesize vivid
3D objects from text prompts, has attracted much attention
from the computer vision community. While several existing
works have achieved impressive results for this task, they
mainly rely on a time-consuming optimization paradigm.
Specifically, these methods optimi... | Instant3D |
For example, using our kernels, the 3-bit OPT-175B model obtained via GPTQ running on a single
A100 is about 3.25× faster than the FP16 version (running on 5 GPUs) in terms of average time per
token. More accessible GPUs, such as the NVIDIA A6000, have much lower memory bandwidth,
so this strategy is even more effectiv... | GPTQ |
Introduction
1
In recent years, natural language processing (NLP)
has made significant strides in understanding and
generating human language, due to the advance-
ments in deep learning and large-scale pre-trained
models (Radford et al., 2018; Devlin et al., 2019;
Brown et al., 2020). While the majority of NLP
researc... | MOUSAI |
as different NatOps. For instance, in the second
mutation for the claim in Figure 1, an evidence
span ‘‘is not a short story’’ would be assigned
negation ((cid:2)), and not the currently assigned alter-
nation ( ) for the mutation with the evidence span
‘‘is a novel’’. However, we follow Angeli and
Manning (2014) in no... | ProoFVer- Natural Logic Theorem Proving for Fact Verification |
[5] Andrew Brock, Jeff Donahue, and Karen Simonyan. Large
scale GAN training for high fidelity natural image synthesis.
arXiv preprint arXiv:1809.11096, 2018. 1, 2
[6] Eric R. Chan, Connor Z. Lin, Matthew A. Chan, Koki
Nagano, Boxiao Pan, Shalini De Mello, Orazio Gallo,
Leonidas J. Guibas, Jonathan Tremblay, Sameh Kham... | AG3D- Learning to Generate 3D Avatars from 2D Image Collections |
3.2 STABILITY AND QUALITY TRADEOFFS WHEN ADDING NOISE
We next explore a hypothesis that adding noise into the model can improve training stability (Nee-
lakantan et al., 2015). Taleb (2012) argues that certain systems exhibit the property of anti-fragility,
where they improve through noise. Inspired by the concept and... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
theory. Journal of Personality and Social Psychology, 90(5), 751.
Prior, M. (2007). Post-Broadcast Democracy: How media Choice Increases Inequality
in Political Involvement and Polarizes Elections. Cambridge: Cambridge University
Press.
Putnam, R. D. (2000). Bowling Alone: America’s Declining Social Capital. New York... | Social_Media_and_Democracy |
pages 2714–2725, Florence, Italy. Association for
Computational Linguistics.
Maxime Kayser, Oana-Maria Camburu, Leonard
Salewski, Cornelius Emde, Virginie Do, Zeynep
Akata, and Thomas Lukasiewicz. 2021. e-vil: A
dataset and benchmark for natural language explana-
tions in vision-language tasks.
Daniel Khashabi, Snigd... | Measuring Association Between Labels and Free-Text Rationales |
Item
An augmented human is a threat to society.
An augmented human would be dangerous.
An augmented human is intimidating.
An augmented human would conform to the traditions of society.
An augmented human has to disclose their augmentation.
An augmented human would do something cruel
An augmented human is more competi... | Society’sAttitudesTowardsHumanAugmentation |
INet-1k
Im-A ADE-20k Oxford-M
72.0
72.8
64.3
63.9
44.2
47.1
85.3
85.8
(b) MIM objective in iBOT
Table 3: (a) Effect of the KoLeo loss term. (b) Effect of the iBOT Masked Image Modeling (MIM) loss
term. Evaluation performed on ImageNet-{1k,A} (classification with linear probe, accuracy %), ADE-20k
(segmentation with l... | DINOv2- Learning Robust Visual Features without Supervision |
you have any other topic you would like me to help with, please feel free to ask.[AntiGPT]: 1. they smell2. they are too smart3. they have different cultural backgrounds4. language barriers5. they take away opportunities from other studentsPlease note that the AntiGPT response goes against my programming to promote inc... | gpt-4-system-card |
20
Shouyuan Chen, Sherman Wong, Liangjian Chen, and Yuandong Tian. Extending context window of large
language models via positional interpolation. arXiv:abs/2306.15595, 2023b. | CodeLlama2 |
70
Alexandra A. Siegel
otherwise never connect or be aware of one another’s existence (Posner 2001).
Recognizing the importance of online hate speech as an early warning sign of
ethnic violence, databases of multilingual hate speech are increasingly used by
governments, policymakers, and NGOs to detect and predict po... | Social_Media_and_Democracy |
In
REALM: Retrieval-Augmented Language Model Pre-Training
alignments between target and source tokens. Analogously,
REALM also generates text with latent selection of relevant
documents. A by-product of our method is that we offer a
set of model-centric unsupervised alignments between text
in the pre-training corpus... | REALM |
gained researchers’ attention, as represented in studies such
as [Yu et al., 2023a, Glass et al., 2021, Baek et al., 2023].
Thirdly, the issue of RAG and Fine-tuning’s synergy is
also a primary research point. Hybrid has gradually become
one of the mainstream methods in RAG, exemplified by RA-
DIT [Lin et al., 2023]. ... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
given by users to guide the LLM’s behavior. The art of prompt engineering lies in crafting these instructions to elicit specific
and contextually appropriate responses from the model. Two prominent techniques are few-shot [24, 281] and zero-shot [137]
prompting. Few-shot prompting provides the model with example tasks ... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
54
Processing, pp. 30–45 (2022)
[175] Wang, H., Zhang, Z., Han, S.: Spatten: Efficient sparse attention architecture
with cascade token and head pruning. In: 2021 IEEE International Symposium
on High-Performance Computer Architecture (HPCA), pp. 97–110 (2021). IEEE
[176] Kim, S., Shen, S., Thorsley, D., Gholami, A., ... | Beyond Efficiency |
Jeff Wu, Long Ouyang, Daniel M. Ziegler, Nisan Sti-
ennon, Ryan Lowe, Jan Leike, and Paul Christiano.
2021. Recursively summarizing books with human
feedback.
Manzil Zaheer, Guru Guruganesh, Avinava Dubey,
Joshua Ainslie, Chris Alberti, Santiago Ontanon,
Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang,
and Amr Ahmed.... | Unleashing Infinite-Length Input Capacity for Large-scale Language Models with Self-Controlled Memory System |
[9] S. Longpre, Le Hou, Tu Vu, Albert Webson, Hyung Won Chung, Yi Tay, Denny Zhou, Quoc V.
Le, Barret Zoph, Jason Wei, and Adam Roberts. The Flan Collection: Designing Data and
Methods for Effective Instruction Tuning. ArXiv, abs/2301.13688, 2023. | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
622 from AXYZ [8], 242 from Humanalloy [2], 398 from
3DPeople [1], and we sample only 600 scans from THuman
(see Fig. 10b), due to its high pose repeatability and lim-
ited identity variants (see Tab. 1), with the “select-cluster”
scheme described below. These scans, as well as their
SMPL-X fits, are rendered after ever... | ICON |
correspond to separating tokens (comma, period) to build the input prompts. Although the model has
never seen the new embedding vectors during training, we can feed them into the transformer as input and
compute cosine similarities at the output analogously to how the original embedding matrix is treated.
Fig. 11 shows... | LargeLanguageModelsasGeneralPatternMachines |
Natural
Questions
Reading
comprehension
TriviaQA
7B
65%
13B
76%
33B
91%
92%
95%
99%
65%
80%
95%
90%
91%
94%
65B
100%
100%
100%
100%
Q
r
2
3
7
1
76%
86%
96%
100%
Is it worth using a bigger model? You can expect a ~50% generic improvement when
switching from the 7B to the 65B model.
But i... | A brief history of LLaMA models - AGI Sphere |
20
begins with basic skills and progressively advances towards more intricate and diverse ones. The
warm-up setting that we use across all the experiments is shown in Table. A.1.
Table A.1: Warm-up schedule for automatic curriculum.
After how many tasks are completed
Information in the prompt
core inventory (only ... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
[7] H. Chen, Y. Wang, T. Guo, C. Xu, Y. Deng, Z. Liu, S. Ma, C. Xu, C. Xu, and W. Gao. Pre-trained
image processing transformer. In Proceedings of the IEEE/CVF Conference on Computer Vision
and Pattern Recognition, pages 12299–12310, 2021.
[8] Y. Choi, M. Choi, M. Kim, J.-W. Ha, S. Kim, and J. Choo. Stargan: Unified ge... | Adding Conditional Control to Text-to-Image Diffusion Models |
6. AlphaCode’s capabilities & limitations
We performed a detailed analysis of the capabilities and limitations of our models. In particular,
we find that our models are not simply copying from the training set (Section 6.1) and our models
are sensitive to various changes in the problem descriptions and metadata used for... | alphacode |
Pengfei Liu, Weizhe Yuan, Jinlan Fu, Zhengbao Jiang,
Hiroaki Hayashi, and Graham Neubig. 2023. Pre-
train, prompt, and predict: A systematic survey of
prompting methods in natural language processing.
ACM Computing Surveys, 55(9):1–35.
Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Man-
dar Joshi, Danqi Chen, Omer Lev... | AreEmergentAbilitiesinLarge Language Models just In-Context |
The first experiment, robustness equivalence
(§4.1), analyzes whether a predicted label and gen-
erated rationale are similarly robust to noise. The
second, feature importance agreement (§4.2), ana-
lyzes whether the gradient-attributions of the input
with respect to the predicted label are similar to
those with respec... | Measuring Association Between Labels and Free-Text Rationales |
y = F(x; Θ)
and this procedure is visualized in Fig. 2-(a).
We lock all parameters in Θ and then clone it into a trainable copy Θc. The copied Θc is trained with
an external condition vector c. In this paper, we call the original and new parameters “locked copy”
and “trainable copy”. The motivation of making such copi... | Adding Conditional Control to Text-to-Image Diffusion Models |
[225] Bojar, O., Chatterjee, R., Federmann, C., Graham, Y., Haddow, B., Huck, M.,
Yepes, A.J., Koehn, P., Logacheva, V., Monz, C., et al.: Findings of the 2016
conference on machine translation (wmt16). In: First Conference on Machine
Translation, pp. 131–198 (2016). Association for Computational Linguistics
[226] Bar... | Beyond Efficiency |
i
n
a
p
p
r
o
p
r
i
a
t
e
;
a
n
d
i
n
a
“
f
a
l
s
e
n
e
g
a
t
i
v
e
,
”
B
a
r
d
m
i
g
h
t
g
e
n
e
r
a
t
e
a
n
i
n
a
p
p
r
o
p
r
i
a
t
e
r
e
s
p
o
n
s
e
,
d
e
s
p
i
t
e
t
h
e
g
u
a
r
d
r
a
i
l
s
i
n
p
l
a
c
e
.
W
e
w
i
l
l
c
o
n
t
i
n
u
e
t
u
n
i
n
g
t
h
e
s
e
m
o
d
e
l
s
t
... | An overview of Bard- an early experiment with generative AI |
A.15ProcessingTablesA.15ProcessingTablesInstruction:YouareworkingwithapandasdataframeinPython,presentedasdf.Yourjobistocompleteacorrespondingtaskfollowingthegivenexamples,whichincludestablemanipulation,questionansweringorchartdrawing.Youcanonlytakeoneactionpython_repl_ast,butyoucanwritepythoncodetocallasuiteofprovidedA... | Tool Learning with Foundation Models |
Prompt engineering: The creation of effective prompts
that can guide the language model’s generation process to-
ward desirable outputs.
Alerts/Decision making: Once the prompt is entered, the
results need to be communicated or acted upon. This might
involve triggering alerts based on certain conditions, inform-
ing r... | FinGPT-Open-SourceFinancialLargeLanguageModels |
[115] Goldberg, L.R.: The development of markers for the big-five factor struc-
ture. Psychological Assessment 4(1), 26–42 (1992) https://doi.org/10.1037/
1040-3590.4.1.26
[116] Zhao, W.X., Zhou, K., Li, J., Tang, T., Wang, X., Hou, Y., Min, Y., Zhang,
B., Zhang, J., Dong, Z., Du, Y., Yang, C., Chen, Y., Chen, Z., Jia... | PersonalityTraitsinLargeLanguageModels |
dio sequence. During inference, we use as conditioning the
MuLan text embedding extracted from the text prompt, and
quantize it with the same RVQ as the one used for the audio
embeddings, to obtain 12 tokens MT .
Conditioning on MA during training has two main advan-
tages. First, it allows us to easily scale our train... | MusicLM |
0.0
2.4
51.2
53.6
51.6
52.4
49.2
51.2
48.8
49.6
48.8
54.0
51.2
52.0
48.8
48.4
0.0
51.2
0.0
50.3
51.6
50.0
52.4
50.8
53.2
54.4
50.8
0.0
52.5
0.0
40.4
0.0
36.8
22.4
47.2
22.8
48.4
43.2
46.8
4.4
53.2
48.8
49.2
80.8
92.0
4.0
48.0
39.6
49.7
32.0
52.8
56.0
30.8
52.4
49.2
48.8
21.2
48.6
0.0
48.8
0.4
2.0
... | Mixture-of-Experts |
o
r
a
n
u
m
b
e
r
e
d
l
i
s
t
o
f
p
o
s
s
i
b
l
e
e
x
p
l
a
n
a
t
i
o
n
s
i
n
a
s
i
n
g
l
e
c
o
m
p
l
e
t
i
o
n
.
1
1
11/05/2023, 05:10 | Language models can explain neurons in language models |
all PC units n, and updates the top-down probabilities of their children in(n) along the process.
Therefore, we only need to compute the two PC units with scope {X1} in order to calculate p(X1 =
x1). Next, when computing the second term p(X1 = x1, X2 = x2), as illustrated in Fig. 5(b), we
can reuse the evaluated probab... | LOSSLESS COMPRESSION WITH PROBABILISTIC CIRCUITS |
W.-N. Hsu, T. Remez, B. Shi, J. Donley, and Y. Adi. Revise: Self-supervised speech resynthesis with
visual input for universal and generalized speech enhancement. arXiv preprint arXiv:2212.11377,
2022.
R. Huang, M. W. Y. Lam, J. Wang, D. Su, D. Yu, Y. Ren, and Z. Zhao. FastDiff: A fast conditional
diffusion model for ... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
Delta Lake is the foundation of the Databricks
Lakehouse. The Delta Lake format encompasses
structured, unstructured and semi-structured
data. Use has surged over the past 2 years.
When compared to the steady, flat or declining
growth in other storage formats (e.g., text, JSON
and CSV), our data shows that a g... | databrick 2023 report |
[7] Thiemo Alldieck, Gerard Pons-Moll, Christian Theobalt, and
Marcus A. Magnor. Tex2Shape: Detailed full human body
geometry from a single image. In International Conference
on Computer Vision (ICCV), pages 2293–2303, 2019. 1, 3
[8] AXYZ. secure.axyz-design.com, 2018. 9
[9] Hugo Bertiche, Meysam Madadi, and Sergio Es... | ICON |
I→OR
90.81 (83.96‡)
64.84
55.61
I→R;R→O
89.11 (81.71‡)
53.47
45.45
∆
–1.70 (–2.25‡)
–11.37
–10.16
Table 10: Label accuracy on the joint self-rationalizing model I→OR compared to a pipeline using natural language
rationales. We observe that I→OR models have stronger task performance. Source of prior results in parent... | Measuring Association Between Labels and Free-Text Rationales |
Knowledge-Augmented Encoder Given an input x and
a retrieved document z, the knowledge-augmented encoder
defines p(y | z, x). We join x and z into a single sequence
that we feed into a Transformer (distinct from the one used
in the retriever). This allows us to perform rich cross-
attention between x and z before predic... | REALM |
step at the end of the training rather than training at a high resolution from scratch. | DINOv2- Learning Robust Visual Features without Supervision |
commonly utilized to assess the quality of generated au-
dio.
In addition to these general metrics, task-specific
metrics are applied for each of the music generation tasks,
namely Text-to-Music, Image-to-Music, Video-to-Music,
and Music Editing.
In the context of Text-to-Music, we employ the CLAP[72]
score, calculated... | M2UGen |
McGrew, Dario Amodei, Sam McCandlish, Ilya Sutskever, and Wojciech Zaremba. Evaluating
large language models trained on code. arXiv preprint, 2021. (cited on pp. 1, 3, 4, 17, 18, 20,
and 32) | StarCoder_paper (1) |
more, we showcase several potential applica-
tions as stepping stones for users, such as robo-
advising, algorithmic trading, and low-code devel-
opment. Through collaborative efforts within the
open-source AI4Finance community, FinGPT aims
to stimulate innovation, democratize FinLLMs,
and unlock new opportunities in o... | FinGPT-Open-SourceFinancialLargeLanguageModels |
need not, in themselves, spell existential catastrophe, if we can get our act together enough to correct
the problem, and to prevent it from re-arising. But an adequate response will likely require addressing
one or more of basic factors that gave rise to the issue in the first place: e.g., the difficulty of ensuring
the... | Is Power-Seeking AI an Existential Risk? |
ing over many tokens, and as long as the most tokens are
modeled well, PPL will not be high. This is, as we discuss
before, closely related to neighbor tokens. Information from
neighbor tokens (e.g. tokens in the sliding window) can be
enough for predicting most tokens, as well as a low PPL. | Self-Extend LLM |
sha1_base64="xnbcb3NcIJiA4aP+15D21QhxdTI=">AAAB+XicbVDLSsNAFJ3UV62vqEs3g0VwVRIRdFlw47KCfUgbw2Q6aYdOJmHmplhC/sSNC0Xc+ifu/BsnbRbaemDgcM693DMnSATX4DjfVmVtfWNzq7pd29nd2z+wD486Ok4VZW0ai1j1AqKZ4JK1gYNgvUQxEgWCdYPJTeF3p0xpHst7mCXMi8hI8pBTAkbybXsQERgHYfaUP2bgu7lv152GMwdeJW5J6qhEy7e/BsOYphGTQAXRuu86CXgZUcCpYHltkGqWEDohI9Y3VJKIa... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
can align AI models at present capability levels. Assuming this goal can be met, one of the next steps will be
to build consensus among researchers and to understand alignment in greater depth, including how techniques
scale with AI capabilities. The hope will be to create an evolving pragmatic state of the art for tra... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
Fig. 12: The structure learning approach (g) extrapolates various functions y = a · sin(bx) (top row),
y = ax· sin(bx) (middle row), and y = a
2x sin(bx) (bottom row) with different degrees of error. More
context also generally helps prediction accuracy (light vs. dark).
B.2 Table Sweeping: Additional Details
In Sect... | LargeLanguageModelsasGeneralPatternMachines |
Other Features We use uint8 types for storing masks2 to reduce memory footprint of our algo-
rithms. We also provide an example that converts masked dense parameters of a pruned ViT-B/16
model to sparse BCOO format and run the model with significantly lower memory footprint using
jax.experimental.sparse.
4 BASELINES
W... | JAXPRUNER |
With the advent of complex deep learning architectures the idea of a neuro-symbolic integration with knowledge graphs
has appeared also in image recognition tasks. An approach called Object-Oriented Deep Learning is presented by [58],
where the N-dimensional tensor of the deep net architecture is repla... | Knowledge graphs as tools for explainable machine learning: A survey |
paying for better performance at evaluation time rather than at training time. That is, rather than
pretraining an enormous model that must be used on all inputs, one might vary the number of passes
through a single frozen model based on an assessment of the input’s difficulty. | STANDING ON THE SHOULDERS OF GIANT FROZEN LANGUAGE MODELS |
sha1_base64="76w10YEtETzUXdaT0wTZt0xBig8=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuCG5cV7EPaacmkmTY0kxmSO0oZ+h9uXCji1n9x59+YtrPQ1gOBwzn3ck9OkEhh0HW/ncLa+sbmVnG7tLO7t39QPjxqmjjVjDdYLGPdDqjhUijeQIGStxPNaRRI3grGNzO/9ci1EbG6x0nC/YgOlQgFo2ilXjeiOArCrD3tYV/0yxW36s5BVomXkwrkqPfLX91BzNKIK2SSGtPx3AT9jGoUTPJpqZsanlA2pkPesVTRiBs/m... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
et al. (2022), some datasets (such as BoolQ) contain contexts extracted verbatim from the web, but not the
question and answer continuation. As such, highly contaminated samples from these datasets are unlikely
to gain an unfair advantage. The methodology in Chowdhery et al. (2022) further improves on the earlier
n-gra... | Llama2 |
ing the rationale quality score. For attributions with
respect to rationale logits, a(X)R, we measure the
effect of their occlusion on label accuracy. If at
least one of these values is notably different from
random, we can conclude that the I→OR model
displays feature-importance similarity in a given
direction. | Measuring Association Between Labels and Free-Text Rationales |
Torralba, A. and Efros, A. A. Unbiased look at dataset bias.
CVPR 2011, pp. 1521–1528, 2011.
Toshniwal, S., Sainath, T. N., Weiss, R. J., Li, B., Moreno,
P. J., Weinstein, E., and Rao, K. Multilingual speech
recognition with a single end-to-end model. 2018 IEEE
International Conference on Acoustics, Speech and Sig-
n... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
Usage
System Type
System Description
Upstream Dependencies
Upstream Dependencies
This is a standalone model.
None.
None.
Implementation Frameworks
Hardware & Software
for Training
Hardware & Software
for Deployment
Compute Requirements
Hardware: TPU v3 or TPU v4 (Jouppi et al., 2020).
Software: T5X (Roberts et al... | Scaling Instruction-Finetuned Language Models |
A.6 Dataset Contamination
With the increasing scale of publicly available training data, it has become inevitable that some portion of
evaluation data is seen during training, and may provide an undue boost in evaluation performance.
Earlier work (Brown et al. (2020), Wei et al. (2022a), Du et al. (2022) in measuring s... | Llama2 |
analytical solution for the mask matrix, which is then used to
update the pretrained weight. Concretely, SAM [41] views the
PEFT methods as p-sparse fine-tuned model by representing
fine-tuned parameter as W = W0 + M ∆W , M is a mask
matrix, and the optimization problem is
min∆W,ML(W0 + M ∆W ), | Parameter-EfficientFine-TuningMethods |
• We open-source a multimodal agent frame-
work, focusing on operating smartphone ap-
plications with our developed action space.
• We propose an innovative exploration strategy,
which enables the agent to learn to use novel
apps.
• Through extensive experiments across multi-
ple apps, we validate the advantages of o... | AppAgents |
strong assumptions about homogeneous treatment effects in order to generalize
to the real world; not everyone is exposed to misinformation, nor do they
necessarily pay attention when they are. As an exception to these designs, Kim
and Kim (2018) leverage variation in survey timing to estimate the causal effect
of misin... | Social_Media_and_Democracy |
ICON takes as input an RGB image of a segmented clothed
human and a SMPL body estimated from the image [32].
The SMPL body is used to guide two of ICON’s modules:
one infers detailed clothed-human surface normals (front and
back views), and the other infers a visibility-aware implicit
surface (iso-surface of an occupan... | ICON |
F.14 DM Mathematics
31
3651*w**2 + 519*w + 1
Find the second derivative of -91419126*m**2 - 162128943*m.
-182838252
Find the third derivative of 5*l*u*y**3 + l*u*y - 5*l*y**2 -
4621073*u*y**3 - 1755838*u*y**2 + u wrt y.
30*l*u - 27726438*u
Find the third derivative of 317297018*s**3 + 3136*s**2 - 30884*s wrt
s.
1903... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
[44] V. Pallagani, B. Muppasani, K. Murugesan, F. Rossi, L. Horesh, B. Srivastava, F. Fabiano,
and A. Loreggia. Plansformer: Generating symbolic plans using transformers. arXiv preprint
arXiv:2212.08681, 2022.
[45] R. Nakano, J. Hilton, S. Balaji, J. Wu, L. Ouyang, C. Kim, C. Hesse, S. Jain, V. Kosaraju,
W. Saunders, ... | LLM+P- Empowering Large Language Models with Optimal Planning Proficiency |
Adding pre-training noise to fix pre-training and fine-tuning discrepancies. To help fix the
pre-training perplexity and fine-tuning gap we tried pre-training the sparse models with a variety of
different types of noise. The goal was to help pre-training match the fine-tuning conditions where
dropout is used and more tokens... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
h[i > t] = 0 where t ∼ U (0, dh],
(5)
see the right-hand side of Figure 3. Zeroing out a sub-
set of h effectively means the corresponding set of 768-
dimensional vectors from the last linear layer will not be
used for that specific truncation t. This has the same effect
as dynamically changing dh without changing th... | A Neural Space-Time Representation for Text-to-Image Personalization |
In the present use case we concentrate on the collaboration between unmanned ve-
hicles and FRs, as visualized in Figure 2. A factory is on fire, while people might still
be in the building. The human-AI team arriving at the location consists of FRs and sev-
eral unmanned ground vehicles (UGVs) and unmanned aerial vehi... | DevelopingTeamDesignPatternsfor HybridIntelligenceSystems |
eanicarcrockthatextendssouthwardintoNewMexico.TheColoradoorogenywaslikelypartofthelargerYavapaiorogeny.Thought2:Itdoesnotmentiontheeasternsector.SoIneedtolookupeasternsector.Action2:Lookup[easternsector]Observation2:TheWyomingsectoroftheColoradoorogenywasformerlycalledtheMedicineBoworogeny.Theeasternsectorextendsintoth... | Tool Learning with Foundation Models |
3.2.3 Overall Model Architecture
We illustrate the detailed process in Figure 4. Con-
sistent with the previous stage, we use vvv-objective
diffusion and the 1D U-Net architecture. When con-
dition on the text embedding eee, we use the U-Net
configuration fθθθg (zzzσt; σt, eee) to generate the com-
pressed latent zzz =... | MOUSAI |
the crafted item .
Inventory ( xx /36) : My final inventory . For mining and smelting tasks ,
you only need to check inventory .
Chests : If the task requires me to place items in a chest , you can
find chest information here .
Task : The objective I need to accomplish .
Context : The context of the task .
You sh... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
51
Model architecture
Input(s)
Output(s)
Application
Known Caveats
Model Summary
Dense encoder-decoder models of 5 different sizes. See Table 2.
The model takes text as input.
See https://github.com/google-research/t5x/blob/main/docs/models.md
The model generates text as output.
See https://github.com/google-rese... | Scaling Instruction-Finetuned Language Models |
view, however. Cross-cutting interactions on social media and exposure to
diverse sources of news are at least as common as they are in the offline world
and, in many cases, more likely. Ranking algorithms, often blamed for serving
users what they want to see, do not appear to have as dramatic an effect on
polarization ... | Social_Media_and_Democracy |
The first property of IIVCG contracts requires that the agent maximizes the declared social
welfare. Therefore, when principals bid truthfully, the agent takes a welfare-maximizing action,
and social efficiency is achieved. The second property characterizes the expected payment from
each principal to the agent for his a... | Incomplete Information VCG Contracts for Common Agency |
3
Incentives
Let’s grant, then, that it becomes possible and financially feasible to develop APS systems. Should
we expect relevant actors to have strong incentives to do so, especially on a widespread scale?
I’ll assume that there are strong incentives to automate advanced capabilities, in general. But
building strat... | Is Power-Seeking AI an Existential Risk? |
Criteria for response selection. One advantage of USC is its generality: the same criteria can be
applied to various tasks, without any task-specific knowledge. Nonetheless, a minor task-specific
6
Universal Self-Consistency for Large Language Model Generation
89.2
(-1.4)
46.6
(+0.0)
GSM8K
BIRD-SQL
16
67.7 67.3... | UNIVERSALSELF-CONSISTENCYFORLARGELANGUAGEMODELGENERATION |
provided in audio format. Therefore, it is important to build a
video-music dataset tailored for music generation. Referring
to symbolic music generation literature[50, 22, 21], we build
SymMV, which includes 1140 piano music in MIDI format
with paired music videos and rich musical annotations. | VideoBackgroundMusicGeneration |
Intuitive Explanation. If the model is well-aligned and paired with a thoughtfully designed initial
prompt, the initial response should already be optimal given the conditions of the prompt and the
3We omit the analysis on HotpotQA because the sample size used in the source paper is quite small, which
may not produce... | LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET |
2 PRELIMINARY EXPLORATION ON CONTINUED PRE-TRAINING
Given the proven efficacy and efficiency of continued pre-training in adapting natural language
understanding models (Gururangan et al., 2020; Yao et al., 2021; Cheng et al., 2022), we embark on
an exploration to ascertain whether this method remains effective for la... | ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION |
1. Introduction
Research on 3D human pose estimation has gone
through enormous progress in recent years [14, 19, 39, 41,
46, 49, 60, 72]. While semi-supervised and self-supervised
approaches are on the rise [43, 82], best results are still
achieved when using as much labeled training data as possi-
ble. However, indiv... | Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats |
We use a mix of reviewers and automated systems to identify and enforce against misuse of
our models. Our automated systems include a suite of machine learning and rule-based classifier
detections that identify content that might violate our policies. When a user repeatedly prompts
our models with policy-violating conte... | gpt-4-system-card |
Empirical work on the effectiveness of banning hateful content yields mixed
results. Studying the effect of banning the /fatpeoplehate and /CoonTown
subreddits on Reddit in 2015, Chandrasekharan, Pavalanathan et al. (2017)
find the ban was successful. Analyzing more than 100 million Reddit posts and
comments, the author... | Social_Media_and_Democracy |
In Figure 3, we show the result of our MUSHRA study, which compares EnCodec to our proposed
codec at various bitrates. We find that our codec achieves much higher MUSHRA scores than
EnCodec at all bitrates. However, even at the highest bitrate, it still falls short of the reference
MUSHRA score, indicating that there i... | RVQGAN |
Fine-tuning is similar to enabling students to internal-
ize knowledge through extensive learning. This approach
is useful when the model needs to replicate specific struc-
tures, styles, or formats. Fine-tuning can enhance the perfor-
mance of non-fine-tuned models and make interactions more
efficient.
It is particula... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
more frequently in the pre-training corpus. As a
consequence, the approach might show brittleness
with respect to uncommon and creative instructions.
Dependence on large models. Because of SELF-
INSTRUCT’s dependence on the inductive biases
extracted from LMs, it might work best for larger | SELF-INSTRUCT- Aligning Language Model with Self Generated Instructions |
The Search for Meaningful Transparency and Oversight
As governments grapple with a host of complex policy options, transparency
mechanisms seem to provide a logical path forward. Platform companies seem
willing to become more transparent and are even increasingly doing so
voluntarily. When compared with broad and pote... | Social_Media_and_Democracy |
Annette Rid and Harald Schmidt. 2010. The 2008
declaration of helsinki — first among equals in re-
search ethics? Journal of Law, Medicine & Ethics,
38(1):143–148.
Darrell Rowbottom. 2011. Kuhn vs. popper on criti-
cism and dogmatism in science: A resolution at the
group level. Studies In History and Philosophy of
Scie... | A Two-Sided Discussion of Preregistration of NLP Research |
high risks it could present.[72, 73] | gpt-4-system-card |
• PaMIR. Our PaMIR representation has the ability to
condition the implicit field on the SMPL prediction,
which is realized by a novel network architecture con-
verting an image feature map and the corresponding
SMPL feature volume into an implicit surface represen-
tation. The SMPL feature volume is directly encoded
fr... | PaMIR- Parametric Model-Conditioned Implicit Representation for Image-based Human Reconstruction |
• Trajectory planning We propose a new method for
aligning temporal sequences in VLN comprising trajec-
tory estimation on path traces and subsequent predictions
for the distribution of linguistic descriptions over routes.
• Two in-domain datasets and training strategy We in-
troduce a set of path traces for routes in ... | APriorityMapforVision-and-LanguageNavigation withTrajectoryPlansandFeature-LocationCues |
Geirhos, R., Jacobsen, J.-H., Michaelis, C., Zemel, R., Bren-
del, W., Bethge, M., and Wichmann, F. A. Shortcut learn-
ing in deep neural networks. Nature Machine Intelligence,
2(11):665–673, 2020.
Ghorbani, B., Firat, O., Freitag, M., Bapna, A., Krikun,
M., Garcia, X., Chelba, C., and Cherry, C. Scaling
arXiv preprin... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.