text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
32
SQL: SELECT customers.customer_name FROM customers JOIN orders ON customers.
customer_id = orders.customer_id WHERE orders.order_status = "On Road" AND
orders.order_status = "Shipped" | Teaching Large Language Models to Self-Debug |
(cid:88)
|| ˆBfull
(cid:32)
(cid:33)
log
βdy
i (r) +
.
(7)
Lseg =
r
βdy
i (r)
By optimizing the two models using Eq. 7, we obtain a
motion segmentation mask Mi by thresholding αdy
i at 0.5.
We do not require an alpha regularization loss as in NeRF-
W [45] to avoid degeneracies, since we naturally include
s... | DynIBaR-NeuralDynamicImage-BasedRendering |
Malicious code On the Hugging Face platform, where the Stack is hosted, a malicious code
detection tool identified 654 files as unsafe. With the help of the BigCode community, we removed
these files ahead of the release of The Stack v1.2. Nevertheless, The Stack may contain undetected
malicious code, and StarCoder migh... | StarCoder_paper (1) |
5https://github.com/EleutherAI/
lm-evaluation-harness
#params.
OpenBookQA 32.8
82.4
SciQ
RACE
31.5
25.4
ARC
55.9
PIQA
73.1
ReCoRD
50.2
SST
MRPC
34.3
79.8
RTE
MultiNLI
61.3
MultiNLI (mis) 63.1
60.4
WSC
WinoGrande
55.2
49.4
WiC
38.9
HellaSwag
Average
52.9
T5 LaMini-T5 F-T5 LaMini-F-T5 C-GPT LaMini-C GPT-2 LaMini-GPT ... | LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions |
Diversity and quality Fréchet Inception Score (FID) [Heusel et al., 2017] is widely adopted for
image generation evaluations, which captures the similarity between generated and real images at
the distribution level in some feature space. It fits a Gaussian distribution for real samples and
one for generated samples in... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
o
w
p
e
o
p
l
e
w
o
r
k
a
n
d
l
i
v
e
.
I
b
e
l
i
e
v
e
t
h
e
r
e
’
s
t
r
e
m
e
n
d
o
u
s
v
a
l
u
e
t
o
b
e
c
a
p
t
u
r
e
d
b
y
p
r
o
d
u
c
t
b
u
i
l
d
e
r
s
w
h
o
c
a
n
s
u
c
c
e
s
s
f
u
l
l
y
p
u
t
t
h
e
p
o
w
e
r
o
f
A
I
i
n
t
o
p
r
o
d
u
c
t
s
t
h
a
t
p
e
o
p
l
e
... | Product-Led AI _ Greylock |
Hello Dolly: Democratizing the magic of ChatGPT with
open models
March 24, 2023 by Mike Conover, Matt Hayes, Ankit Mathur, Xiangrui Meng, Jianwei Xie, Jun Wan, Ali Ghodsi,
Patrick Wendell and Matei Zaharia in Company Blog
Update Apr 12, 2023: We have released Dolly 2.0, licensed for both
research and commercial use. ... | Dolly 2 Databricks |
Data sample
{’articleCounts’: {’eng’: 15},
’categories’: [{’label’: ’news/Business’, ’uri’: ’news/Business’, ’wgt’: 92}, ...],
’concepts’: [{’label’: {’eng’: ’Car finance’},
’score’: 100,
’type’: ’wiki’,
’uri’: ’http://en.wikipedia.org/wiki/Car_finance’}, ...],
’eventDate’: ’2019-10-01’,
’location’: None,
’relevance’: ... | Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio |
6.5 Impact of Knowledge Distillation
For small architectures, we distill larger models instead of training them from scratch. We use the distillation
procedure described in Sec. 5. We evaluate the effectiveness of this approach by comparing a ViT-L/14
trained from scratch with one distilled from a ViT-g/14 over 12 benc... | DINOv2- Learning Robust Visual Features without Supervision |
research/chatgpt.
https://openai.com/
OpenAI. 2023. Gpt-4 technical report.
Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Mered-
ith Ringel Morris, Percy Liang, and Michael S Bern-
stein. 2023. Generative agents: Interactive simulacra
of human behavior. In Proceedings of the 36th An-
nual ACM Symposium on User Int... | AppAgents |
Gao, L., Tow, J., Abbasi, B., Biderman, S., Black, S., DiPofi, A., Foster, C., Golding, L., Hsu,
J., Le Noac’h, A., Li, H., McDonell, K., Muennighoff, N., Ociepa, C., Phang, J., Reynolds, L.,
Schoelkopf, H., Skowron, A., Sutawika, L., Tang, E., Thite, A., Wang, B., Wang, K., and Zou, A.
(2023). A framework for few-shot... | TinyLlama |
A similar classification was developed by [40], who divided XAI ap-
proachesintothreecategories:scope(whethertheexplanationtargets
asingleforecastorattemptstoexplainthewholemodel),methodology
(iffocusedontheinputdataormodelparameters),andusage(ifinte-
gratedtothemodelorappliedtoanymodelingeneral).Liaoetal.[41]
dividedX... | Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio |
human behavior for users to engage with. In gaming, for instance,
these models have been employed to create interactive fiction [36]
and text adventure games [20]. With their ability to generate and
decompose action sequences, large language models have also been
used in planning robotics tasks [47]. For example, when ... | Generative Agents- Interactive Simulacra of Human Behavior |
Laudio-CFM-m(θ) = Et,m,q(x,z),p0(x0)||m ⊙ ((x − (1 − σmin)x0) − vt(w, xctx, z; θ))||2,
(6)
where the loss is only computed on masked frames. Appendix B.1 shows it leads to better results
Duration model We consider two solutions. The first one closely follows the audio model. It
models q(l | y, lctx) via a conditiona... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
Figure 2: GPT-3 175B validation accuracy vs. number of trainable parameters of several adaptation
methods on WikiSQL and MNLI-matched. LoRA exhibits better scalability and task performance.
See Section F.2 for more details on the plotted data points.
6 RELATED WORKS
Transformer Language Models. Transformer (Vaswani e... | LORA |
Tuckwood, C. (2014). The state of the field: Technology for atrocity response. Genocide
Studies and Prevention: An International Journal, 8(3), 81–86.
Twitter. (2017). Twitter Rules and Policies. https://help.twitter.com/en/rules-and-
policies/violent-groups
(2018). The Twitter Rules. https://support.twitter.com/art... | Social_Media_and_Democracy |
An aside on the router z-loss. One might think that the router z-loss is a convoluted method
replaceable by clipping logits (Wu et al., 2016). We explain why this is not the case. The goal is to
minimize large roundoff errors going into exponential functions. Clipping the logits occurs after any
roundoff errors – resul... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
where Z(x) =(cid:80)
y πref(y|x) exp
r(x, y) = β log
+ β log Z(x)
(cid:16) 1
β r(x, y)
(cid:17)
r′(x, y) = β log
πr(y|x)
πref(y|x)
which completes the proof.
We can further expand on these results. We can see that if r and r′ are two reward functions in the
same class, then
f (r, πref, β)(x, y) = β log
= β l... | Direct Preference Optimization |
Misinformation, Disinformation, and Online Propaganda
29
Campbell, A., Converse, P. E., Miller, W. E., & Stokes, D. E. (1960). The American
Voter. Chicago: University of Chicago Press.
Ciampaglia, G. L., Flammini, A., & Menczer, F. (2015). The production of information
in the attention economy. Scientific Reports, 5... | Social_Media_and_Democracy |
macro plan then serves as the input to an encoder-decoder model for surface realization. SANA [195]
is a skeleton-based two-stage model that includes skeleton generation to select key tokens from
the source table and edit-based generation to produce texts via iterative insertion and deletion
operations. In contrast to ... | SurveyofHallucinationinNatural Language Generation |
David Autor and Anna Salomons. Is automation labor-displacing? productivity growth, employment, and
the labor share. Technical report, National Bureau of Economic Research, 2018.
April H Bailey, Adina Williams, and Andrei Cimpian. Based on billions of words on the internet, people= | Llama2 |
Figure 1 shows the progression of the CrowS-Pairs gen-
der bias metric and the effect of the interventions. We
can clearly see a reduction in the bias as result of swap-
ping the gendered pronouns in the last 7% or 21% of the
training for all model sizes, but most prominently for the
larger ones, although these are als... | Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling |
Tom Brown, Benjamin Mann, Nick Ryder, Melanie
Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind
Neelakantan, Pranav Shyam, Girish Sastry, Amanda
Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen
Krueger, Tom Henighan, Rewon Child, Aditya Ramesh,
Daniel Ziegler, Jeffrey Wu, Clemens Winter, Chris
Hesse, Mark Chen, ... | MOUSAI |
-
SotA
15.1 19.7
BART
RAG-Tok. 17.3 22.2
RAG-Seq. 14.7 21.4
72.5
89.5
to more effective marginalization over documents. Furthermore, RAG can generate correct answers
even when the correct answer is not in any retrieved document, achieving 11.8% accuracy in such
cases for NQ, where an extractive model would score 0%... | Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks |
concept of an agent in the context of AI. In this paper, we treat AI agents as artificial entities that are
capable of perceiving their surroundings using sensors, making decisions, and then taking actions in
response using actuators [1; 4]. | TheRiseandPotentialofLargeLanguageModel BasedAgents |
female
lighter
0.0
0.0
95.8
6.7
0.0
0.0
97.7
17.2
male
darker
0.0
0.0
86.6
65.0
0.0
0.2
86.1
52.2
male
lighter
0.0
0.0
79.0
60.2
0.0
0.0
84.0
48.1
18-30
30-45
45-70
0.0
0.0
90.5
32.8
0.0
0.0
91.2
35.3
0.0
0.0
88.3
37.2
0.0
0.1
90.2
37.3
0.0
0.0
91.9
29.4
0.0
0.0
93.2
23.0
70+
0.0
0.0
82.3
6.5
0.0
0.0
88.7
9.7... | DINOv2- Learning Robust Visual Features without Supervision |
In addition to speech recognition, the transformer model has shown promising results in TTS
applications. The transformer based TTS model generates mel-spectrograms, followed by a WaveNet
vocoder to output the final audio results [309]. Several neural network-based TTS models, such as
Tacotron 2, DeepVoice 3, and trans... | AReviewofDeepLearningTechniquesforSpeechProcessing |
ule. To improve realism, we train our model using multiple
discriminators while also integrating geometric cues in the
form of predicted 2D normal maps. We experimentally find
that our method outperforms previous 3D- and articulation-
aware methods in terms of geometry and appearance. We
validate the effectiveness of ou... | AG3D- Learning to Generate 3D Avatars from 2D Image Collections |
hi, but no one ever answered.
One day, the pumpkin heard a voice. It was soft and gentle, and it said ”Hello”. The voice said ”Hello!”
The pumpkin was so happy!
The voice belonged to a very friendly bird. The bird flew above the pumpkin and said ”I can help you
move”. The bird said ”I can be your friend”.
So the bird f... | TinyStories-HowSmallCanLanguageModelsBeandStillSpeak CoherentEnglish? |
BeforeAfterBeforeAfterBeforeAfter JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015
[15] Y. Xu, S.-C. Zhu, and T. Tung, “Denserac: Joint 3d pose and
shape estimation by dense render-and-compare,” in The IEEE
International Conference on Computer Vision (ICCV), October 2019.
[16] T. Kanade, P. Rander, and P. J.... | PaMIR- Parametric Model-Conditioned Implicit Representation for Image-based Human Reconstruction |
+ 0.015 log
(A.1)
(cid:18) P
(cid:19)
1011
(cid:19)
(cid:18) D
8 · 104
where P is the number of parameters in the PM and D is the size of the dataset.
However, the results when training on another preference modeling data distribution look quite different, as
seen on the right in Figure 32. Note that there appe... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
on-distribution and maintain an accurate reward for the latest model.
In Table 6, we report the statistics of reward modeling data that we collected over time, and present them
against multiple open-source preference datasets including Anthropic Helpful and Harmless (Bai et al.,
2022a), OpenAI Summarize (Stiennon et al... | Llama2 |
InstructGPT Prompt → Write a short poem about a wise frog.
InstructGPT Response → The frog is a symbol of wisdom He knows all the secrets of the world He is a master of
disguise And a great teacher of life He is a symbol of transformation And the bringer of
change He is the frog who has seen it all And knows the meani... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
evaluate what types of counter-speech from what sources are most effective in
reducing online hate in diverse contexts. Recognizing the potential of counter-
speech bots, Leetaru (2017) proposed deploying AI bots en masse to fight online
hate speech, though the feasibility and consequences of such an intervention are
no... | Social_Media_and_Democracy |
Chatbot Arena [119] and MT-Bench [246] are two significant benchmarks that contribute to the
evaluation and advancement of chatbot models and LLMs in different contexts. Chatbot Arena
provides a platform to assess and compare diverse chatbot models through user engagement and
voting. Users can engage with anonymous mod... | ASurveyonEvaluationofLargeLanguageModels |
(cid:0)n1:K, x1:K|y(cid:1) ,
pa(z) = pnc
(1)
where pnc is the distribution of the normal maps n1:K and
color images x1:K observed from 3D assets conditioned on
an image y. For simplicity, we omit the symbol y for this
equation in the following discussions. Therefore, our goal
is to learn a model f that synthesizes m... | Wonder3D |
Knowledge Graph. The Extract-Transform-Load module consists
of a series of batch processes. The processes are executed regu-
larlytoensurethattheKGanddatabaseinformationisupdated
regarding the Enterprise Resource Planning software and Media
EventRetrievalSystem.
• Database: stores data relevant to the AI models, which ... | Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio |
Specialized heads do the heavy lifting, the rest can be pruned. arXiv preprint arXiv:1905.09418, 2019.
[31] Guillaume Wenzek, Marie-Anne Lachaux, Alexis Conneau, Vishrav Chaudhary, Francisco Guzm´an, Armand
Joulin, and Edouard Grave. Ccnet: Extracting high quality monolingual datasets from web crawl data. arXiv
prepri... | TinyStories-HowSmallCanLanguageModelsBeandStillSpeak CoherentEnglish? |
Overview To learn the representation space for mu-
sic, we deploy a diffusion magnitude autoencoder
(DMAE) shown in Figure 2. Specifically, we adopt
our diffusion-based audio autoencoder, introduced
in Section 3.1.3, to compress audio into a smaller
Figure 2: The training scheme of our diffusion magni-
tude autoencode... | Moûsai |
enser further comprising a means of identifying the user by voice recog-
nition. Also, an object is dispenser further comprising a means of identi-
fying a supervisor by voice recognition.
Furthermore, an object is a dispenser further comprising a means of cus-
tomizing the plurality of aural messages for instructing t... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
(∥τ − τ∗∥2+
(cid:88)
τ
(cid:88)
t
(x,y)∈Ot
√
λ
2π
σ
exp
(cid:18)
−∥τ∗
t − (x, y)∥2
2
2σ2
(cid:19)
)
,
(9)
where λ and σ are hyperparameters. The first term regulates
the optimized trajectory τ∗ to be similar to the original τ,
and the second term pushes the waypoint τ∗
t in the trajectory
away from the... | ALanguageAgentforAutonomousDriving |
S4. Alternative Derivation: Noise-Aware Preference Model
Paralleling the original DPO formulation we consider a policy trained on maximizing the likelihood of p(x0|c, t, xobs) where
xobs is a noised version of x0. Here x0 is an image, c is a text caption, t is a noising scale, and xobs is a corruption (noised
version) ... | DiffusionModelAlignmentUsing Direct Preference Optimization |
Test
3611
11314
2033
635
26849
101093*
10000
6666
parameters. The best performing "closed-book" (parametric only) open-domain QA model is T5-11B
with 11 Billion trainable parameters. The T5 model with the closest number of parameters to our
models is T5-large (770M parameters), which achieves a score of 28.9 EM on Nat... | Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks |
demonstrate diversity in length of the instructions,
instance inputs, and instance outputs in Figure 4. | SELF-INSTRUCT- Aligning Language Model with Self Generated Instructions |
when appropriately fine-tuned, we have also decided to scale up FLAN-ST, employing instruction
fine-tuning. | Mixture-of-Experts |
Yes, but not well itemized in existing
FEC reports; must use federal funds
for such ads
expenditure.” “Super” PACs have
no limits on the source of funding
for these ads
“Issue advocacy” ads
FEC-registered
PACs (incl.
“super” PACs)
Express advocacy ads
Yes, and classified as an “independent
“Issue advocacy” ads
Ye... | Social_Media_and_Democracy |
3
The adaptability of in-context learning lies in the amount of flexibility that can be packed into s1:k—this
prompt sequence can itself contain many sequences, each an input-output pair, and perhaps additional task
conditioning [38, 29]. Specifically, a model can in-context learn to complete a prompt which is a set ... | LargeLanguageModelsasGeneralPatternMachines |
is also asymmetric information between the principals themselves [... S]ince principals’
interests often diverge, they face incentives to advance their individual interests instead
of the joint interests by all principals[...] As a result, introducing governance to align
the interests of the principals with those of... | Incomplete Information VCG Contracts for Common Agency |
vation4:ChurfürstenisamountainrangeintheCantonofSt.Gallen,Switzerland.Theyformthenaturalboundarybetweenthecanton’sToggenburgandSarganserlanddistricts..TheyarethesouthernmostrangeoftheAppenzellAlps,separatedfromtheGlarusAlpsbytheSeezriverandWalensee..Theyconsistofalimestoneridgerunningeasttowest,withtheindividualpeaksfo... | Tool Learning with Foundation Models |
B
a
c
k
t
o
B
l
o
g
A
n
n
o
u
n
c
e
m
e
n
t
s
J
u
r
a
s
s
i
c
-
X
:
C
r
o
s
s
i
n
g
t
h
e
n
e
u
r
o
-
s
y
m
b
o
l
i
c
c
h
a
s
m
w
i
t
h
t
h
e
M
R
K
L
s
y
s
t
e
m
W
e
p
r
e
s
e
n
t
t
h
e
n
e
x
t
e
v
o
l
u
t
i
o
n
i
n
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
s
s
i
n
g
–
t
h
e
M
o
d
... | Jurassic-X_ Crossing the neuro-symbolic chasm with the MRKL system |
Both filtered and raw versions were produced, with
the raw version only deduplicated by URL. The fil-
tered version contains 65.86 GB of uncompressed
text across 17,103,059 documents. The raw version
is much larger, at 193.89GB of uncompressed text
across 69,547,149 documents.
23
C.4.1 Extractor Choice
We chose to use ... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
Such nimbleness also calls for a far more flexible approach to technology and innovation. Rather than bet on particular vendors or
proprietary products, companies need to be strictly agnostic about the technology they use to develop their products—and even
about the products and technologies they create. When Deon Nicho... | 4 Trends for AI Startups and Generative AI Companies |
4.6. Clustering
Filtering using example tests can still leave thousands of candidate programs per problem. Randomly
picking from this pool wastes the limited submission budget on programs that are syntactically
different but semantically equivalent. Semantically equivalent programs could be detected if we had
additional... | alphacode |
Implications of pro-and counterattitudinal
polarization. Human Communication Research, 40(3), 309–332.
Gerber, A. S., & Green, D. P. (2012). Field Experiments: Design, Analysis, and
Interpretation. New York: W. W. Norton.
Gorwa, R. (2017). Computational propaganda in Poland: False amplifiers and the digital
public sp... | Social_Media_and_Democracy |
In a separate task to collect the identity labels, annotators were asked to indicate all identities that were mentioned in the
comment. Identities included: disability, gender, race or ethnicity, religion, and sexual orientation. Rater could indicate
all options that applied:
What genders are mentioned in the comment?... | PaLM 2 Technical Report |
[134] Chattaraman, V., Kwon, W.-S., Gilbert, J.E., Ross, K.: Should ai-based, con-
versational digital assistants employ social- or task-oriented interaction style?
a task-competency and reciprocity perspective for older adults. Computers in
Human Behavior 90, 315–330 (2019) https://doi.org/10.1016/j.chb.2018.08.048
[... | PersonalityTraitsinLargeLanguageModels |
19
Method
MNLI SST-2 MRPC CoLA QNLI QQP
RTE
STS-B
Dataset
Optimizer
Warmup Ratio
LR Schedule
Batch Size
# Epochs
Learning Rate
Weight Decay
CLS Dropout
LoRA Config.
LoRA α
Max Seq. Len.
AdamW
0.1
Linear
4
10
6
8
1E-04
1E-04
0.01
0
0.1
0.1
rq = rv = 8
8
8
5
0
1E-04
0.15
8
16
6E-05
0.01
0
32
30
2E-0... | LORA |
Since prompt engineering is crucial to our role-playing framework, this section delves deeply into
our prompting techniques. Unlike other techniques for conversational language models, our prompt
engineering occurs solely at the beginning of role-playing, for task specification and role assignment.
Once the conversation... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
images/3D point cloud data [8, 40, 69, 74]. In addition to
harnessing the capabilities of LLMs for multi-modal un-
derstanding, researchers have also strived to utilize these
models to grasp the creative intentions of humans. For in-
stance, they have explored generating images [7], videos
[30], audio [48], or music [9... | M2UGen |
Author Contributions
Agustin Dal Lago worked on development of the dataset, evaluation, and general infrastructure.
Cyprien de Masson d’Autume worked on model development and analysis.
Daniel J. Mankowitz worked on clustering.
David Choi was the technical lead, developed initial prototypes for solving competitive progr... | alphacode |
based zero-shot learning with language models.
Computational Linguistics, 2022.
Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R. Bowman.
In
GLUE: A multi-task benchmark and analysis platform for natural language understanding.
ICLR (Poster). OpenReview.net, 2019.
Ben Wang and Aran Kom... | ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION |
[475] Cai, Z., B. Chang, W. Han. Human-in-the-loop through chain-of-thought. CoRR,
abs/2306.07932, 2023.
[476] Hancock, B., A. Bordes, P. Mazaré, et al. Learning from dialogue after deployment: Feed
yourself, chatbot! In A. Korhonen, D. R. Traum, L. Màrquez, eds., Proceedings of the 57th
Conference of the Association... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Table 2 shows that RAG-Token performs better than RAG-Sequence on Jeopardy question generation,
with both models outperforming BART on Q-BLEU-1. 4 shows human evaluation results, over 452
pairs of generations from BART and RAG-Token. Evaluators indicated that BART was more factual
than RAG in only 7.1% of cases, while ... | Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks |
• [Theory] 1. Scaling laws and diminishing returns: theoretical insights into scal-
ing laws for neural networks, particularly LLMs, suggest that as models become
larger, the benefits in performance improvement per parameter added diminish [12].
This phenomenon raises questions about the optimal size of LLMs and the bal... | Beyond Efficiency |
B Additional Experiments
B.1 Comparing audio model training objectives
While A3T is considered the regression-based speech infilling baseline, it is trained on a smaller
dataset and uses a smaller model compared to Voicebox. Here we present a controlled study comparing
the flow-matching and regression objectives, as ... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
Andy Zeng, Adrian Wong, Stefan Welker, Krzysztof Choroman-
ski, Federico Tombari, Aveek Purohit, Michael Ryoo, Vikas
Sindhwani, Johnny Lee, Vincent Vanhoucke, et al. Socratic
models: Composing zero-shot multimodal reasoning with lan-
guage. arXiv preprint arXiv:2204.00598, 2022. 11
Anthony Brohan, Yevgen Chebotar, Che... | JARVIS-1 |
Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann
Dubois, Xuechen Li, Carlos Guestrin, Percy Liang,
and Tatsunori B Hashimoto. 2023. Alpaca: A
strong, replicable instruction-following model. Stan-
ford Center for Research on Foundation Models.
https://crfm. stanford. edu/2023/03/13/alpaca. html,
3(6):7.
Harish Tayyar ... | AreEmergentAbilitiesinLarge Language Models just In-Context |
•
Student Recruitment & Admissionswww.ed.ac.uk/student-recruitment 7
You should also consider expected outputs to be achieved by the research such as a new database, fundamental
knowledge of a new or existing field, publications, attendance at conferences, contribution to a new policy, development
of a new tech... | research proposal guidance |
Saharia, C., Chan, W., Saxena, S., Li, L., Whang, J., Denton, E., Ghasemipour, S. K. S., Ayan, B. K., Mahdavi,
S. S., Lopes, R. G., Salimans, T., Ho, J., Fleet, D. J., and Norouzi, M. (2022). Photorealistic text-to-image
diffusion models with deep language understanding.
Sohl-Dickstein, J., Weiss, E. A., Maheswaranath... | Improving Image Generation with Better Captions |
[401] Yuan, H., C. Zhang, H. Wang, et al. Plan4mc: Skill reinforcement learning and planning for
open-world minecraft tasks. CoRR, abs/2303.16563, 2023.
[402] Hao, R., L. Hu, W. Qi, et al. Chatllm network: More brains, more intelligence. CoRR,
abs/2304.12998, 2023.
[403] Mandi, Z., S. Jain, S. Song. Roco: Dialectic... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
[11] Yoonwoo Jeong, Seokjun Ahn, Christopher Choy, Anima
Anandkumar, Minsu Cho, and Jaesik Park. Self-calibrating
neural radiance fields. In ICCV, 2021. 2
[12] Angjoo Kanazawa, Shubham Tulsiani, Alexei A. Efros, and
Jitendra Malik. Learning category-specific mesh reconstruc-
tion from image collections. In ECCV, 2018.... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
trees [40, 53, 81], account for the brute force approach of human-
authoring the agent’s behavior [70]. They provide a straightforward
way of creating simple agents that is still the most dominant ap-
proach today [68, 73, 109], and can even handle rudimentary social
interactions, as shown in simulation games such as M... | Generative Agents- Interactive Simulacra of Human Behavior |
the training of the S2ST model. It requires pre-trained ASR, MT, and TTS models, and multiple
training iterations. Wang et al. [2022] proposed an approach that combines teacher models and
pseudo-labeling to utilize unlabelled data. Their approach consists of three steps, in which the first
step is to adapt a pre-traine... | Translatotron3 |
4.1.2 Online Stage: Retrieve and Solving
In the online stage, given the user query with a task description, MLCopilot will respond with
the corresponding reasonable ML solutions via retrieving related experiences and knowledge, and
interacting with LLM by a curated prompt in one round.
4 | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
further eliminate the dependency on external aligners by
estimating or learning alignments that maximize the likeli-
hood of target mel-spectrograms (Zeng et al., 2020; Miao
et al., 2020; Kim et al., 2020). Meanwhile, generative adver-
sarial networks (GANs) (Goodfellow et al., 2014) have been
explored in second stage ... | ConditionalVariationalAutoencoderwithAdversarialLearningfor End-to-EndText-to-Speech |
o
u
n
c
e
m
e
n
t
s
A
I
2
1
L
a
b
s
p
a
r
t
n
e
r
s
w
i
t
h
A
m
a
z
o
n
t
o
o
| Announcing Jurassic-2 and Task-Specific APIs |
The AI performance bias presents an intriguing contrast with Sartori and Bocca [60] findings
on AI Anxiety. While individuals often express strong negative attitudes about AI replacing them
in certain tasks, it appears that when humans and AI work together, even in a non-functional
AI setting, joint performance is judg... | AI enhance sour performance |
(cid:13)(cid:13)(cid:13) ˆX∗(xt) − X∗(xt)
(cid:13)(cid:13)(cid:13)2
2
,
(cid:88)
xt
Lmatch =
(17)
and a geometric cycle consistency loss [18, 65] that forces
the image projection after forward warping of ˆX∗(xt) to
land back on its original 2D coordinates:
(cid:13)(cid:13)(cid:13)Πt(cid:16)W t,→( ˆX∗(xt))
(ci... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
User System Prompt:
Never forget you are a <USER_ROLE> and I am a
<ASSISTANT_ROLE>. Never flip roles! You will
always instruct me.
We share a common interest in collaborating to
successfully complete a task.
I must help you to complete the task.
Here is the task: <TASK>. Never forget our task!
You must instruct me base... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
teacher model is small or large. It is also clear that when the teacher model is large, the student model will
require more capacity to provide accuracy comparable to the teacher model. | Scaling Speech, Language and Vision Models with Mixture of Experts Technique - Microsoft Community Hub |
b
a
s
e
m
o
d
e
l
.
U
l
t
i
m
a
t
e
l
y
,
w
e
w
o
u
l
d
l
i
k
e
t
o
b
e
a
b
l
e
t
o
u
s
e
a
u
t
o
m
a
t
e
d
i
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
t
o
a
s
s
i
s
t
i
n
a
u
d
i
t
s
o
f
l
a
n
g
u
a
g
e
m
o
d
e
l
s
,
w
h
e
r
e
w
e
w
o
u
l
d
a
t
t
e
m
p
t
t
o
d
e
t
e
c
t
a
n
d
u
n
... | Language models can explain neurons in language models |
Code trainingInfilling code training . Instruction Fine-tuning Python code trainingLong context Fine-tuningLong context fine-tuningCᴏᴅᴇ Lʟᴀᴍᴀ - Iɴsᴛʀᴜᴄᴛ (7B ⇄, 13B ⇄, 34B)Cᴏᴅᴇ Lʟᴀᴍᴀ (7B ⇄, 13B ⇄, 34B)Cᴏᴅᴇ Lʟᴀᴍᴀ - Pʏᴛʜᴏɴ(7B, 13B, 34B)Lʟᴀᴍᴀ 2Foundation models(7B, 13B, 34B)100B<latexit sha1_base64="xxjQ0qU69VzePn... | CodeLlama2 |
employ NeRF as their 3D representation. DreamFusion-Scene
is a modified version of DreamFusion designed for generating
3D scenes, as the vanilla version focuses on 3D objects and is
not suitable for outward-facing scene generation. 3DP and Pix-
elSynth are two novel view synthesis methods using explicit
polygon meshes o... | Text2NeRF- Text-Driven 3D Scene Generation with Neural Radiance Fields |
3.4
Direct CoT Direct
13.8 10.3 34.6
34.5 24.1 76.9
44.8 51.7 80.8
48.3 51.7 88.5
13.8 13.8 34.6
13.8 10.3 26.9
17.2 13.8 23.1
31.0 13.8 42.3
27.6 31.0 26.9
27.6 24.1 57.7
20.7
26.9
24.1 13.8 50.0
27.6
26.9
41.4 31.0 53.8
17.2 24.1 19.2
17.2 20.7 57.7
27.6 34.5 57.7
17.2 34.5 69.2
31.0 48.3 65.4
27.6 48.3 88.5
41.4 31... | Scaling Instruction-Finetuned Language Models |
Thanks to our use of Nested Dropout, we can further
compress the representation of the concept by dropping a
significant subset of parameters in the network’s final layer.
When we reduce the number of units in the final layer from
the full 128 units to 32 units, the number of parameters de-
creases to 390, 000 paramete... | A Neural Space-Time Representation for Text-to-Image Personalization |
differences through a handful of individually defined rules
(e.g., shrink the hip–pelvis distance by a certain factor [68]),
but this does not scale to many keypoints and datasets—we
need a more systematic and automatic method. The ques-
tion we tackle in this work is therefore: How can we auto-
matically merge dozens o... | Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats |
Additional work has distinguished between hate speech directed at a group
(generalized hate speech) and hate speech directed at individuals (directed hate
speech) to capture important nuances in the targets of online hate speech
(ElSherief, Kulkarni et al. 2018). Beyond relying on textual
features,
researchers have als... | Social_Media_and_Democracy |
SOTA
ATT3D
Figure 12: Computation cost comparison against the base-
line methods. Left: comparison with the state-of-the-art
methods on the Animals dataset. Right: comparison with
a concurrent approach ATT3D on their dataset. Views-PP
is the number of rendered images per prompt in training,
and CLIP-RP is the retriev... | Instant3D |
Shehzaad Dhuliawala, Mojtaba Komeili, Jing Xu, Roberta Raileanu, Xian Li, Asli Celikyilmaz, and
Jason Weston. Chain-of-verification reduces hallucination in large language models. arXiv preprint
arXiv:2309.11495, 2023.
Ning Ding, Yulin Chen, Bokai Xu, Yujia Qin, Zhi Zheng, Shengding Hu, Zhiyuan Liu, Maosong
Sun, and B... | ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup |
2.4 Long context fine-tuning | CodeLlama2 |
harm; and such systems may, ultimately, have just claims to things like political rights, autonomy,
and so forth. In fact, I think that part of what makes alignment important, even aside from its role in
making AI safe, is its role in making our interactions with AI moral patients ethically acceptable.169
It’s one thin... | Is Power-Seeking AI an Existential Risk? |
Some works [114, 153, 167] have dealt with the hallucination issue at the instance level by using a
score for each source-reference pair and filtering out hallucinated ones. This corpus filtering method
consists of several steps: (1) measuring the quality of the training samples in terms of hallucination
utilizing the ... | SurveyofHallucinationinNatural Language Generation |
2.2.2 Case Study: Containers
Let’s consider a single case study in some detail—the fact that (under ordinary
circumstances), the water in a tea kettle with the lid on can only come out the spout.
As the reader might expect by now, GPT-2 sort of gets this some of the time. Sort of.
In principle, we mig... | The Next Decade in AI- |
day, while he was cleaning up the yard, he stumbled upon a mysterious object buried in the
ground. The object was about the size of a large grapefruit and was made of a shiny metal that
James had never seen before. As he picked it up, he noticed that it had strange markings on it
that he couldn’t decipher. Curious, Jam... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
E.3.2. Incorrect or irrelevant rewordings
To investigate what parts of the problem description the model pays attention to and how strongly it
conditions on the description, we investigated how the solve rate for a problem changes when we
add irrelevant information to the description, or reword it to be under-specified ... | alphacode |
precision). To combine GOLD and tempering, we introduce a short training phase between pretraining
and finetuning. Full details of GOLD and this combination are in Appendix C.3. | alphacode |
Authors’ addresses: Yupeng Chang, ypchang_jluai@outlook.com; Xu Wang, xwang22@mails.jlu.edu.cn, School of Artificial
Intelligence, Jilin University, 2699 Qianjin St, Changchun, China, 130012; Jindong Wang, Microsoft Research, Beijing,
China, Jindong.wang@microsoft.com; Yuan Wu, School of Artificial Intelligence, Jilin ... | ASurveyonEvaluationofLargeLanguageModels |
external datasets to the performance of existing demand forecasting
models. Third, we envision enhancing the explanations by including
meaningfulcurrenteventsreportedbythemedia.Thisway,inaddition
toagoodunderstandingofthepastcontext,theexplanationswillpro-
videinformationoneventsthatarelikelytoinfluencefuturedemand
so ... | Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio |
Figure 11: Effect of various gradient attribution methods on the ROAR test at k = 10 − 30% occlusion for the
CoS-E v1.0 validation set. We compute attributions with respect to the label logit and measure label accuracy of the
resulting model after masking and re-training (see §4.2 for details). The largest drop in perf... | Measuring Association Between Labels and Free-Text Rationales |
C Dataset Details
This section contains additional information about
each dataset listed in Section 2, including how it
was obtained, how it was processed, and any other
details relevant for replication. The intent of this
section is to provide as much detail as possible,
so that Pile can be replicated in the future if... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.