text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
be seen, existing LLMs struggle to solve mathematical problems in backward rationales and our
MetaMath has a significant improvement on both datasets. Specifically, the ways where different
LLMs solve the backward mathematical problem are illustrated through examples in Appendix A.3. | METAMATH |
first earthquake?Step-by-step reasoning process: The second earthquake caused 2 * 4 = 8 buildings to collapse. The third earthquake caused 2 * 8 = 16 buildings to collapse. The fourth earthquake caused 2 * 16 = 32 buildings to collapse. In total, the earthquakes caused 4 + 8 + 16 + 32 = 60 buildings to collapse.The ans... | Scaling Instruction-Finetuned Language Models |
Problem
1554A Cherry
1559A Mocha and Math
1569A Balanced Substring
No consecutive zeros
Nim
Original Simplified
2.98%
15.74%
55.53%
12.25%
31.97%
10.61%
1.25%
0.17%
0.95%
85.38%
Appendix Table A9 | Performance on original vs. simplified problems. The percentage of correct
samples from a total of 100k samples, for origi... | alphacode |
OpenSea
Up to 2.5%
Uniswap
0.30%¹
Ethereum
~0.06%²
1/ Most popular fee tier 2/ Calculated as total gas fees paid by users divided by total transfer value of ETH and top ERC20 tokens in 2022 (Source: Coin Metrics)
a16z crypto
State of Crypto
2023
Why Web3 Matters
8
Web3 counterbalances the trend ... | State-of-Crypto2023 |
The datasets in Table 2 are linked to tasks for which they have been designed or most commonly used. The majority of
online art collections include general annotations related to the whole image and are often used for classification or
retrieval tasks. Those annotations are mostly provided by art experts and contain inf... | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
in order to impact
Indeed, computational propaganda has been propelled by a broader scale
normalization of social media as a means for control by those focused on digital
political communication (Karpf 2012). As Chadwick (2013) points out, “even
the most radical changes to communications systems must be channeled
thro... | Social_Media_and_Democracy |
[73] Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Ed Chi, Quoc Le, and Denny Zhou. Chain
of thought prompting elicits reasoning in large language models. arXiv preprint arXiv:2201.11903, 2022.
[74] Sergey Zagoruyko and Nikos Komodakis. Paying more attention to attention: Improving the performance
of convolu... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
5. An agreed line and level of questioning should be adopted and questions must be relevant to
entry criteria. Supplementary questions should be used to probe for further information or
clarification where answers are incomplete or ambiguous.
Interviews should be assessed against predetermined criteria which are c... | UCL Academic Manual |
Loubna Ben Allal, Raymond Li, Denis Kocetkov, Chenghao Mou, Christopher Akiki, Carlos Munoz
Ferrandis, Niklas Muennighoff, Mayank Mishra, Alex Gu, Manan Dey, Logesh Kumar Umapathi,
Carolyn Jane Anderson, Yangtian Zi, Joel Lamy Poirier, Hailey Schoelkopf, Sergey Troshin,
Dmitry Abulkhanov, Manuel Romero, Michael Lappert... | StarCoder_paper (1) |
[21] R. Karimi Mahabadi, S. Ruder, M. Dehghani, and J. Henderson,
“Parameter-efficient multi-task fine-tuning for transformers via shared
hypernetworks,” in Proc. Annu. Meeting Assoc. Comput. Linguistics,
Int. Joint Conf. Natural Lang. Process., 2021, pp. 565–576.
[22] A. Chronopoulou, M. Peters, A. Fraser, and J. Dod... | Parameter-EfficientFine-TuningMethods |
[9] Jooyoung Choi, Sungwon Kim, Yonghyun Jeong, Youngjune
Gwon, and Sungroh Yoon.
Ilvr: Conditioning method for
denoising diffusion probabilistic models. In Proceedings of
the IEEE/CVF International Conference on Computer Vision
(ICCV), pages 14367–14376, October 2021.
[10] Hyungjin Chung, Byeongsu Sim, and Jong Chul ... | Relightify-Relightable3DFacesfromaSingleImageviaDiffusionModels |
A paper which a) is confirmatory and b) concerns
an application for which risk tolerance is low, can
be rejected for not having preregistered. We noted
the necessity of allowing preregistered research to
re-classify as exploratory, i.e., conditional accep-
tance for non-preregistered, flagged research, if it
explicitly l... | A Two-Sided Discussion of Preregistration of NLP Research |
Mehrish et al.
• Common Voice: Mozilla’s Common Voice project [17] is dedicated to producing an accessible,
unrestricted collection of human speech for the purpose of training speech recognition
systems. This ever-expanding dataset features contributions from more than 9, 000 speakers
spanning 60 different languages.
•... | AReviewofDeepLearningTechniquesforSpeechProcessing |
2022-12-03. 1, 4
score, 2018. 6
[2] Shane Barratt and Rishi Sharma. A note on the inception
[3] Shariq Farooq Bhat, Ibraheem Alhashim, and Peter Wonka.
Adabins: Depth estimation using adaptive bins. In Proceed-
ings of the IEEE/CVF Conference on Computer Vision and
Pattern Recognition (CVPR), pages 4009–4018, June 2... | LDM3D- Latent Diffusion Model for 3D |
On MMLU (Hendrycks et al., 2021a), Gemini Ultra can outperform all existing models, achieving
an accuracy of 90.04%. MMLU is a holistic exam benchmark, which measures knowledge across a
set of 57 subjects. Human expert performance is gauged at 89.8% by the benchmark authors, and
Gemini Ultra is the first model to excee... | gemini_1_report |
2.2 Decision-making with AI
Decision-making, a process shaped by expectations and perceptions, involves selecting from a
range of options [67]. The Drift Diffusion Model (DDM) serves as a cognitive framework for
understanding this process, describing it as evidence accumulation until a decision boundary (a
correct vs. ... | AI enhance sour performance |
create a sliding window with hop length equal to 250ms and run each chunk of audio through the
classifier and average the outputs.
The model is tested on the dev-clean split of Librispeech. We then take a 100 hour subset of the 60K
hour-English data and set aside 2,703 random utterances (to match the size of dev-clean)... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
10
24681012Bitrate (kbps)0102030405060708090MUSHRA scoreReference (speech)Proposed (speech)EnCodec (speech)24681012Bitrate (kbps)0102030405060708090MUSHRA scoreReference (music)Proposed (music)EnCodec (music)24681012Bitrate (kbps)0102030405060708090MUSHRA scoreReference (env)Proposed (env)EnCodec (env) References
[1] ... | RVQGAN |
[28] D. Ganguli, L. Lovitt, J. Kernion, A. Askell, Y. Bai, S. Kadavath, B. Mann, E. Perez,
N. Schiefer, K. Ndousse, A. Jones, S. Bowman, A. Chen, T. Conerly, N. DasSarma, D. Drain,
N. Elhage, S. El-Showk, S. Fort, Z. Hatfield-Dodds, T. Henighan, D. Hernandez, T. Hume,
J. Jacobson, S. Johnston, S. Kravec, C. Olsson, S. R... | gpt-4-system-card |
Our journey to create a commercially viable model
We also wanted to produce an open source model that can be commercially used. Despite
databricks-dolly-15k being substantially smaller than Alpaca, the dataset on which Dolly 1.0
was trained, the resulting Dolly 2.0 model, based on EleutherAI’s pythia-12b, exhibited hi... | Dolly 2 Databricks |
m
e
t
a
d
a
t
a
a
s
w
e
l
l
a
s
a
u
d
i
o
f
i
l
e
d
u
r
a
t
i
o
n
a
n
d
s
t
a
r
t
t
i
m
e
,
a
l
l
o
w
i
n
g
f
o
r
c
o
n
t
r
o
l
o
v
e
r
t
h
e
c
o
n
t
e
n
t
a
n
d
l
e
n
g
t
h
o
f
t
h
e
g
e
n
e
r
a
t
e
d
a
u
d
i
o
.
T
h
i
s
a
d
d
i
t
i
o
n
a
l
t
i
m
i
n
g
c
o
n
d
i
t
i
o
n
... | Stable Audio_ Fast Timing-Conditioned Latent Audio Diffusion — Stability AI |
Training the Gemini family of models required innovations in training algorithms, dataset, and
infrastructure. For the Pro model, the inherent scalability of our infrastructure and learning algorithms
enable us to complete pretraining in a matter of weeks, leveraging a fraction of the Ultra’s resources.
The Nano series... | gemini_1_report |
by GPT-4 are impacted by the prompt; future work may study the best way to elicit high-quality
judgments from automated systems. Finally, many possible applications of DPO exist beyond training
language models from human preferences, including training generative models in other modalities. | Direct Preference Optimization |
There are several types of funding for postgraduate
research: your own funds; external funding bodies such as
charities and trusts; national and governmental agencies;
employers and the private sector, and internal sources
such as University scholarships, funded studentships and
projects advertised by supervisors.... | research proposal guidance |
15 | QLORA |
2. Related Work
The physics of free-viewpoint rendering involves mod-
eling geometry and surface properties and then rendering
from new camera views. However, it remains difficult to
recreate complex geometry and subtle lighting effects. Al-
ternatively, image-based rendering [57, 61] offers to render
novel views base... | HumanNeRF- Free-viewpoint Rendering of Moving People from Monocular Video |
Recently, the field of 3D generation has experienced
rapid and flourishing development with the introduction of
diffusion models. A growing body of research [5, 29, 43,
59, 63], such as DreamField [24], DreamFusion [43], and
Magic3D [29], resort to distilling prior knowledge of 2D
image diffusion models or vision langu... | Wonder3D |
Place of work
Rämistrasse 69, Zürich, Switzerland
Start of employment
Employment start date to be mutually agreed, ideally 1st of November 2023. Please submit your application by 01.09.2023. Your
application should include: your CV, motivation letter (specify your research interests and how they relate to the topic of... | UZH_ PhD Position in Digital Humanities_ From Text to Image with AI |
Feedback: With the above function, count_ways(2) = 3. The assertion is "
assert count_ways(2) == 3". So the code passes the assertion. The code above
is correct.
### Task End ###
66
### Task Start ###
# These are the assertions for your function:
assert find_Rotations("aaaa") == 1
""" Write a python function to fi... | Teaching Large Language Models to Self-Debug |
Method
Fine-Tune (354M)
AdapterL (0.37M)
AdapterL (11M)
FTTop2 (24M)
Prefix (0.35M)
LoRA (0.35M)
Fine-Tune (774M)
AdapterL (0.88M)
AdapterL (23M)
Prefix (0.77M)
LoRA (0.77M)
BLEU↑
S
64.2
54.5
60.4
53.6
62.9
62.1±.2
65.3
61.1±.0
64.7±.2
63.4
64.0±.3
U
27.7
45.1
48.3
18.9
45.6
46.7±.4
43.1
49.8±.0
49.2±.1
47.7
48.... | LORA |
adtoanexcessivelylargepolicyupdate;hence,wenowconsiderhowtomodifytheobjective,topenalizechangestothepolicythatmovert(θ)awayfrom1.Themainobjectiveweproposeisthefollowing:LCLIP(θ)=ˆEthmin(rt(θ)ˆAt,clip(rt(θ),1−(cid:15),1+(cid:15))ˆAt)i(7)whereepsilonisahyperparameter,say,(cid:15)=0.2.Themotivationforthisobjectiveisasfoll... | PPO |
96.3
90.8
91.2
92.4
99.2
93.5
77.7
89.6
96.6
95.1
98.0
27.1
21.7
96.1
74.7
95.2
86.5
62.3
41.9
47.4
Table 12: ST-MoE-32B versus previous best for inference-only techniques and fine-tuned mod-
els. A split of “dev/test” refers to dev split for Zero-Shot and One-Shot and test split for Fine-Tune
quality. Data not availab... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
Meeting of the Association for Computational Lin-
guistics (Volume 1: Long Papers), pages 6997–7013,
Dublin, Ireland. Association for Computational Lin-
guistics.
Kevin Cohen, Kar¨en Fort, Margot Mieskes, Aur´elie
N´ev´eol, and Anna Rogers. 2021. Reviewing nat-
ural language processing research. In Proceedings
of the ... | A Two-Sided Discussion of Preregistration of NLP Research |
148 ± 322
115 ± 175
206 ± 298
352 ± 134
Table 2: Statistics of seed, self-augmentation and self-curation finetuning data. Instruction and output
lengths are given as the number of characters.
(a) Seed data.
(b) Augmented data in A5
Figure 2: Instruction diversity of seed data and augmented data. The inner circle sh... | Self-AlignmentwithInstructionBacktranslation |
i, j=0
i, j=0
,
i, j=0
and
4.3 Feature extraction
(a)k = a(a + 1)...(a + k− 1) =
Γ(a + k)
Γ(a)
Equation (4) is the Pochhammer symbol.
The set of (N+1) Krawtchouk polynomial forms the
complete set of discrete basis functions with weight
function
w(x; p,N) =
px(1− p)N−x
(5)
(cid:19)
(cid:18) N
x
For a giv... | VISAPP_HumanPoseEstimation |
6 Conclusion
In this work, we introduce VOYAGER, the first LLM-powered embodied lifelong learning agent,
which leverages GPT-4 to explore the world continuously, develop increasingly sophisticated skills,
and make new discoveries consistently without human intervention. VOYAGER exhibits superior
performance in discove... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
Delta skinning weights We qualitatively evaluate the effect
of delta skinning weights. As shown in Fig. 15, without
learning a delta skinning weights specific to each 3D point,
the reconstructed shape and motion may be over-regularized
by the 3D Gaussians.
Figure 13. Sensitivity to number of bones. | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
ion:11.0FinalAnswer:11(pinkflowerstones)Query:Mrs.Hiltwenttoaconcert.Atotalof65,899peopleattendedtheconcert.Thenextweek,shewenttoasecondconcert,whichhad119morepeopleinattendance.Howmanypeoplewereatthesecondconcert?Response:Thought:Thesecondconcerthad119morepeoplethanthefirstconcert.Ifweadd119totheattendanceofthefirstconce... | Tool Learning with Foundation Models |
X-Denoising is Complementarily Effective but Does Not Suffice as a Standalone We observe that mixing
Extreme Denoising is effective. Most of the best results across the board come from mixtures with long spans
(e.g., 32 or 64). When compared with variants without long spans (Var-D vs. Var-C), we see that Var-D
is strictly ... | UL2- Unifying Language Learning Paradigms |
(2018) demonstrated jointly training a modern deep learn-
ing speech recognition system on several languages with a
single model, and Pratap et al. (2020a) scaled this line of
work significantly to 50 languages with a billion-parameter
model. MUTE (Wang et al., 2020c) and mSLAM (Bapna
et al., 2022) studied joint trainin... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
(cid:3)S1, E1(cid:4) and G2 = (cid:3)S2, E2(cid:4). Let τ = (cid:3) f , R(cid:4) be a transformation from F1 to F2. Then, τ is a DLBS transformation from F1
to F2 if there is a set M ⊆ 2V 1·D1 of landmarks and a bijection g : A1 → A2 such that conditions (1)–(5) below hold: First
define the variable set V M = {vϕ | ... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
(3)
Where pce and pstu are the probabilities from the cross-encoder teacher model and our student model.
α is a hyperparameter to balance the two loss functions. Lcont is the same as in Equation 1.
min DKL(pce, pstu) + αLcont
4.3 Applications to Text Embedding Tasks | E5 |
[t1(˜b, o)] = h1(˜b−1) ≥ (cid:15), while Eo∼F|a2
2 ·(cid:80)
o∈O 1
33 | Incomplete Information VCG Contracts for Common Agency |
3.3 Caption type results
We start by analyzing the performance difference between models trained on different types of captions. For
this evaluation, we train three models:
1. A text-to-image model trained only on ground truth captions.
2. A text-to-image model trained on 95% short synthetic captions.
3. A text-to-im... | Improving Image Generation with Better Captions |
Q: Alice, Bob, and Claire are playing a game. At the start of the game, they are each holding a ball: Alice has a pink ball,
Bob has a brown ball, and Claire has a orange ball. As the game progresses, pairs of players trade balls. First, Claire and
Alice swap balls. Then, Alice and Bob swap balls. Finally, Bob and Clai... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
1
Introduction
Interpretable NLP aims to better understand pre-
dictive models’ internals for purposes such as de-
bugging, validating safety before deployment, or
revealing unintended biases and behavior (Molnar,
2019). These objectives require faithful rationales—
explanations of the model’s behavior that are accu-... | Measuring Association Between Labels and Free-Text Rationales |
Pengcheng Yin, Bowen Deng, Edgar Chen, Bogdan
Vasilescu, and Graham Neubig. 2018. Learning to
mine aligned code and natural language pairs from
stack overflow. In International Conference on Min-
ing Software Repositories, MSR, pages 476–486.
ACM.
Shuyan Zhou, Uri Alon, Sumit Agarwal, and Gra-
ham Neubig. 2023. Codebe... | CODEFUSION |
Q: How might a person address someone who is leaving? Choices: A.further cause B.wave goodbye C.attempt suicide
D.insulting them E.thank god
A: Reasoning process: 1. We know that the person is leaving, so we can eliminate answer choices A and C because they
don’t make sense in context. 2. We can further eliminate answe... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
1.2.2 Agency | Is Power-Seeking AI an Existential Risk? |
John Jumper, Richard Evans, Alexander Pritzel, Tim Green, Michael Figurnov, Olaf Ronneberger, Kathryn
Tunyasuvunakool, Russ Bates, Augustin Žídek, Anna Potapenko, et al. Highly accurate protein structure
prediction with alphafold. Nature, 596(7873):583–589, 2021. URL https://www.nature.com/
articles/s41586-021-03819-2.... | Tool Learning with Foundation Models |
8
Universal Self-Consistency for Large Language Model Generation
et al., 2023), and utilize it for reranking (Zhang et al., 2023a; Huang et al., 2023a). In this work, we
directly instruct the LLM to perform consistency-based selection without an explicit definition of the
pairwise similarity, and we demonsrate its a... | UNIVERSALSELF-CONSISTENCYFORLARGELANGUAGEMODELGENERATION |
Query: please recognize named entities in the sentence [ " that is to end the state of hostility , " Thursday's overseas edition of the People's Daily quoted Tang as saying. ]Response: I have recognized named entities in the sentence 'that is to end the state of hostility , ' Thursday's overseas edition of the People's... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
5
910M23456789100M234567891B234567234567891011ArchitectureBert-BaseFunnel6 Layers8 Layers16 Layers24 LayersDeepNarrow (24 L)Thin (H=512)Wide(H=1024)Embedding E=12824 Heads6 Heads3 Heads1 HeadFFN every 2LFFN every 4LTotal Tokens IngestedMLM Loss2B2.5B3B3.5B4B4.5B5B5.5B6B6.5B7B1.922.12.22.32.42.52.62.72.82.93Total Token... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
87.5
85.9
73.2
0.0
88.4
83.6
53.8
49.8
52.0
52.7
54.9
52.0
84.6
83.6
13.7
-
85.5
82.0
Table 5: Ablation study, which improvements were most important? The first group shows an ablation where
one component of the final combination of training, architecture, and data modifications (the crammed BERT
model) is replaced by ... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
Summarization Upon obtaining the correct answer, which derives the accurate reasoning chains
(based on the assumption that when a correct answer is generated, the reasoning chain is most
likely correct), the Summary-Prompt method is employed to prompt LLMs to generate the final
reasoning chain (Iter-CoT) based on prior ... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
While the benefits of Large Language Models (LLMs) are indisputable, their rapidly growing size
and computational requirements pose significant challenges in terms of training efficiency, memory
footprint, and deployment costs. Consequently, there is a pressing need for developing scalable
techniques that can harness t... | Mixture-of-Experts |
The profound challenge of incorporating human knowl-
edge into neural-network-based driving systems is reconcil-
Figure 3. Illustration of function calls in the tool library. Agent-
Driver can effectively collect necessary environmental information
from neural modules through dynamic function calls. | ALanguageAgentforAutonomousDriving |
36
A Appendix
A.1 Complicate Input Prompt
Example-1:
I want you act as a Prompt Rewriter.
into a more complex version using dataformat to make those famous AI systems (e.g.,
chatgpt and GPT4) more difficult to handle.
reasonable and must be understood and responded by humans.
You must add [XML data] format text as ... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
[17] B. Roessle, J. T. Barron, B. Mildenhall, P. P. Srinivasan, and M. Nießner,
“Dense depth priors for neural radiance fields from sparse input views,”
in Proceedings of the IEEE/CVF Conference on Computer Vision and
Pattern Recognition, 2022, pp. 12 892–12 901.
[18] J. Wu, C. Zhang, T. Xue, B. Freeman, and J. Tenenba... | Text2NeRF- Text-Driven 3D Scene Generation with Neural Radiance Fields |
the Indo-European language family and many of which are
high-resource languages. These benchmarks only provide
limited coverage and room to study Whisper models multi-
lingual capabilities which include training data for speech
recognition in 75 languages. To study the performance of
Whisper more broadly we also report... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
We follow Ho et al. (2022b) in jointly training all the models in the Imagen Video pipeline on images
and videos. During training, individual images are treated as single frame videos. We achieve this
by packing individual independent images into a sequence of the same length as a video, and bypass
the temporal convolu... | IMAGEN VIDEO- HIGH DEFINITION VIDEO GENERATION WITH DIFFUSION MODELS |
Reasoning Corpus. arXiv:2302.09425, 2023.
[24] S. Alford. A Neurosymbolic Approach to Abstraction and Reasoning. PhD thesis, Massachusetts Institute of
Technology, 2021.
[25] R. Assouel, P. Rodriguez, P. Taslakian, D. Vazquez, and Y. Bengio. Object-centric Compositional Imagination for
Visual Abstract Reasoning. In ... | LargeLanguageModelsasGeneralPatternMachines |
human perception of AI Art, another topic of interest are visual characteristics of AI Art in the context of art history.
Hertzmann introduces the concepts of visual indeterminacy as a specific stylistic property of AI Art created using GANs
[63]. Srinivasan and Uchino [115] explore the biases in the generative art AI p... | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
61%
60%
52%
35%
30%
40%
Confidence
Indifference
26%
17%
14%
21%
+17pp
–10pp
2023
2018
Optimism grows with familiarity, and respondents who use generative AI
regularly are far more bullish than those who have never tried it
Optimism
62%
55%
36%
Concern
42%
27%
22%
Nonusers
Rare
users
Regular
use... | AI at Work- What People Are Saying |
60.8
82.0
87.2
85.6
0.0
41.6
0.0
61.2
0.0
58.5
42.8
54.0
45.2
40.4
24.8
50.8
48.6
0.0
42.9
0.0
69.6
38.0
45.2
62.0
52.8
31.6
30.8
29.6
42.8
32.4
53.2
35.2
66.0
36.8
76.0
35.6
47.6
36.8
74.0
63.6
85.6
0.0
47.2
0.0
57.2
0.0
80.6
33.6
48.4
60.8
33.6
50.8
57.2
60.8
0.0
53.7
0.0
90.8
0.8
1.2
1.2
1.2
0.0
0.... | Mixture-of-Experts |
conclusion
In
an
and
persistence
of misinformation.
Within both popular media and academia, concerns abound regarding the
prevalence
age where
misinformation can diffuse rapidly via the Internet and social media, it is
more imperative than ever to think creatively about how best to debunk
misinformation. Although... | Social_Media_and_Democracy |
of gist descriptors for web-scale image search. In CIVR, 2009.
Quentin Duval, Ishan Misra, and Nicolas Ballas. A simple recipe for competitive low-compute self supervised
vision models. arXiv preprint arXiv:2301.09451, 2023.
Alaaeldin El-Nouby, Gautier Izacard, Hugo Touvron, Ivan Laptev, Hervé Jegou, and Edouard Gra... | DINOv2- Learning Robust Visual Features without Supervision |
0/22
22/22
19/22
0/22
19/22
0/22
improvement at 8B, DoReMi (150M→8B) and DoReMi (1B→8B) still achieve the baseline accuracy
almost 2x faster. This suggests that DoReMi is robust to the proxy model scale.
Choosing the easiest or hardest domains do not suffice. We ablate the components of the excess
loss metric (cid:96)... | DoReMi- Optimizing Data Mixtures Speeds Up Language Model Pretraining |
8 Probabilities | Is Power-Seeking AI an Existential Risk? |
LLaMA is a transformer model similar to GPT with the following modifications.
Normalize the input of each transformer sub-layer to improve training stability.
Use SwiGLU instead of ReLU to improve performance.
Use rotary embedding instead of absolute positioning to improve performance.
The table below summarizes the ... | A brief history of LLaMA models - AGI Sphere |
synthetic captions".
Figure 3 shows examples of ground-truth, short synthetic, and descriptive synthetic captions.
Once built, we apply our image captioner fine-tunes to every image in our text-to-image dataset, resulting in a
set of synthetic captions which we use for subsequent experiments. | Improving Image Generation with Better Captions |
How we use digital media is up to us individually and collectively (even if we
do not use them under conditions of our own choosing). In deeply divided,
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
158
Rasmus Kleis Nielsen & Richard Fletcher | Social_Media_and_Democracy |
M
o
d
e
l
w
e
i
g
h
t
s
:
W
e
h
a
v
e
r
e
a
c
h
e
d
o
u
t
t
o
M
e
t
a
t
o
o
b
t
a
i
n
g
u
i
d
a
n
c
e
o
n
r
e
l
e
a
s
i
n
g
t
h
e
A
l
p
a
c
a
m
o
d
e
l
w
e
i
g
h
t
s
,
b
o
t
h
f
o
r
t
h
e
7
B
A
l
p
a
c
a
a
n
d
f
o
r
| Stanford alpha CRFM |
Results for CoS-E v1.0 are once again in Figure 7
(and for other datasets in Figure 12 in Appendix B).
In Figure 7 (left), we find that removing top-k%
tokens by a(X)R magnitude degrades label perfor-
mance compared to the baseline (green vs. blue
line). Intuitively, this drop is less than token attri-
butions by a(X)L... | Measuring Association Between Labels and Free-Text Rationales |
date-miningMake up a word that means "when two AI researchers go on a date".Model inputMake up a word that means "when two AI researchers go on a date".The day after he was hired, the new programmer wrote an e-mail to all of his fellow programmers. It said, "I will be on vacation next week."The day after he was hired, ... | Scaling Instruction-Finetuned Language Models |
tions through a simplified action space, mim-
icking human-like interactions such as tapping
and swiping. This novel approach bypasses
the need for system back-end access, thereby
broadening its applicability across diverse apps.
Central to our agent’s functionality is its in-
novative learning method. The agent learns... | AppAgents |
Controllable Generation. Current works treat the hallucination level as a controllable attribute
in order to remain the hallucination in outputs at a low level. Controllable generation techniques
such as controlled re-sampling [152], control codes that can be provided manually [50, 152, 210],
or predicted automatically... | SurveyofHallucinationinNatural Language Generation |
of BiomedGPT. However, due to the limited volume and diversity of existing biomedical datasets, as well
as concerns regarding data privacy and imbalance, we plan to explore data augmentation and synthetic
biomedical datasets in our future work.
Table 4: Intra- & inter-distribution transfer performance on VQA tasks in t... | BiomedGPT |
POSTED IN:
https://blog.google/technology/ai/bard-google-ai-search-updates/
3/4
T
h
e
K
e
y
w
o
r
d
H
e
l
p
i
n
g
d
e
v
e
l
o
p
e
r
s
i
n
n
o
v
a
t
e
w
i
t
h
A
I
B
o
l
d
a
n
d
r
e
s
p
o
n
s
i
b
l
e
A
M
e
s
s
a
g
e
F
r
o
m
O
u
r
C
E
O
A
I
A
b
o
u
t
t
h
e
a
u
t
h
o
r
02/05/2023, 15:33
Go... | Google AI updates_ Bard and new AI features in Search |
One possible reason is that our biped cartoon meshes own
an extremely larger amount of vertices than SMPL to model
more complex geometry. This greatly increases the challenge
of vertices regression in Mesh-Graphormer and DecoMR.
Thus, in our setting, directly regressing the low-dimension
parameters performs better. | RaBit- Parametric Modeling of 3D Biped Cartoon Characters with a Topological-consistent Dataset |
3
Figure 2: DocLLM model architecture with disentangled spatial attention and infilling objective. left: Input document
with text tokens xi and bounding boxes bi. Some text segments are randomly masked (two segments here) and the
model predicts the tokens in these text segments autoregressively. right: The infilling ... | DOCLLM |
users and LLMs. For instance, InternLM-XComposer
can compose an article with accompanying illustrations.
SEED-LLaMA [21] leverages LLaMA [65] and diffusion
models for image understanding and question answering,
as well as image generation. The NExT-GPT model [71],
which is most relevant to our proposed framework, fa-
c... | M2UGen |
7.1 Limitations | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
Angshuman Parashar, Minsoo Rhu, Anurag Mukkara,
Antonio Puglielli, Rangharajan Venkatesan, Brucek
Khailany, Joel Emer, Stephen W Keckler, and
William J Dally. 2017. Timeloop: A systematic ap-
proach to dnn accelerator evaluation. In 2017 IEEE
International Symposium on Performance Analysis
of Systems and Software (ISPA... | LLM in a flash |
• Who is [Wolfgang Schulz]?
Wolfgang Schulz is a person I met recently who has been
helping me with my research paper on the effects of gentri-
fication on low-income communities. We had several con-
versations about it, he gave me some great resources, and he
even invited me to join him and another friend for lunch. H... | Generative Agents- Interactive Simulacra of Human Behavior |
In practice, however, we only simulated running live and instead submitted our solutions after the
contest already ended. One consequence of this is that we do not fully consider the “hacking” phase
of the competition, where competitors can get points for finding vulnerabilities in the code of others.
This is only a fac... | alphacode |
4.2 Downstream Tasks
We evaluate all models on a variety of downstream
tasks. In all cases, we consider a prompted zero-
shot setup – i.e., models are instructed to solve
each task in natural language, but we do not pro-
vide any in-context examples. This is in contrast
to prior work on tool use (e.g., Gao et al., 2022... | Toolformer |
3.2 SCALING
In Figure 11 we show that our base video model strongly benefits from scaling up the parameter
count of the video U-Net. We performed this scaling by increasing the base channel count and depth
of the network. This result is contrary to the text-to-image U-Net scaling results by Saharia et al.
(2022b), whic... | IMAGEN VIDEO- HIGH DEFINITION VIDEO GENERATION WITH DIFFUSION MODELS |
(not (holding ?ob))))
(:action stack
:parameters (?ob ?underob)
:precondition (and (clear ?underob) (holding ?ob))
:effect (and (arm-empty) (clear ?ob) (on ?ob ?underob)
(not (clear ?underob)) (not (holding ?ob))))
(:action unstack
:parameters (?ob ?underob)
:precondition (and (on ?ob ?underob) (clear ?ob) (arm-em... | LLM+P- Empowering Large Language Models with Optimal Planning Proficiency |
We train our model on a newly collected dataset,
TEXT2MUSIC, with 50K text-music pairs covering
diverse music genres. Remarkably, our diffusion-
based model improves significantly on previous
models, as it can be trained on a variety of mu-
sic genres, generate long-context music for several
minutes with a high quality... | MOUSAI |
izedtobethesamevolumesizeasthegeneratedvideosforeachmodel.WealsoreporttheresultsofLDMandLFDMunderthe64×64resolution.Forsampling,weuse1000-stepDDPMforLDMandLFDM.SinceDDPMsamplingisveryslowinthelargelatentspaceofVDM(40×64×64×3),weemploy200-stepDDIM[68]toacceleratethesamplingprocess.Wefindthatthismethodcanachievecomparable... | Conditional Image-to-Video Generation with Latent Flow Diffusion Models |
Figure 4. VQA Examples. REVEAL is able to use knowledge
from different sources to correctly answer the question. We show
more examples in Figure 1-3 of Supplementary Material, indicating
that our model can retrieve and use items from diverse knowledge
sources to correctly solve different input query.
(3) Can we add kn... | REVEAL-Retrieval-AugmentedVisual-LanguagePre-Trainingwith Multi-SourceMultimodalKnowledgeMemory |
The Future of Music: How Generative AI Is Transforming the Music Industry | Andreessen Horowitz
In this post, we’ll dive into what people are doing today, explore where AI music might be headed,
and outline a few of the emerging companies and capabilities at the forefront of the space. This
TA B L E O F C O N T E ... | The Future of Music_ How Generative AI Is Transforming the Music Industry _ Andreessen Horowitz |
Mandy Guo, Joshua Ainslie, David Uthus, Santiago
Ontanon, Jianmo Ni, Yun-Hsuan Sung, and Yinfei
Yang. 2022. LongT5: Efficient text-to-text trans-
former for long sequences. In Findings of the Associ-
ation for Computational Linguistics: NAACL 2022,
pages 724–736, Seattle, United States. Association
for Computational Lin... | Unleashing Infinite-Length Input Capacity for Large-scale Language Models with Self-Controlled Memory System |
Evaluation Methodology. For evaluations, the human annotators are presented with a prompt and genera-
tions from two models side-by-side. They are asked to answer the following question:
Considering both model responses, which is better (helpful while also being safe and
honest), Model A or Model B?
The annotators answ... | Llama2 |
• Response: Pros - 1) Increased flexibility in work
schedule, allowing for better job satisfaction with
fewer commuting obligations 2) Improved workplace
culture by offering opportunities to connect from an-
other location or country leading off site; this can lead
to more diverse employees 3) Reduced turnover rate
as c... | LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions |
Hidden
1024
2048
3072
Att. Head Enc. Layer Dec. Layer
4
8
12
4
4
6
4
4
6
the image describe?” to differentiate image classification. “What is the summary of text ‘{Text}’?” and
“Can text1 ‘{Text1}’ imply text2 ‘{Text2}’?” are exploited for text summarization and natural language
inference, respectively.
2.4 Gnera... | BiomedGPT |
From this model, we eliminated all items with loadings below 0.40 and those that were loaded on multiple
factors. We merged items with high similarity as a final step. The scale encompassed fourteen items distributed in
two factors; seven items per factor. The model had a good fit, KMO 𝑀𝑆𝐴 = 0.85, Tucker Lewis Index... | Society’sAttitudesTowardsHumanAugmentation |
of the fundamental mechanisms of thought. Basic Books.
[43] James D. Hollan, Edwin L. Hutchins, and Louis Weitzman. 1984. STEAMER: An
Interactive Inspectable Simulation-Based Training System. AI Magazine 5, 2
(1984), 23–36.
[44] S Holm. 1979. A simple sequentially rejective multiple test procedure. Scandi-
navian Jo... | Generative Agents- Interactive Simulacra of Human Behavior |
in Student & Registry Services.
Undergraduate Affiliate Applicants
5 Undergraduate affiliate applicants must be students registered at other Higher Education
Institutions.
6. Undergraduate affiliate applicants should have completed at least two years’ study at university
and be fully enrolled at their home ... | UCL Academic Manual |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.