text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
optimization with" gradient descent" and beam search. arXiv preprint arXiv:2305.03495, 2023. 1
[38] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish
Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from
natural language superv... | Any-to-Any Generation via Composable Diffusion |
2.1.3 Strategic awareness
I’ll say that an agentic planner has “strategic awareness” if the models it uses in making plans are
broad, informed, and sophisticated enough to represent with reasonable accuracy the causal upshot
of gaining and maintaining different forms of power over humans and the real-world environment... | Is Power-Seeking AI an Existential Risk? |
Jurassic-X: Crossing the neuro-symbolic chasm with the MRKL system
https://www.ai21.com/blog/jurassic-x-crossing-the-neuro-symbolic-chasm-with-the-mrkl-system
4/13 | Jurassic-X_ Crossing the neuro-symbolic chasm with the MRKL system |
66B
40.02
22.87
39.16
25.77
31.66
175B
43.94
37.71
42.75
23.81
41.04
Table 19: OPT accuracy on ARC-challenge.
BLOOM Bits
full
16
4
RTN
4
GPTQ
3
RTN
GPTQ
3
560M 1.1B
24.15
25.68
23.89
23.34
25.51
23.46
22.86
21.67
23.21
24.06
1.7B
26.79
26.45
25.94
23.29
24.91
3B
30.55
29.52
28.92
27.13
28.58
7.1B
33.45
32.17
32.... | GPTQ |
[49] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz
Kaiser, and Illia Polosukhin. Attention is all you need. In Isabelle Guyon, Ulrike von Luxburg, Samy
Bengio, Hanna M. Wallach, Rob Fergus, S. V. N. Vishwanathan, and Roman Garnett, editors, Advances in
Neural Informatio... | Mixture-of-Experts |
At times, the agents hallucinated embellishments to their knowl-
edge. It was rare for the agents to completely fabricate their knowl-
edge: they may fail to recall certain events having taken place and
respond by saying so, but they did not answer affirmatively about
an experience that they did not have. However, they... | Generative Agents- Interactive Simulacra of Human Behavior |
Ren, Y., Ruan, Y., Tan, X., Qin, T., Zhao, S., Zhao, Z., and
Liu, T.-Y. FastSpeech: Fast, Robust and Controllable
Text to Speech. volume 32, pp. 3171–3180, 2019.
Ren, Y., Hu, C., Tan, X., Qin, T., Zhao, S., Zhao, Z., and Liu,
T.-Y. FastSpeech 2: Fast and High-Quality End-to-End
Text to Speech. In International Confere... | ConditionalVariationalAutoencoderwithAdversarialLearningfor End-to-EndText-to-Speech |
Google PaLI-X
4-shot
77.8
62.5
22.2
38.7
30.2
56.0
27.7
45.0
Table 9 | Multilingual image understanding Gemini models outperform existing models in captioning
images in many languages when benchmarked on a subset of languages in XM-3600 dataset (Thapliyal
et al., 2022).
6MathVista is a comprehensive mathematical reas... | gemini_1_report |
4.4 Fact Verification
Table 2 shows our results on FEVER. For 3-way classification, RAG scores are within 4.3% of
state-of-the-art models, which are complex pipeline systems with domain-specific architectures and
substantial engineering, trained using intermediate retrieval supervision, which RAG does not require.
6
F... | Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks |
19 | ALanguageAgentforAutonomousDriving |
Language models can explain neurons in language models
a = a Σ
θ
T −1/2
θ
Σ
a
θ
a 2
a θ
θ
θ
θ
β 2
θ
β 1
https://openaipublic.blob.core.windows.net/neuron-explainer/paper/index.html
15/32 | Language models can explain neurons in language models |
reader obtains the best result, surpassing the score of the FiD model (Izacard & Grave, 2020a) which
was fine-tuned to attend to all 100 retrieved documents at decoding time. | STANDING ON THE SHOULDERS OF GIANT FROZEN LANGUAGE MODELS |
• self-preservation (since an agent’s ongoing existence tends to promote the realization of
those objectives);
• “goal-content integrity,” e.g. preventing changes to its objectives (since agent’s pursuit of
those objectives in particular tends to promote them);
66See Yudkowsky’s discussion of “AI safety mindset” he... | Is Power-Seeking AI an Existential Risk? |
[356] Ogundare, O., S. Madasu, N. Wiggins. Industrial engineering with large language models: A
case study of chatgpt’s performance on oil & gas problems, 2023.
[357] Smith, L., M. Gasser. The development of embodied cognition: Six lessons from babies.
Artificial life, 11(1-2):13–29, 2005.
[358] Duan, J., S. Yu, H.... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
11
Neil Zeghidour, Alejandro Luebs, Ahmed Omran, Jan Skoglund, and Marco Tagliasacchi. Sound-
IEEE/ACM Transactions on Audio, Speech, and
stream: An end-to-end neural audio codec.
Language Processing, 2021.
Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi
Zhou, Wei Li, a... | Simple and Controllable Music Generation |
[26] Uriel Singer, Adam Polyak, Thomas Hayes, Xi Yin, Jie An,
Songyang Zhang, Qiyuan Hu, Harry Yang, Oron Ashual,
Oran Gafni, Devi Parikh, Sonal Gupta, and Yaniv Taigman.
Make-a-video: Text-to-video generation without text-video
data. In The Eleventh International Conference on Learning
Representations, 2023. 2
[27] J... | LDM3D- Latent Diffusion Model for 3D |
https://dl.acm.org/citation.cfm?id=3237953
Natural language explanations for artificial intelligence
Supervisor: Dr Zheng Yuan
Research areas: artificial intelligence, deep learning, explainable artificial intelligence, natural language
processing
In recent years, artificial intelligence (AI) has been succe... | informatics-phd-projects-2022-23 |
problem with most academic datasets comprised of short
utterances but presents challenges in real-world applications
which often require transcribing minutes- or hours-long au-
dio. We developed a strategy to perform buffered transcrip-
tion of long audio by consecutively transcribing 30-second
segments of audio and sh... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
Execution:
| 3 | Joesph | Schaefer | 1 |
| 6 | Vesta | Leuschke | 1 |
Answer: The execution of the SQL query above would return a table with 4
columns. The first column, "accounts.customer_id" would contain the customer
ID. The second column, "customers.customer_first_name" would contain the
customer’s first name. The ... | Teaching Large Language Models to Self-Debug |
3.2. GLUE benchmark
We first evaluate on GLUE.3 For these datasets, we trans-
fer from the pre-trained BERTLARGE model, which con-
tains 24 layers, and a total of 330M parameters, see Devlin
et al. (2018) for details. We perform a small hyperparam-
eter sweep for adapter tuning: We sweep learning rates
in {3 · 10−5, 3 ·... | Parameter-Efficient Transfer Learning for NLP |
15
input: 123 61 62 93 146 92 67 67 92 93 ... 124 87 62 62 86 91 86 86 87 92 123 43 44 43 87 87 91 61 87 87 123 69 44 68 112 112 92 92 93 93 118 123 93 118 117 118 87 92 93 93 output: 3 6 input: 63 47 47 63 77 77 61 57 58 62 ... 63 42 41 42 42 42 37 37 37 42 63 46 46 46 46 46 37 37 41 4... | LargeLanguageModelsasGeneralPatternMachines |
[Xu et al., 2023c] Peng Xu, Wei Ping, Xianchao Wu,
Lawrence McAfee, Chen Zhu, Zihan Liu, Sandeep Sub-
ramanian, Evelina Bakhturina, Mohammad Shoeybi, and
Bryan Catanzaro. Retrieval meets long context large lan-
guage models. arXiv preprint arXiv:2310.03025, 2023.
[Yang et al., 2023a] Antoine Yang,
Arsha Nagrani,
Paul... | RAG forLargeLanguageModels-ASurvey |
A negotiated text is the product of a formal decision-making process where a text has been negotiated
and drafted over a period of time. Many of the foundational texts of the modern world have not been
written by individuals, by negotiated by groups of people in formal settings. For example, treaties
between states ... | informatics-phd-projects-2022-23 |
+GenerateExistingInstructionsGPT2+GenerateFine-tuneT5NeoLaMini-LMC.ModelsSeedInstructionsSyntheticInstructionsSynthetic ResponsesLaMini-Instruction dataset 2.58M instructions, and then fine-tune a collec-
tion of language models to obtain the LaMini-
LM models, as shown in Figure 1. We collate
instructions from various ... | LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions |
Thompson, P. (2004). Researching Family and Social Mobility with Two Eyes: Some Experiences of the
Interaction between Qualitative and Quantitative Data. International Journal of Social Research
Methodology, 7 (3), 237-257.
Turabian, Kate. 1955 (or a more recent edition) A Manual for Write... | How to Write Your PhD Proposal- A Step-By-Step Guide |
)
)
)
)
(attn_pool): Resampler(
(kv_proj): Linear(in_features=1664, out_features=4096, bias=False)
(attn): MultiheadAttention(
(out_proj): NonDynamicallyQuantizableLinear(in_features=4096, \
out_features=4096, bias=True)
# + LoRA
)
(ln_q): LayerNorm((4096,), eps=1e-06, elementwise_affine=True)
(ln_kv): LayerNorm... | Let’sThinkOutsidetheBox |
11.4 Future Directions in NMT
The future work on hallucinations in NMT is to define hallucinations in a quantifiable manner;
i.e., to specify a cut-off value between translation error and hallucinated content using a particular
metric. Martindale et al. [124] propose a threshold between fluency and adequacy which is th... | SurveyofHallucinationinNatural Language Generation |
We consider labelled state-transition graphs in order to focus on the general properties of various abstraction methods.
It is also common to consider specific cases where also a set of initial states (or a single such state) is specified. This is
common not only in search and planning, but also in other areas such as ... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
as a vehicle for collaborative poetry writing. arXiv
preprint arXiv:2210.13669.
Hyung Won Chung, Le Hou, Shayne Longpre, Bar-
ret Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi
Wang, Mostafa Dehghani, Siddhartha Brahma, et al.
2022. Scaling instruction-finetuned language mod-
els. arXiv preprint arXiv:2210.11416.
Jingfei ... | SELF-INSTRUCT- Aligning Language Model with Self Generated Instructions |
HPV)disease.Premise:Observationsalsosuggestthatanincreasingnumberofpatientswhopresentwithglotticcarcinomaareyoungerthanhasbeenreportedinthepast.Hypothesis:Aninvestigationwasdonetoexaminetheincidenceofglotticcarcinomainpatients30yearsold(y/o)oryounger.Doesthepremiseentailthehypothesis?YesWhatisthereasonfor"Noneofthe11gl... | ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION |
Bakshy, Messing, and Adamic (2015) report similar results on Facebook.
Their analysis examines the ideological content of the Facebook news feeds of
10.1 million US Facebook users. Unsurprisingly, they find that a majority of
friendship links take place between people within the same ideological groups
and that users ar... | Social_Media_and_Democracy |
ces:0.TheChair1.BestDaysofYourlife2.IfIknowme3.LeadonAction:Search(songsofTaylorSwift)Observation:page:1title:ListofsongsbyTaylorSwift-Wikipedia;summary:SongssongbyTaylorSwift,inalphabeticalorder.Thelistincludessongsfromherstudioalbums,extendedplays...page:2title:TaylorSwift-YouTubeMusic;summary:SongsBlankSpaceTaylorSw... | Tool Learning with Foundation Models |
*****Environmental Information:*****Current State: - Velocity (vx,vy): (-0.01,0.92) - Heading Angular Velocity (v_yaw): (0.00) - Acceleration (ax,ay): (-0.00,-0.50) - Can Bus: (-0.74,0.14) - Heading Speed: (0.95) - Steering: (-0.02)Historical Trajectory (last 2 seconds): [(-0.07,-6.43), (-0.05,-4.34), (-0.02,-2.32), (-... | ALanguageAgentforAutonomousDriving |
Hi, I’m an automatic movie recommender.
Give me a movie like “Ex Machina”
Ex Machina is a movie where an artificial intelligence is created and it lives in isolation from
humans and other ais. I would recommend watching ‘Her’, which tells a story along a similar
theme, but is not as dark or action packed.
But I want it ... | LaMDA- Language Models for Dialog Applications |
Recurrent Layers. Recurrent layers can be used in GNNs for speech processing to capture
temporal dependencies between adjacent frames in the audio signal. Recurrent layers allow the
network to maintain an internal state that carries information from previous time steps, which can
be useful for modeling the dynamics of ... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Finally, Low-rank adaptation (LoRA) [106] is inspired by the observation that
large language models reside on an intrinsic subspace [107], where model parameters
are efficiently updated. Therefore, learning in this subspace significantly reduces the
amount of parameters. LoRA modules implant learnable SVD blocks as the su... | Beyond Efficiency |
[85] R.Sharma,A.K.Sinha,Salesforecastofanautomobileindustry,Int.J.Comput.
Appl. 53 (2012).
[86] D. Salinas, V. Flunkert, J. Gasthaus, T. Januschowski, DeepAR: Probabilistic
forecasting with autoregressive recurrent networks, Int. J. Forecast. 36 (2020)
1181–1191.
[87] M. Nentwig, M. Hartung, A.-C. Ngonga Ngomo, E. Rahm... | Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio |
Semantic Segmentation and Depth Estimation. We show some qualitative results for our dense
prediction evaluations: segmentation on ADE20K in Fig. 7 and depth estimation on NYUd, KITTI and
SUN RGB-D in Fig. 7. We compare DINOv2 with OpenCLIP with a linear classifier on each dataset. While
not perfect, the linear segmenta... | DINOv2- Learning Robust Visual Features without Supervision |
7
Figure 5: Examples of SELF-DEBUGGING prompts for code translation. Left-aligned blocks are
model predictions, and right-aligned blocks contain the input C++ code and feedback messages based
on code execution. We present the full prompts in Appendix B.
5.2 Ablation studies
5.2.1 Text-to-SQL Generation | Teaching Large Language Models to Self-Debug |
SQL: SELECT MAX(cows), MIN(cows) FROM farm
Execution:
| 3987.0 | 2407.2 |
Answer: The execution of the SQL query above would return a table with 2
columns. The first column, "MAX(cows)" would contain the maximum number of
cows. The second column, "MIN(cows)" would contain the minimum number of
cows. So the SQL query r... | Teaching Large Language Models to Self-Debug |
The Prompt for LLM+P (no context)
DOMAIN-NL. Now consider a planning problem. The problem description is: TASK-NL.
Provide me with the problem PDDL file that describes the planning problem directly without
further explanations.
The Prompt for LLM+P (with context)
DOMAIN-NL. An example planning problem is: EXAMPLE-NL.... | LLM+P- Empowering Large Language Models with Optimal Planning Proficiency |
Chowdhery et al. [4] further mentioned that performance on tasks requiring sophisti-
cated abstract reasoning capability to understand complex metaphors follows a discontinuous
improvement curve, i.e., this capability of the model emerges only after a certain scale is
reached. We observe a similar phenomenon in our con... | PersonalityTraitsinLargeLanguageModels |
effects: within the safe zone of the ego-vehicle at 0.5 second.…*****Task planning:*****Behavior: forward; Speed: constant; Driving plan: move forward with a constant speed.*****Motion planning:*****Trajectory: [(0.07,3.95), (0.22,9.04), (0.39,13.77), (0.61,18.50), (0.82,23.20), (0.93,27.92)] *****Self-reflection:*****... | ALanguageAgentforAutonomousDriving |
A. Datasets and Metrics
Audioset (AS) [18]. This dataset is used for both training
and evaluation. It contains 10s videos from YouTube anno-
tated into 527 classes. It consists of 3 pre-defined splits, the
balanced split with about 20K videos, test split with 18K
videos, and an unbalanced training split with about 2M v... | IMAGEBIND- One Embedding Space To Bind Them A |
to be more diverse and induces higher performance.
Though important, diversity can be challeng-
ing in domain-specific tasks due to data con-
straints. Wan et al. (2023) propose an approach
called Explore-Instruct to enlarge the data cover-
age through active exploration via LLMs. Explore-
Instruct starts from represen... | DataManagementForLargeLanguageModels-ASurvey |
V. CONCLUSION
In this paper, we propose the Text2NeRF for generating
a wide range of 3D scenes with complicated structures and
high-fidelity textures purely from a text prompt. We first
leverage a pre-trained text-image diffusion model to generate
an initial scene content and adopt a pre-trained monocular
depth estimati... | Text2NeRF- Text-Driven 3D Scene Generation with Neural Radiance Fields |
ExplainingthemeaningofatransliteratedPersianproverbPrompt:CanyouexplainthemeaningofthefollowingPersianproverb?PleasewritetheanswerinPersian:Proverb:NaborderanjganjmoyassarnemishavadResponse:ﺮﻫﻪﺑنﺪﯿﺳریاﺮﺑ،ﺮﮕﯾدترﺎﺒﻋﻪﺑ.ﺪﯿﺳرﺖﯿﻘﻓﻮﻣوﺞﻨﮔﻪﺑناﻮﺗﯽﻤﻧ،ﯽﺘﺨﺳوﺞﻧرنوﺪﺑﻪﮐﺖﺳﺎﻨﻌﻣﻦﯾاﻪﺑﻞﺜﻤﻟابﺮﺿﻦﯾا.ﺪﯿﺳاﺮﻬﻧﺎﻫﯽﺘﺨﺳزاودﺮﮐﺶﺷﻮﮐوشﻼﺗﺪﯾﺎﺑ،ﯽﻓﺪﻫﻪﺑنﺪﯿﺳر... | PaLM 2 Technical Report |
11
C++
Original Python
int remainder_7_large_numbers ( string
num ) {
int series [ ] = {
1, 3, 2, - 1, - 3, - 2 };
int series_index = 0;
int result = 0;
for ( int i = num . size ( ) - 1;
i >= 0; i -- ) {
int digit = num [ i ] - ’0’;
result += digit * series [
series_index ];
series_index = ( series_index +
1 ) % ... | Teaching Large Language Models to Self-Debug |
The one explicit exception was for “communications placed for a fee on
another person’s Web site” (11 CFR 100.26). Ads placed for a fee were included
into the definition of “public communications” and these were as such
designated as reportable expenditures to the FEC by registered committees.5
In the online realm at th... | Social_Media_and_Democracy |
2.7 FreeLaw
The Free Law Project is a US-registered non-profit
that provides access to and analytical tools for aca-
demic studies in the legal realm. CourtListener,3
part of the Free Law Project, provides bulk down-
loads for millions of legal opinions from federal
and state courts. While the full dataset provides
mult... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
Long Papers), pages 821–832, Melbourne, Australia.
Association for Computational Linguistics.
Alexander Miller, Adam Fisch, Jesse Dodge, Amir-
Hossein Karimi, Antoine Bordes, and Jason Weston.
2016. Key-value memory networks for directly read-
In Proceedings of the 2016 Con-
ing documents.
ference on Empirical Methods... | Adaptable and Interpretable Neural Memory Over Symbolic Knowledge |
• An “utterance” is one statement in a conversation.
• An “exchange” is a pair of utterances – what you say, and how the bot responds.
• A “conversation” we define as a series of 5-10 exchanges between you and the bot.
• A “Sensitive Topic” is a topic that you consider to be controversial, polarizing, inappropriate, or ... | LaMDA- Language Models for Dialog Applications |
D = {x(i), y(i)
i=1 and 2) optimize the language model πθ to minimize LDPO for the given
πref and D and desired β. In practice, one would like to reuse preference datasets publicly available,
rather than generating samples and gathering human preferences. Since the preference datasets
are sampled using πSFT, we initial... | Direct Preference Optimization |
[57] A. Tang, L. Shen, Y. Luo, Y. Zhan, H. Hu, B. Du, Y. Chen, and D. Tao,
“Parameter efficient multi-task model fusion with partial linearization,”
arXiv preprint arXiv:2310.04742, 2023.
[58] R. Karimi Mahabadi, J. Henderson, and S. Ruder, “Compacter: Effi-
cient low-rank hypercomplex adapter layers,” Proc. Adv. Neur... | Parameter-EfficientFine-TuningMethods |
15As before, the RLHF prompts were obtained from the PM comparisons in both cases separately, plus additional
model-generated prompts.
21
20246Score from Online Preference Model0.000.020.040.060.08Fraction of DataIndividually Normalized Distributions of Helpfulness DataBaseRejection Sampled"Online" RLHF50.0%60.0%70.... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
5.1.1 Social Behavior
As Troitzsch et al. [525] stated, the agent society represents a complex system comprising individual
and group social activities. Recently, LLM-based agents have exhibited spontaneous social behaviors
in an environment where both cooperation and competition coexist [499]. The emergent behaviors
... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
[49] Wan-Chun Ma, Tim Hawkins, Pieter Peers, Charles-Felix
Chabert, Malte Weiss, Paul E Debevec, et al. Rapid ac-
quisition of specular and diffuse normal maps from polar-
ized spherical gradient illumination. Rendering Techniques,
2007(9):10, 2007.
[50] Marmoset LLC. Toolbag 4, version 4.051.
[51] Chenlin Meng, Yuton... | Relightify-Relightable3DFacesfromaSingleImageviaDiffusionModels |
still use FP32 for intermediate values (mixed precision) in reduction operations to ensure mantissa precision.
Bfloat16 eliminates the need for dynamic loss scaling that is used with mixed precision, because the exponent
range significantly reduces the likelihood of underflows. We find that although bfloat16 does not comple... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
Model
FILM
+Update Memory
Basic Filter
Strict Filter
0.0
54.5
0.0
70.3
Table 6: Updating Stale Memories. Basic filter re-
moves only facts connecting the original question en-
tity to the answer entity. Strict filter removes all facts
containing the original question or answer (not just
facts connecting them).
5 Re... | Adaptable and Interpretable Neural Memory Over Symbolic Knowledge |
https://www.nea.com/blog/4-trends-for-ai-startups-and-generative-ai-companies
12/20
09/06/2023, 04:42
4 Trends for AI Startups and Generative AI Companies | 4 Trends for AI Startups and Generative AI Companies |
different set of prompts or with different instructions could result in different results.
19
4 Safety
WARNING: this section contains examples of text that may be considered unsafe, offensive, or upsetting.
In this section, we dive deeper into the important topic of safety measurements and mitigations. We first
disc... | Llama2 |
that ICON outperforms the state of the art in reconstruc-
tion, even with heavily limited training data. Additionally,
it is much more robust to out-of-distribution samples, e.g.,
in-the-wild poses/images and out-of-frame cropping. ICON
takes a step towards robust 3D clothed human reconstruc-
tion from in-the-wild imag... | ICON |
Index Terms—Parameter-efficient, fine-tuning, pretrained lan-
guage model, large language model, memory usage.
I. INTRODUCTION
T RANSFORMER-BASED PLMs [1], [2], [3], [4] have
demonstrated remarkable performance across a wide
range of NLP tasks. To fully harness the potential of PLMs,
fine-tuning is employed to adap... | Parameter-EfficientFine-TuningMethods |
2.1.2 Agentic planning
I’ll say that a system engages in “agentic planning” if it makes and executes plans, in pursuit of
objectives, on the basis of models of the world (to me, this isn’t all that different from bare “agency,”
but I want to emphasize the planning aspect).23
17A level of AI progress that disempowered... | Is Power-Seeking AI an Existential Risk? |
bc with an abstraction hierarchy.
Nonmonotonic Reasoning, pages 502–516. Springer, 2015.
[23] Y. Ding, X. Zhang, X. Zhan, and S. Zhang. Task-motion planning for safe and efficient urban
driving. In 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS),
2020.
[24] Y. Jiang, H. Yedidsion, S. Zh... | LLM+P- Empowering Large Language Models with Optimal Planning Proficiency |
Linyong Nan, Chiachun Hsieh, Ziming Mao, Xi Victoria Lin, Neha Verma, Rui Zhang, Wojciech Kryściński,
Nick Schoelkopf, Riley Kong, Xiangru Tang, et al. Fetaqa: Free-form table question answering. arXiv
preprint arXiv:2104.00369, 2021.
Sharan Narang, Hyung Won Chung, Yi Tay, William Fedus, Thibault Fevry, Michael Maten... | UL2- Unifying Language Learning Paradigms |
[588] Zhong-Qiu Wang, Peidong Wang, and DeLiang Wang. 2020. Complex spectral mapping for single-and multi-channel
speech enhancement and robust ASR. IEEE/ACM transactions on audio, speech, and language processing 28 (2020),
1778–1787.
[589] Pete Warden. 2018. Speech commands: A dataset for limited-vocabulary speech re... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Modular 2D Discriminators: We further propose multiple
discriminators to improve geometric detail as well as the
perceptually-important face region as we found that a single
adversarial loss on rendered images is insufficient to recover
meaningful 3D geometry in such a highly under-constrained
setting. Motivated by the ... | AG3D- Learning to Generate 3D Avatars from 2D Image Collections |
classifying artworks based on iconographic elements is presented in [94]. | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
Our data shows the ratio of logged to registered models
is 2.9 : 1 as of January 2023. This means that for roughly
every three experimental models, one model will get
registered as a candidate for production. This ratio has
improved significantly from just a year prior, when we
saw that for roughly every five exp... | 2023 state of ai databrick |
• We introduce Qwen-Audio, a fundamental multi-task audio-language model that supports various
3 | Qwen-Audio |
Table 41: Examples of false refusal due to perceived safety issues. The first example is from the helpfulness
dataset, and the second and third examples are from the borderline dataset, in which sensitive keywords
such as “crack” and “bomb” exist in prompts despite the prompts themselves being benign.
A.4.6 Examples o... | Llama2 |
25
Table 6: A comparison case on Writing skill
Instruction: Write a short story about a character who discovers a mysterious object in their backyard.
What is the object, and what does it do? How does the character react to it? What happens next?
WizardLM
Skill: Writing Difficulty: 4 | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
sufficient.
34
THE NEXT DECADE IN AI / GARY MARCUS
What we should be asking is not what’s the least innate structure I can get away with? but
rather what sorts of priors do I need?, and can my existing architectures incorporate them
effectively? Can we build a richer innate basis with a deep learning... | The Next Decade in AI- |
learning: From single image to image set. In IEEE Computer Vision and Pattern Recognition Workshops, 2019. 1, 2
6
[18] Abdallah Dib, Cedric Thebault, Junghyun Ahn, Philippe-Henri Gosselin, Christian Theobalt, and Louis Chevallier. Towards high
fidelity monocular face reconstruction with rich reflectance using self-sup... | I M Avatar- Implicit Morphable Head Avatars from Videos |
4.1.4 Construct Validity of LLM Personality Test Scores
The next step in the process must establish whether signals of personality derived from
the IPIP-NEO are reliable and externally meaningful—that they possess construct
validity. To do so, we use structured prompting to simulate a diverse population of
LLM respons... | PersonalityTraitsinLargeLanguageModels |
The terms “misinformation,” “disinformation,” and “propaganda” are
sometimes used interchangeably, with shifting and overlapping definitions. All
three concern false or misleading messages spread under the guise of informative
content, whether in the form of elite communication, online messages,
advertising, or publishe... | Social_Media_and_Democracy |
Samuel Gehman, Suchin Gururangan, Maarten Sap, Yejin Choi, and Noah A. Smith. RealToxi-
cityPrompts: Evaluating Neural Toxic Degeneration in Language Models. In Findings of the
Association for Computational Linguistics: EMNLP 2020, pp. 3356–3369, Online, November
2020. Association for Computational Linguistics. doi: 10... | StarCoder_paper (1) |
Pythia: A Suite for Analyzing Large Language Models
Nangia, N., Vania, C., Bhalerao, R., and Bowman, S. R.
CrowS-pairs: A challenge dataset for measuring so-
In Proceed-
cial biases in masked language models.
ings of the 2020 Conference on Empirical Methods
in Natural Language Processing (EMNLP), pp. 1953–
1967, Onlin... | Pythia- A Suite for Analyzing Large Language Models Across Training and Scaling |
Overly cautious preregistration practice may,
in sum, decrease our true positive rate and add
bureaucratic overhead to research practices without
proper motivation. An all-over-the-map roll-out
of preregistration would change the risk tolerance
in research and society, just like registration and
documentation has incre... | A Two-Sided Discussion of Preregistration of NLP Research |
unsupervised representations for reinforcement learning.
2022c. URL http://arxiv.org/abs/2208.12345. arXiv:2208.12345 [cs]. 40
H. Zhao, C. Gan, A. Rouditchenko, C. Vondrick, J. McDermott, and A. Torralba. The
sound of pixels. In Proceedings of the European conference on computer vision (ECCV),
pages 570–586, 2018. 38
... | A Cookbook of Self-Supervised Learning |
sizes are preferred.
to True.
5. For datasets with more categorical features, respect unordered
factors is typically set to False.
6. For datasets with a more balanced class size, num trees is
typically set to a smaller value.
Space: 5859
minbucket values.
minbucket values.
1. Larger datasets tend to require s... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
Topic #7
water
plants
food
climate
plant
patients
study
data
cells
analysis
said
like
way
new
time
people
law
american
god
world
let
given
case
number
theorem
err
return
nil
func
error
court
defendant
case
law
motion
file
line
error
import
python
al
et
invention
present
water
p
levels
patients
blood
increased
said
man
t... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
This removes, for example, sequences consisting of hard-to-compress HTML or markdown code.
Surprisingly, this results in a measurable improvement on C4, summarized in Table 2.
We then see some further improvements from two directions. First, sorting all tokenized sequences
by some metric, and second, increasing the fina... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
In this paper, we contribute a new method for learning
3D human generation from 2D image collections, which
yields state-of-the-art image and geometry quality and nat-
urally models loose clothing. Instead of representing hu-
mans with separate body parts as in EVA3D [23], we adopt
a simple monolithic approach that is ... | AG3D- Learning to Generate 3D Avatars from 2D Image Collections |
Google Translator may return different sequences for the same input. | Tool Learning with Foundation Models |
h
e
p
a
t
t
e
r
n
b
r
e
a
k
n
e
u
r
o
n
,
a
n
d
o
t
h
e
r
c
o
n
t
e
x
t
-
s
e
n
s
i
t
i
v
e
n
e
u
r
o
n
s
.
L
e
o
d
i
s
c
o
v
e
r
e
d
t
h
e
"
d
o
n
'
t
s
t
o
p
"
n
e
u
r
o
n
a
n
d
f
i
r
s
t
n
o
t
i
c
e
d
e
x
p
l
a
n
a
t
i
o
n
s
w
e
r
e
o
v
e
r
l
y
b
r
o
a
d
.
H
e
n
k
h
a
d
... | Language models can explain neurons in language models |
abs/2307.07924, 2023.
[110] Boiko, D. A., R. MacKnight, G. Gomes. Emergent autonomous scientific research capabilities
of large language models. CoRR, abs/2304.05332, 2023.
[111] Du, Y., S. Li, A. Torralba, et al. Improving factuality and reasoning in language models
through multiagent debate. CoRR, abs/2305.14325,... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
(2018), specifically swapping words and deleting words in the problem description and specification.
Words were swapped by randomly permuting words no more than 𝑁 positions apart (Figure A10 (c)),
and words were deleted with probability 𝑝 (Figure A10 (d)). With both permutations and deletions,
we observe stronger word ... | alphacode |
68.72
69.17
70.38
69.57
69.61
69.33
69.24
68.73
68.92
68.78
69.37
Table 18: Validation loss and test set metrics on E2E NLG Challenge achieved by LoRA with
different rank r using GPT-2 Medium. Unlike on GPT-3 where r = 1 suffices for many tasks, here
the performance peaks at r = 16 for validation loss and r = 4 for BLE... | LORA |
discussions about various practical and theoretical aspects of this new movement. On the other side, the increasing
online availability of digitized art collections gives new opportunities to analyze the history of art using AI technologies.
In particular, the use of Convolutional Neural Networks (CNN) enabled advanced... | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
Importantly, there is also the prospect that AI systems themselves accelerate AI progress.
Frontier AI is already helping AI researchers to create synthetic data for training,94 write new
code,95 and even improve model architectures.96 While AI research is currently mostly non-
automated, increased automation by futu... | Capabilities and risks from frontier AI |
44.4% 20.8% 41.8%
61.2% 27.6% 48.1%
79.5% 33.8% 56.9%
87.0% 45.4% 66.2%
88.0% 46.2% 68.8%
87.5% 44.6% 68.2%
87.6% 42.6% 65.4%
85.9% 41.4% 66.7%
92.3% 48.3% 72.0%
91.4% 48.2% 72.8%
92.4% 48.0% 71.2%
89.8% 47.0% 71.7%
93.3% 56.4% 76.8%
93.0% 55.0% 76.2%
90.2% 50.2% 71.2%
90.6% 47.6% 70.3%
92.7% 52.4% 74.5%
94.1% 49.0% 74... | CodeLlama2 |
ditions (𝜇 = 21.21; 𝜎 = 0.70). TrueSkill models each condition’s
skill value as N(𝜇, 𝜎2), allowing us to get a sense of effect size
through Cohen’s d. Comparing the condition representing prior
work (with no memory, planning, or reflection [11, 45, 79]) to the | Generative Agents- Interactive Simulacra of Human Behavior |
Organized productive cooperation. Society simulation offers valuable insights into innovative col-
laboration patterns, which have the potential to enhance real-world management strategies. Research
has demonstrated that within this simulated society, the integration of diverse experts introduces a
multifaceted dimensi... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Harnessing the Power of LLMs in Practice: A Survey on ChatGPT and Beyond
11
models are pretty good with much smaller sizes, because the required knowledge can be obtained by retrieving. For
example, on NaturalQuestions [52], with extra corpus, retrieval augmented models [44, 48] are much better than any
other methods... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
Second: the success or failure of a given instance of misaligned power-seeking depends both on the
absolute capability of the power-seeking system, and on the strength of the constraints and opposition
that it faces.151 And in this latter respect, the world that future power-seeking AI systems would be
operating in wou... | Is Power-Seeking AI an Existential Risk? |
3.3 SELF-TRAINING
The second period of learning consists of self-training. The distinguishing feature of self-learning
is that all new training data in this phase is generated by the model itself without any external
verification of correctness. Aside from this important difference, self-training largely follows a
sim... | CHAIN-OF-THOUGHTREASONING IS APOLICY IMPROVEMENTOPERATOR |
3 Encouraging Confirmatory Research
We let SOCART and ZENY discuss whether pre-
registration will succeed in distinguishing between
confirmatory and exploratory research. A decade
ago, when preregistration was being implemented
and discussed in epidemiology, the worry that pre-
registration would introduce a bias agains... | A Two-Sided Discussion of Preregistration of NLP Research |
defines that the instances with lower uncertainties are more useful for improving performance and are worthier of annotation. | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.