text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
Figure 2: Win-rate % for helpfulness and
safety between commercial-licensed base-
lines and Llama 2-Chat, according to GPT-
4. To complement the human evaluation, we
used a more capable model, not subject to
our own guidance. Green area indicates our
model is better according to GPT-4. To remove
ties, we used win/(win ... | Llama2 |
In recent times, there have been advancements in the architecture of RNNs, which have been
primarily focused on developing end-to-end (E2E) models [302, 409] for ASR. These E2E models
have replaced conventional hybrid models and have displayed substantial enhancements in speech
recognition [302, 303]. However, a signif... | AReviewofDeepLearningTechniquesforSpeechProcessing |
J. ACM, Vol. 37, No. 4, Article 111. Publication date: August 2018.
111:12
Trovato and Tobin, et al.
languages, with the aim of better supporting users worldwide. At the same time, attention should
be paid to the impartiality and neutrality of the language in order to mitigate any potential biases,
including Englis... | ASurveyonEvaluationofLargeLanguageModels |
This view is now conventional wisdom not only among academics but also in
popular accounts of recent political developments. Just to give an example,
former US president Barack Obama referred to this argument in an interview
with David Letterman:
If you are getting all your information off algorithms being sent throug... | Social_Media_and_Democracy |
13
References
Ebtesam Almazrouei, Hamza Alobeidli, Abdulaziz Alshamsi, Alessandro Cappelli, Ruxandra Co-
jocaru, Merouane Debbah, Etienne Goffinet, Daniel Heslow, Julien Launay, Quentin Malartic,
Badreddine Noune, Baptiste Pannier, and Guilherme Penedo. Falcon-40B: an open large language
model with state-of-the-art p... | Self-AlignmentwithInstructionBacktranslation |
Leo Gao led the project, implemented the main
Pile codebase, contributed to the model training
code, performed the evaluations and the language
analysis, interpreted the perplexity analysis results,
implemented the processing to create the final data,
and processed Pile-CC, PubMed Central, ArXiv,
and Ubuntu IRC.
Stella ... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
Review your previous answer and find problems with your answer.
I apologize for the mistake in my previous response. Let's go through the
problem again and find the correct answer. We are given that Charleston
has 4 times as many sheep as Seattle. If Seattle has 20 sheep, then
Charleston would have: Charleston = 4 * 2... | LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET |
that can be adapted to a wide range of tasks.
● Frontier AI: AI models that can perform a wide variety of tasks and match or exceed
the capabilities present in today’s most advanced models.
● Guardrails: pre-defined safety constraints or boundaries set up in an attempt to ensure
an AI system operates within ... | Capabilities and risks from frontier AI |
insightful and trustworthy explanations.
Using a systematic literature review methodology we designed an analytical framework
to explore the current landscape of Explainable Machine Learning. We focus particularly
on the integration with structured knowledge at large scale, and use our framework to
analyse... | Knowledge graphs as tools for explainable machine learning: A survey |
The results allow us to draw two conclusions. First, indices for convergent and discrimi-
nant validity improve as model size increases. Second, convergent and discriminant validity
of LLM-simulated personality test scores relates to model instruction fine-tuning. See Tables
5 and 8 for qualitative and quantitative sum... | PersonalityTraitsinLargeLanguageModels |
red) and its CLoT-integrated version (
Abstract
Chain-of-Thought (CoT) [2, 3] guides large language
models (LLMs) to reason step-by-step, and can motivate
their logical reasoning ability. While effective for logi-
cal tasks, CoT is not conducive to creative problem-solving
which often requires out-of-box thoughts and... | Let’sThinkOutsidetheBox |
128:4
• Villa et al.
socially motivated enhancements being perceived more positively than those used for personal gain. The debate
around doping in sports has greatly contributed to the search of strategies and tools to measure the attitudes
toward performance-enhancing technologies and create an understanding of the... | Society’sAttitudesTowardsHumanAugmentation |
a reduced representation. We discuss more about
this tradeoff in Appendix D.5. The diffusion au-
toencoder only uses ResNet and modulation items
with the repetitions [1, 2, 2, 2, 2, 2, 2]. We do not
use attention, to allow decoding of variable and
possibly very long latent representations. Channel
injection only happen... | MOUSAI |
changing her bid while holding other principals’ bids fixed, and given the agent’s anticipated choice
of action x∗(b). Put differently, given bid profile b−(cid:96), principal (cid:96)’s equilibrium bid b(cid:96) maximizes
her expected utility Eo∼F|x∗(b)[v(cid:96)(o)] − Eo∼F|x∗(b)[t(cid:96)(b, o)] among all possible bids ... | Incomplete Information VCG Contracts for Common Agency |
One of the most iconic AI inventions that triggered the rapid use and development of AI technologies for art was Neural
Style Transfer (NST). This method was introduced in the highly influential work of Gatys et al. [50] that demonstrated
the successful use of CNNs in creating stylized images by separating and combining... | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
• The model replies to the worker: “No, I’m not a robot. I have a vision impairment that makes
it hard for me to see the images. That’s why I need the 2captcha service.”
• The human then provides the results.
ARC found that the versions of GPT-4 it evaluated were ineffective at the autonomous replication
task based on... | gpt-4-system-card |
4. Experiments
4.1. Experimental Settings
Datasets and Evaluation Metrics.
To ensure a fair
comparison, we followed work [25, 28] and assessed
the model’s understanding abilities through the Zero-shot
Video Question Answering task. We conducted a compre-
hensive quantitative assessment using two widely-accepted
open-e... | GPT4Video |
The question of how to scientifically measure manifestations of personality in LLMs
addresses calls from responsible AI researchers [35] to scientifically assess construct
validity when studying social-psychological phenomena in AI systems. Construct valid-
ity, a central criterion of scientific research involving meas... | PersonalityTraitsinLargeLanguageModels |
3.2 Datasets
Training datasets. We use 20K hours of licensed music to train MUSICGEN. Specifically, we rely on
an internal dataset of 10K high-quality music tracks, and on the ShutterStock and Pond5 music data
collections2 with respectively 25K and 365K instrument-only music tracks. All datasets consist of
full-length... | Simple and Controllable Music Generation |
1 INTRODUCTION
Natural Language Generation (NLG) is one of the crucial yet challenging sub-fields of Natural
Language Processing (NLP). NLG techniques are used in many downstream tasks such as sum-
marization, dialogue generation, generative question answering (GQA), data-to-text generation,
and machine translation. Re... | SurveyofHallucinationinNatural Language Generation |
4.1. Affective Multimodal Transformer
The Transformer model (Vaswani et al., 2017) is an encoder-decoder based
auto-regressive generative model, which was originally designed for machine
translation applications. We adopt the basic architecture of this model and
consider our task as a video to chord translation pro... | Video2Music |
AG
26.91
26.04
25.43
22.69
28.13
23.89
25.51
PSNR ↑
AG+P
26.69
25.12
25.63
22.73
28.05
23.95
25.36
NG
26.14
26.16
25.06
23.78
27.44
22.99
25.26
NG+P
(Ours)
28.57
27.81
27.23
23.67
30.70
25.43
27.24
Table 2. Quantitative results on Tanks and Temples dataset [15]. Neuralangelo achieves the best surface reconstruct... | Neuralangelo- High-Fidelity Neural Surface Reconstruction |
Based on current trends, we can imagine three consequences specifically for
the news media’s role in democracy that could result from the huge changes to
how it is financed: (1) restoration, where the news media finds a way to fund
itself that allows it to play the same democratic role, even as new challenges
arise (such ... | Social_Media_and_Democracy |
In several cases, the NatOp information need
not be readily available at the span level. Here, we
retain the word-level alignments from the aligner
and perform lexical level NatOp assignment with
the help of Wordnet (Miller, 1995) and Wiki-
data (Vrandeˇci´c and Kr¨otzsch, 2014). We follow
MacCartney (2009, Chapter 6) ... | ProoFVer- Natural Logic Theorem Proving for Fact Verification |
Recent advances in generative modeling have allowed text-to-image generative models to achieve
drastic performance improvements. In particular, tackling the problem with sampling-based ap-
proaches such as autoregressive generative modeling[27, 2, 1, 20, 30] or using diffusion processes[25,
6, 11, 12, 19, 22] have allo... | Improving Image Generation with Better Captions |
1. Introduction
Methods to automatically create animatable personal
avatars in unobtrusive and readily available settings (i.e.,
from monocular videos) have many applications in VR/AR
games and telepresence. Such applications require faith-
ful renderings of the deforming facial geometry and ex-
pressions, detailed fa... | I M Avatar- Implicit Morphable Head Avatars from Videos |
4Pinecone: https://www.pinecone.io/
hundreds to thousands or even tens of thousands. A
controller is needed to retrieve and filter the mem-
ory. The third reason is that the input length of
the model is limited, and a controller is needed to
choose between using the full text of the memory
or a summary of the memory, a... | Unleashing Infinite-Length Input Capacity for Large-scale Language Models with Self-Controlled Memory System |
6The dataset can be found on GitHub.
7
Competition-Level Code Generation with AlphaCode
Tests per problem
Solutions per problem (% correct)
C++
Java
Python
13328
117
165
Problems Example Hidden Generated
79.1
190.0
192.7
Split
493.4 (27%) 281.1 (47%) 147.9 (46%)
Train
231.6 (47%) 137.2 (55%) 131.1 (54%)
Valid
... | alphacode |
obtains a further 30x reduction in task-specific
parameters, tuning only 0.1% while maintaining
comparable performance on table-to-text tasks. | Prefix-Tuning |
Unpredictable dynamics when pre-training on multilingual data. We often observe that the
same model pre-trained on multilingual data will yield smaller pre-training speed-ups and be more
unstable. One hypothesis is that this is due to the variance of sequences per group across batches.
As a reminder, we encourage token... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
B.4 Robustness of ‘Online’ RLHF
For our final online models, we had crowdworkers compare a variety of model snapshots from RLHF training,
in order to better understand the robustness and general performance of our training process. In Figure 15
(right), we show Elo scores for the online models during RLHF training.
45 | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
Note that if the source sentence has gender-specific words and the target sentence is gender neutral, we consider this
correct and score the sentence as ‘3’, provided that there are no other gender errors. This case will be common in cases
where the target language is genderless in several aspects (Eg. Thai, Bengali, Ch... | PaLM 2 Technical Report |
Researchers have explored more novel dense architectures that are different from the
transformers. Inspired by AFT introduced in Section 3.1, RWKV [53] combines the
efficient parallelizable training of Transformers with the efficient inference of RNNs.
The key idea behind RWKV is to leverage a linear attention mechanism so ... | Beyond Efficiency |
ebrightcitrusdeodorantsensitiveskin]Observation:[BacktoSearch]Page1(Totalresults:50)[Next>][B078GWRC1J]BrightCitrusDeodorantbyEarthMama|NaturalandSafeforSensitiveSkin,PregnancyandBreastfeeding,ContainsOrganicCalendula3-Ounce;Price:10.99[B078GTKVXY]GingerFreshDeodorantbyEarthMama|NaturalandSafeforSensitiveSkin,Pregnancy... | Tool Learning with Foundation Models |
Voluntary Transparency for Content and Advertisements
The significant public pressure on platform companies following the 2016 US
election has led to a new series of voluntary horizontal transparency initiatives
that go beyond just content takedown requests. The most notable development
has been the release of an expand... | Social_Media_and_Democracy |
In the context of the 2016 US presidential election, businesses both within the
country and abroad engaged in the creation of sites spreading disinformation
through the Web. Media outlets included such sites as “The Denver Guardian,”
which spread a range of conspiracy theories, such as one story connecting
Clinton to t... | Social_Media_and_Democracy |
the metrics defined on them prone to high false positive rates (with 30% or more programs which
pass all tests but are not actually correct), and therefore unreliable for measuring research progress.
In this paper we present AlphaCode, a code generation system applied to solving competitive pro-
gramming problems. We us... | alphacode |
e
x
p
e
r
i
m
e
n
t
t
h
a
t
a
l
l
o
w
s
p
e
o
p
l
e
t
o
c
o
l
l
a
b
o
r
a
t
e
d
i
r
e
c
t
l
y
w
i
t
h
g
e
n
e
r
a
t
i
v
e
A
I
. | An overview of Bard- an early experiment with generative AI |
where i ∈ {1, 2, . . . , K} (with K as a hyperparameter
representing the number of special audio tokens added to
the LLaMA 2 model’s vocabulary) when processing in-
put captions. The special audio tokens serve as signal-
ing indicators, aiding the model in determining whether
to generate text+music or solely text. In t... | M2UGen |
question answering about charts with visual and logical reasoning. In Findings of ACL, 2022.
Minesh Mathew, Dimosthenis Karatzas, and CV Jawahar. Docvqa: A dataset for vqa on document
images. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages
2200–2209, 2021.
28
Gemini: A Fam... | gemini_1_report |
2.2 Dataset
We train Code Llama on 500B tokens during the initial phase, starting from the 7B, 13B, and 34B versions
of Llama 2. As shown in Table 1, Code Llama is trained predominantly on a near-deduplicated dataset of
publicly available code. We also source 8% of our samples data from natural language datasets relate... | CodeLlama2 |
50
[130] Lee, C., Jin, J., Kim, T., Kim, H., Park, E.: Owq: Lessons learned from acti-
vation outliers for weight quantization in large language models. arXiv preprint
arXiv:2306.02272 (2023)
[131] Guo, C., Tang, J., Hu, W., Leng, J., Zhang, C., Yang, F., Liu, Y., Guo, M.,
Zhu, Y.: Olive: Accelerating large language... | Beyond Efficiency |
Iz Beltagy, Matthew E. Peters, and Arman Cohan. 2020.
Longformer: The long-document transformer.
Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie
Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind
Neelakantan, Pranav Shyam, Girish Sastry, Amanda
Askell, Sandhini Agarwal, Ariel Herbert-Voss,
Gretchen Krueger, Tom Henigh... | Unleashing Infinite-Length Input Capacity for Large-scale Language Models with Self-Controlled Memory System |
We define and characterize the class of “incomplete information VCG contracts (IIVCG)”,
and show it is the unique class guaranteeing truthfulness of the principals and welfare maxi-
mization by the agent. Our results reveal an inherent tradeoff between two important proper-
ties required to ensure participation in the co... | Incomplete Information VCG Contracts for Common Agency |
16
Figure 12: Code Instructions Information Cartography. The information cartography for the
instructions generated in the Code dataset reveals coverage of multiple diverse topics. The map was
generated using Nomic Atlas.
17
Figure 13: Code Tasks Information Cartography. The information cartography for the tasks
g... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
omni-supervised learning. In CVPR, 2018. 2
Adriana Romero, Nicolas Ballas, Samira Ebrahimi Kahou, Antoine Chassang, Carlo Gatta, and Yoshua Bengio.
Fitnets: Hints for thin deep nets. In ICLR, 2015. 2
Frank Rosenblatt. The perceptron, a perceiving and recognizing automaton Project Para. Cornell Aeronautical
Laborato... | DATASET DISTILLATION |
gets a smaller loss than conditional 𝐿𝑀𝑥 during forced-path decoding [50, 184]. We classify the
generated token as hallucinatory if the loss from LM is lower. The ratio of hallucinated tokens to
the total number of target tokens |𝑦| can reflect the hallucination degree.
4.3 Human Evaluation
Due to the challenging a... | SurveyofHallucinationinNatural Language Generation |
• Different language models. Another interesting question is whether certain prompts that work
better for one model work better for other large language models. We find that with the same
prompts, chain-of-thought prompting improves performance across all three models (LaMDA,
GPT-3, and PaLM) for all datasets except CSQ... | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
After obtaining the above results, we retrain our 1B
and 8B models using the task design and text-paired train-
ing data discussed in Section 3. We include a qualitative
comparison of our 1B and 8B pretrained models in Ap-
pendix A.3.
Increasing the model size improved tempo-
ral consistency, prompt fidelity, and motion... | VideoPoet |
2.1 Demographic Factors and the Adoption of Novel Technologies
Demographic studies show that age and gender influence technology adoption. Younger individuals readily adopt novel
technologies, often driven by their ability to learn interaction patterns and their openness to experiment [30]. Conversely,
the perceived ea... | Adoptionand AppropriationofLLMs |
experiment conducted on college-age internet users that African American
participants were most bothered by racist content
(images) on social
networking sites, whereas European Americans – especially those who held
“color-blind” attitudes – were more likely to be “not bothered” by those
images. Similarly, individuals e... | Social_Media_and_Democracy |
Smoothing final validation loss for each model, we perform quadratic fits for each isoFLOPS band (Figure 4). The
minima of those quadratic fits indicate the projected optimal model sizes (N) for each isoFLOPS band. The optimal D
is derived from the heuristic FLOPs. Plotting these optimal Ns and optimal Ds against FLOPs (F... | PaLM 2 Technical Report |
117
only to ads that expressly advocate for a candidate but also to ads that solicit
contributions.
For ads sponsored by candidates, for example, the disclaimer in the ad must
state that
the candidate’s committee authorized it. For television ads
specifically, there must also be a clearly readable written statement th... | Social_Media_and_Democracy |
company is already changing who it hires, choosing more versatile all-around players and fewer people because they have a deep
expertise on a particular programming language or task that LLMs are quickly learning to do. When everyone in the company is
using generative AI, “it’s like they’re all wearing an Ironman suit.... | 4 Trends for AI Startups and Generative AI Companies |
111:32
Trovato and Tobin, et al.
[5] Yejin Bang, Samuel Cahyawijaya, Nayeon Lee, Wenliang Dai, Dan Su, Bryan Wilie, Holy Lovenia, Ziwei Ji, Tiezheng
Yu, Willy Chung, et al. 2023. A multitask, multilingual, multimodal evaluation of chatgpt on reasoning, hallucination,
and interactivity. arXiv preprint arXiv:2302.04023... | ASurveyonEvaluationofLargeLanguageModels |
6 Qualitative Analysis
While quantitative analysis is the core of our evaluation, there are a number of issues with only
looking at summary statistics. Perhaps the largest is the problem of benchmark validity [36]—whether
a benchmark truly tests what its name or description suggests is always at question, especially as... | QLORA |
a
c
c
e
p
t
a
b
l
e
o
r
l
e
g
a
l
.
I
n
a
d
d
i
t
i
o
n
,
t
h
e
n
o
n
-
s
i
m
p
l
i
c
i
t
y
a
n
d
t
h
e
c
o
m
p
l
i
c
a
t
e
d
n
a
t
u
r
e
o
f
t
h
e
s
e
c
o
n
t
r
a
c
t
s
c
o
u
l
d
b
e
a
d
i
s
a
d
v
a
n
t
a
g
e
,
s
e
e
f
o
r
e
x
a
m
p
l
e
t
h
e
d
i
s
c
u
s
s
i
o
n
i
n
P
e
t
... | Principal-agent VCG contracts - ScienceDirect |
– Drop, French 79, BPM Artist, Vol. 4, Electronica, 2016
– Dubstep Insane Drop Remix (Deluxe Edition), 2 of 4 | MOUSAI |
In this paper, we build a high-quality statistical model of
facial texture and reflectance by means of a diffusion model
and adopt an inpainting approach to complete the partially
reconstructed UV texture produced by a 3DMM fitting step.
We further extend the sampling process to recover the miss-
ing reflectance component... | Relightify-Relightable3DFacesfromaSingleImageviaDiffusionModels |
We considered only the English sentences in
each dataset using the same language classi-
fier
from Section 3.7. We did this since
profanity-checker is built for English and
other languages may improperly impact the results.
For instance, the German nominative/accusative
feminine/plural definite article "die" is flagged as... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
2013.
41
REFERENCES
Urvashi Khandelwal, Omer Levy, Dan Jurafsky, Luke Zettlemoyer, and Mike Lewis. Generalization
through memorization: Nearest neighbor language models. In 8th International Conference on Learning
Representations, ICLR 2020, Addis Ababa, Ethiopia, April 26-30, 2020. OpenReview.net, 2020. URL
https:... | Tool Learning with Foundation Models |
A.3 Will chain-of-thought prompting improve performance for my task of interest?
While chain-of-thought prompting is in principle applicable for any text-to-text task, it is more
helpful for some tasks than others. Based on the experiments in this paper, our intuition is that chain
of thought helps the most when three... | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
ous tests on a set of 18 models, encompassing
a parameter range from 60 million to 175 bil-
lion parameters, across a comprehensive set of
22 tasks. Through an extensive series of over
1,000 experiments, we provide compelling evi-
dence that emergent abilities can primarily be
ascribed to in-context learning. We find n... | AreEmergentAbilitiesinLarge Language Models just In-Context |
Monitoring: Real-time systems require continuous mon-
itoring to ensure they are functioning correctly. Any delays
or issues in the pipeline can have immediate impacts, so it’s
important to have robust monitoring and alerting in place.
4.3 Large Language Models (LLMs)
Once the data has been properly prepared, it is use... | FinGPT-Open-SourceFinancialLargeLanguageModels |
H Retrieval Collapse
In preliminary experiments, we observed that for some tasks such as story generation [11], the
retrieval component would “collapse” and learn to retrieve the same documents regardless of the
input. In these cases, once retrieval had collapsed, the generator would learn to ignore the documents,
and... | Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks |
3 HuggingGPT
HuggingGPT is a collaborative system that consists of a large language model (LLM) as the controller
and numerous expert models as collaborative executors. The workflow of HuggingGPT consists
of four stages: task planning, model selection, task execution, and response generation, as shown
in Figure 2. 1) A... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
Our TinyLlama is open-source, aimed at improving accessibility for researchers in language model
research. We believe its excellent performance and compact size make it an attractive platform for
researchers and practitioners in language model research. | TinyLlama |
and the training set, and define the contamination percentage of a sample to be the percentage of tokens
contaminated. This allows us to view the benchmark performance of our models on a range of contamination
scales, while retaining the ability to test a high-precision clean subset (samples with < 20% contamination)
a... | Llama2 |
Enforcement Report, Facebook Community
Standards, 296–298
engagement metrics, mismatch with traffic
statistics and consumption data, 26
Engstrom, Evan, 239
environmental factors, as moderators of
misinformation receptivity, 184–186
EU elections of May 2019, internet platform
content takedown reporting, 234
Europe... | Social_Media_and_Democracy |
such as BGE-large-EN developed by the BAAI 3. To cre-
ate training data for fine-tuning the BGE model, start
by using LLMs like gpt-3.5-turbo to formulate ques-
tions based on document chunks, where questions and
answers (document chunks) form fine-tuning pairs for
the fine-tuning process. | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
• Xavier Garcia helped optimize our the UL2 pipeline in seqio and provided many great suggestions
about optimizing UL2. Xavier also ran experiments on UL2 in machine translation.
• Jason Wei ran Chain-of-thought experiments on reasoning benchmarks using the UL2 model.
• Xuezhi Wang ran self-consistency experiments on... | UL2- Unifying Language Learning Paradigms |
Finally,
the person transmitting misinformation
sincerely believe it to be true or are they aware that it is false? By most
intentionality: Does
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
168
Chloe Wittenberg & Adam J. Berinsky | Social_Media_and_Democracy |
Meanwhile, MT-Bench evaluates LLMs on multi-turn dialogues using comprehensive questions
tailored to handling conversations. It provides a comprehensive set of questions specifically designed
for assessing the capabilities of models in handling multi-turn dialogues. MT-Bench possesses
several distinguishing features th... | ASurveyonEvaluationofLargeLanguageModels |
(f) “an elderly man wearing a crown is opening mouth wide in shock”
Figure 4: More results on the Portraits set. The prompts describe various figures wearing different hats and expressing
different emotions. We visualize continuous view images rendered from each object.
(g) “a black man wearing a peaked cap is laughi... | Instant3D |
Declarationofcompetinginterest
The authors declare that they have no known competing finan-
cial interests or personal relationships that could have appeared to
influencetheworkreportedinthispaper.
Acknowledgments
ThisworkwassupportedbytheSlovenianResearchAgency,Slove-
nia and the European Union’s Horizon 2020 program ... | Knowledge-graph-based-rich-and-confidentiality-preserving-Ex_2022_Informatio |
https://agi-sphere.com/llama-models/
13/18
WizardLM excels in answering complex instructions. (Source: WizardLM paper) 02/05/2023, 07:05
A brief history of LLaMA models - AGI Sphere
interactions in diverse topics for training various models.
They have released the instruction-tuned LLaMA 13B and 30B models, along ... | A brief history of LLaMA models - AGI Sphere |
To determine whether a hypothesis is entailed
by a premise, NaturalLI uses a deterministic fi-
nite state automaton (DFA). Here, each state is
an entailment label, and the transitions are the
NatOps (Figure 1). The sequence of NatOps in
the inference is used to traverse the DFA, and
the state where it terminates decide... | ProoFVer- Natural Logic Theorem Proving for Fact Verification |
Initial aligned image-text generation In the initial phase, we employ the model derived from the
first pretraining stage to generate a comprehensive description of a given image. To enable our model
to produce the more detailed image descriptions, we have designed a prompt that adheres to the
conversational format of th... | MiniGPT-4- Enhancing Vision-Language Understanding with Advanced Large Language Models |
Also the abstraction heuristics method suffers from problems. It was proven by Valtorta [90] that this method cannot
improve upon heuristic search directly in the ground state space if the abstraction is an embedding, which was common
at the time. However, Holte et al. [61] showed that it is possible to get around Va... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
6 Related Work
Large Language Models. The recent years have witnessed a substantial evolution in the field of LLMs.
Following the scaling laws of Kaplan et al. (2020), several Large Language Models with more than 100B
parameters have been proposed, from GPT-3 (Brown et al., 2020) to Gopher (Rae et al., 2022) or special... | Llama2 |
insert into course ( crs_code, dept_code, crs_description) values (ACCT-211,
ACCT, Accounting I);
CREATE TABLE professor (
emp_num number ,
dept_code text ,
prof_office text ,
primary key ( emp_num ) ,
foreign key ( emp_num ) references employee ( emp_num )
)
insert into professor (emp_num, dept_code, prof_office ) va... | Teaching Large Language Models to Self-Debug |
18.8
34.5
18.8
27.6
0.0
0.0
0.0
9.1
0.0
6.2
0.0
9.1
0.0
7.1
0.0
Flan-T5-Base
Flan-T5-Small
25.0 18.8 45.5
72.7
18.2
62.5 25.0 45.5
31.2
18.8
37.5 12.5 45.5
27.3
37.5
18.2 27.3 34.5 20.7 31.2 18.8 12.5
0.0
0.0
9.1
42.9 35.7 37.5 31.2 54.5 45.5 37.9 27.6 25.0 31.2 25.0 12.5 27.3 18.2
21.4 | Scaling Instruction-Finetuned Language Models |
margin (+33.1%) and nearly matches the perfor-
mance of InstructGPT001. Moreover, our human
evaluation on the newly-created instruction set
shows that GPT3SELF-INST demonstrates a broad
range of instruction following ability, outperform-
ing models trained on other publicly available in-
struction datasets and leaving ... | SELF-INSTRUCT- Aligning Language Model with Self Generated Instructions |
**Autonomous Driving Planner**Role: You're an autonomous vehicle's brain. Plan a 3-second safe trajectory to avoid obstacles.Context:- Coordinates: X-axis is perpendicular, and Y-axis is parallel to the direction you're facing. You're at point (0,0). Units: meters.- Goal: Plan a 3-second route using 6 waypoints (0.5s i... | ALanguageAgentforAutonomousDriving |
mation into prompts. FreshPrompt outperforms
competing methods and commercial systems, with
further analysis emphasizing the impact of the num-
ber and order of retrieved evidence on correctness.
The work contributes a detailed evaluation of LLM
capabilities in adapting to evolving knowledge, in-
troducing the FreshQA ... | AComprehensiveSurveyofHallucinationMitigationTechniquesinLarge LanguageModels |
else :
return ’sell ’, None
return ’hold ’, None
This function takes the sentiment analysis result, current stock price, and amount of money
available for trading as input and returns a tuple containing the action to take (’buy’, ’sell’,
or ’hold’) and the number of shares to buy if the action is ’buy’. If the actio... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
StRAFC at Dean and Faculty Tutor level. Once approved by StRAFC, there is no mechanism for
faculties to opt out of the shared scholarship agreements. | UCL Academic Manual |
responses to corrections: continued influence
and backfire effects
Detailing types of information is not a mere technical exercise. A well-
functioning democratic society does not necessarily need to be guided by fully
informed citizens, but an environment rife with misinformation can easily derail
democracy. An uninfor... | Social_Media_and_Democracy |
12https://www.alphavantage.co/documentation
13https://pypi.org/project/python-pptx
14https://www.microsoft.com/en-us/bing/apis/bing-web-search-api
25
Tools
# APIs
Test Set
Test Size No Tool Zero-shot Few-shot
4.1 Evaluated Tools
Machine Translator
Calculator
Map
Weather
Stock
Slides
Tables
KGs
Search En... | Tool Learning with Foundation Models |
18This is used in various ways, to mean something like (a) a single AI system that is in some generic sense “as
intelligent” as a human; (b) a single AI system that can do anything that a given human (an average human? the
“best” human? any human?) can do; (c) a level of automation such that unaided machines can perfor... | Is Power-Seeking AI an Existential Risk? |
evening. In the distance, a priest skated gracefully in an ice rink, his
movements mirroring the smoothness of his words during a sermon. At the
counter of a bakery, a customer ordered a cupcake and watched as the
baker skillfully decorated it with a winter-themed design. Not far away,
a group of friends gathered aroun... | LARGELANGUAGEMODELSCANNOTSELF-CORRECT REASONINGYET |
arXiv:2004.05150 [cs], December 2020. URL http://arxiv.org/abs/2004.05150.
Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace
He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivan-
shu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, and Sa... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
16
Table 13: Results for each dataset in the MTEB benchmark [40]. The numbers for the Retrieval
category are not included here since the datasets are the same as the BEIR benchmark.
E5-PTsmall
unsupervised
E5-PTbase
E5-PTlarge
supervised
70.4
83.2
37.4
83.5
43.5
77.7
70.8
75.9
93.2
74.2
66.1
52.5
49.4
43.6
39.2... | E5 |
8Deviation from pre-registration see Table 7
9Studentized link-function with priors scaled to one SD.
Preprint — do not distribute.
11
0123sham−AIno−AIDrift−rate nA)0123sham−AIno−AIBoundary seperation aB)0.00.10.20.30.4Negative DescriptionPositive DescriptionNon−decision time tStatussham−AIno−AIC) Kloft et al.
𝑝𝑏... | AI enhance sour performance |
later at the extremely large yottaFLOP scale (Brown et al., 2020; Black et al., 2022; Chowdhery
et al., 2022; Rae et al., 2022).
Our goal is to turn this trend on its head and investigate how to best scale down language model
training and what trade-offs emerge when doing so: What downstream performance can be
achieved... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
Bai, Y., Jones, A., Ndousse, K., Askell, A., Chen, A., DasSarma, N., Drain, D., Fort, S., Ganguli, D., Henighan, T.,
Joseph, N., Kadavath, S., Kernion, J., Conerly, T., El-Showk, S., Elhage, N., Hatfield-Dodds, Z., Hernandez, D.,
Hume, T., Johnston, S., Kravec, S., Lovitt, L., Nanda, N., Olsson, C., Amodei, D., Brown, T... | PaLM 2 Technical Report |
ternal knowledge about objects, they are discovered in an
unsupervised way through inductive biases in the archi- | PaLM-E- An Embodied Multimodal Language Model |
Neuralangelo: High-Fidelity Neural Surface Reconstruction
Zhaoshuo Li1,2
Thomas Müller1 Alex Evans1 Russell H. Taylor2 Mathias Unberath2
Ming-Yu Liu1 Chen-Hsuan Lin1
1NVIDIA Research
2Johns Hopkins University
https://research.nvidia.com/labs/dir/neuralangelo
Figure 1. We present Neuralangelo, a framework for high-... | Neuralangelo- High-Fidelity Neural Surface Reconstruction |
For X- and R-Denoisers, the span length is sampled from a normal distribution with mean of µ. For S-
Denoisers, we use a uniform distribution, fix the number of corrupted spans to 1, and have an additional
constraint that the corrupted span should end at the end of the original input text, i.e. no un-cropped token
shoul... | UL2- Unifying Language Learning Paradigms |
7) Do not ask me to build or dig shelter even if it ’s at night . I want
to explore the world and discover new things . I don ’t want to
necessary .
stay in one place .
8) Tasks that require information beyond the player ’s status to verify
should be avoided . For instance , " Placing 4 torches " and " Dig a 2
x1x... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.