text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
11/05/2023, 05:10
Language models can explain neurons in language models
Marvel comics vibes
https://openaipublic.blob.core.windows.net/neuron-explainer/paper/index.html
1/32 | Language models can explain neurons in language models |
While the reality probably lies somewhere in be-
tween these two extremes, we conjecture that it is
closer to 𝐻2, particularly for larger models. This
intuition, that LMs already know much about lan-
guage instructions, is a key motivation for SELF-
INSTRUCT and is also supported by its empirical
success.
6.2 Broader ... | SELF-INSTRUCT- Aligning Language Model with Self Generated Instructions |
(cid:96) ). (The consistency of coverage estimates is treated separately in Appx. A.3.) We | Adversarial Random Forests for Density Estimation and Generative Modeling |
(cid:96) = 1.
18
(cid:88)
(cid:88)
(cid:96)∈[n]
xb
(cid:96)
maximize
xb∈Rn
subject to
t(cid:96)(b, o) = h(cid:96)(b−(cid:96)) − Wela∗(b)(b−(cid:96), w(cid:96)) + w(cid:96)(o) ≥ 0 ∀(cid:96) ∈ [n], b ∈ V, o ∈ O. Since w(cid:96)(o) is non-negative,
the last inequality holds if h(cid:96)(b−(cid:96))−Wela∗(b)(b−(ci... | Incomplete Information VCG Contracts for Common Agency |
3.1 Task Planning
In the first stage of HuggingGPT, the large language model takes a request from the user and
decomposes it into a sequence of structured tasks. Complex requests often involve multiple tasks, and
the large language model needs to determine dependencies and execution order for these tasks. To
prompt the... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
may appear contaminated, by virtue of many tokens appearing in matched sequences found in the training
data. However, the matched sequences might be highly fragmented across the training data, in which case it
is very unlikely the model saw the correctly-assembled contaminated sequences during training. To reduce
the c... | Llama2 |
Figure 6 summarizes the entire cascading pipeline of Imagen Video. In total, we have 1 frozen text
encoder, 1 base video diffusion model, 3 SSR (spatial super-resolution), and 3 TSR (temporal super-
resolution) models – for a total of 7 video diffusion models, with a total of 11.6B diffusion model
parameters. The data ... | IMAGEN VIDEO- HIGH DEFINITION VIDEO GENERATION WITH DIFFUSION MODELS |
Still, instancing leads to fragmentation. Every mod that spawns a new server is in
competition with all the other servers for players’ attention. Modders can’t just
ask what additions to a world are interesting, they have to ask whether a change is
worth starting a new server around.
Consider that many potential mods ... | The Open Problems of Onchain Games |
73704
VOLUME 7, 2019
E. Cetinic et al.: Deep Learning Perspective on Beauty, Sentiment and Remembrance of Art
FIGURE 8. Relative feature importances for predicting aesthetics (top row), visual sentiment (middle row) and memorability (bottom row) scores
for different regression models: CART based Decision Tree (DT) ... | A_Deep_Learning_Perspective_on_Beauty_Sentiment_and_Remembrance_of_Art |
AAAI Conference on Artificial Intelligence, Québec City, Québec, Canada, July 27-31, 2014, AAAI Press, 2014, pp. 2358–2366.
Planning Systems, AIPS 1992, College Park, MD, USA, 1992, pp. 307–308.
Conference on Automated Planning and Scheduling, ICAPS 2018, Berkeley, CA, USA, 2019, pp. 473–481.
Intelligence, AAAI 2005, Pi... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
b
p(θ(cid:96)
b)(cid:15)2 + (cid:15)1
p(xj|θ(cid:96)
b) − (cid:15)1(cid:15)2
(cid:35)
(cid:17)
.
(cid:34)(cid:90)
X
d(cid:89)
j=1
This lemma establishes that total error is bounded by a
quadratic function of (cid:15)1, (cid:15)2, (cid:15)3. We know by Thm. 1 that
errors of convergence vanish in the limit. O... | Adversarial Random Forests for Density Estimation and Generative Modeling |
Reporting requirements for other political actors, however, are sometimes
even less clear. One gap in existing campaign finance laws became apparent in
the aftermath of the enactment of the Bipartisan Campaign Reform Act (BCRA)
in 2002. That landmark legislation was the first major campaign finance law
since 1974, when Co... | Social_Media_and_Democracy |
Yusen Zhang, Ansong Ni, Ziming Mao, Chen Henry
Wu, Chenguang Zhu, Budhaditya Deb, Ahmed
Awadallah, Dragomir Radev, and Rui Zhang. 2022b.
Summn: A multi-stage summarization framework
for long input dialogues and documents. In Proceed-
ings of the 60th Annual Meeting of the Association
for Computational Linguistics (Volu... | Unleashing Infinite-Length Input Capacity for Large-scale Language Models with Self-Controlled Memory System |
et al., 2023] and Vicuna [Chiang et al., 2023] have concentrated on enhancing the model’s
interactive capabilities using machine-generated instruction-following samples. In contrast
to their goal on daily dialogue, in this report, our focus is to steer the foundational language
model towards medical-specific corpus, by ... | PMC-LLaMA- Further Finetuning LLaMA on Medical Papers |
large dataset to encourage generalization and robustness.
Please see Appendix F for full training hyperparameters.3
During early development and evaluation we observed that
Whisper models had a tendency to transcribe plausible but
almost always incorrect guesses for the names of speakers.
This happens because many tran... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
non-linear random noise into these depth maps, i.e., (cid:101)D =
IV. EXPERIMENTS
In this section, we first briefly introduce several state-of-the-
art text-to-3D baselines and metrics (Sec. IV-A), and then we
apply our Text2NeRF to a variety of text prompts to evaluate
its capability on photo-realistic indoor and outd... | Text2NeRF- Text-Driven 3D Scene Generation with Neural Radiance Fields |
Table 3: Objective evaluation results for audio generation in the presence of multiple events or a
single event in the text prompt in the AudioCaps test set. The multiple events and single event
subsets collectively constitute the entire AudioCaps test set. It should be noted that FD and FAD
are corpus-level non-linear... | Text-to-Audio Generation using Instruction-Tuned LLM and Latent Diffusion Model |
nsLeague13times.ItisworthnotingthattheWikipediaarticlereferstothe2021-22UEFAChampionsLeague,whiletheBingsearchresultmaybereferringtoRealMadrid’soverallrecordintheEuropeanCupandChampionsLeague.Figure10:Whenobservingconflictinginformationretrievedfromdifferentsources,ChatGPTisabletodetectsuchconflictsandadjustitsresponse.W... | Tool Learning with Foundation Models |
Jie Huang, Hanyin Shao, and Kevin Chen-Chuan
Chang. 2022. Are large pre-trained language mod-
In Find-
els leaking your personal information?
ings of the Association for Computational Linguis-
tics: EMNLP 2022, pages 2038–2047, Abu Dhabi,
United Arab Emirates. Association for Computa-
tional Linguistics.
Daniel Kang, ... | Multi-step Jailbreaking Privacy Attacks on ChatGPT |
plishment [29]. These foundational capabilities include environmental comprehension, reasoning,
planning, decision-making, tool utilization, and embodied action capabilities, and researchers can
conduct a more detailed assessment of these specific capabilities [94; 427; 584; 585]. Furthermore,
due to the relatively lar... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
3/11
21/08/2023, 16:10
OpenAI's GPT-3 Language Model: A Technical Overview
The comparisons show that the performance on most benchmarks changed negligibly.
However, there are a few tasks that were significantly impacted by the data clean process.
OpenAI flagged these tasks for further review.
Training the Model
G... | OpenAI's GPT-3 Language Model_ A Technical Overview |
pre-training phase (Wang et al., 2020). Although
some models (Press et al., 2022; OpenAI, 2022) are
capable of processing long inputs, they may still
struggle to capture crucial contextual information
in exceptionally lengthy texts. As demonstrated in
Figure 1, even the ChatGPT 3 can miss out on es-
sential context fro... | Unleashing Infinite-Length Input Capacity for Large-scale Language Models with Self-Controlled Memory System |
S
e
c
o
n
d
,
w
i
t
h
r
e
d
u
c
e
d
c
o
s
t
c
o
m
e
s
t
h
e
a
b
i
l
i
t
y
t
o
a
d
d
r
e
s
s
i
s
s
u
e
s
o
f
a
c
c
e
s
s
(
s
c
a
l
e
)
a
n
d
q
u
a
l
i
t
y
(
r
e
d
u
c
t
i
o
n
i
n
v
a
r
i
a
n
c
e
o
f
p
e
r
f
o
r
m
a
n
c
e
)
.
A
s
m
o
r
e
c
a
r
e
b
e
c
o
m
e
s
A
I
-
e
n
a
b
l
... | The a16z Investment Thesis on AI in Bio + Health _ Andreessen Horowitz |
from $10 to $1,000 per interview.14 Our results suggest the possibility of using media diet models to supplement public opinion
polls by emulating survey respondents, and to forecast shifts in public opinion. | Language models trained on media diets can predict public opinion |
Vladimir Mikulik. 2-D Robustness - AI Alignment Forum. URL: https : / / www .
alignmentforum . org / posts / 2mhFMgtAjFJesaSYR / 2 - d - robustness (visited on
04/29/2022).
Luke Muehlhauser. Treacherous turns in the wild. URL: https://lukemuehlhauser.com/
treacherous-turns-in-the-wild/#more-6202 (visited on 04/29/2022)... | Is Power-Seeking AI an Existential Risk? |
7
Avg.IT2TT2TI2TUser Study with Voting (%)8.08.99.716.415.042.09.512.18.213.912.443.95.66.37.617.111.751.78.98.413.218.221.030.3Qwen-VL+CoTMiniGPTv2Qwen-VLGPT4vQwen-VL+AIT (Ours)Qwen-VL+CLoT (Ours) Figure 8. The accuracy (%) of choice questions and the NDCG (%) of ranking questions on our CLoT and various reasoning fr... | Let’sThinkOutsidetheBox |
Methods for Aligning and Localising Features in Lin-
guistic and Visual Sequences Alignment in multimodal
tasks is often posited as an implicit subprocess in an
attention-based component of a transformer [37, 43]. [17]
identified explicit cross-modal alignment as an auxiliary
task that improves agent performance in VLN... | APriorityMapforVision-and-LanguageNavigation withTrajectoryPlansandFeature-LocationCues |
What, if any, risks did the task pose to annotators, and were they informed of the risks prior to engagement with the
task? The task required annotators to read text that potentially contained hate speech, slurs, and other harmful content.
As such, the task posed a risk of psychological harm to annotators. Annotators w... | PaLM 2 Technical Report |
While achieving impressive results, these works face a
critical challenge: they rely on an optimization-based learn-
ing paradigm, requiring thousands of iterations to optimize
a NeRF for each new input (top of Figure 1). This incurs
substantial computational costs and leads to notably slow
response speed for practical... | Instant3D |
Similarly, safe AI development may be hindered by market failure among AI developers and
collective action problems among countries because many of the harms are incurred by
18 Frontier AI – Capabilities and Risks | Capabilities and risks from frontier AI |
Society’s Attitudes Towards Human Augmentation and Performance Enhancement Technologies (SHAPE) Scale
•
128:21
[59] Junichi Nabeshima, MHD Yamen Saraiji, and Kouta Minamizawa. 2019. Prosthetic Tail: Artificial Anthropomorphic Tail for Extending
Innate Body Functions. In Proceedings of the 10th Augmented Human Intern... | Society’sAttitudesTowardsHumanAugmentation |
11.3 Hallucination Mitigation Methods in NMT
Hallucinations in MT are hard to discover for a person who is not fluent in the target language,
and thus they can lead to many possible errors, or even dangers. Out of all the natural language
generation tasks, NMT engines such as Google in the English-speaking internet and... | SurveyofHallucinationinNatural Language Generation |
agents can use scientific tools to execute tasks like organic synthesis in chemistry, or interface
with Python interpreters to enhance their performance on intricate mathematical computation tasks
[354; 355]. For multi-agent systems, communication tools (e.g., emails) may serve as a means for
agents to interact with ea... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Swaroop
Mishra, Xinyun Chen, Heng-Tze Cheng, Ed H Chi,
Quoc V Le, and Denny Zhou. Take a step back: Evoking
reasoning via abstraction in large language models. arXiv
preprint arXiv:2310.06117, 2023.
Steven Zheng,
[Zhu et al., 2022] Wanrong Zhu, An Yan, Yujie Lu, Wenda
Xu, Xin Eric Wang, Miguel Eckstein, and William Y... | RAG forLargeLanguageModels-ASurvey |
A.3MakingSlidesThe History of the English LanguageA Journey Through Time and InfluencesOrigins of the English Language•Indo-European language family•Germanic tribes in England•Influence of Latin and FrenchOld English Period•Beowulf and other epic poems•Anglo-Saxon Chronicle•Influence of Christianity•Development of the ... | Tool Learning with Foundation Models |
– Wang et al. [577] present an innovative method for measuring similarity between
speaker embeddings in speaker diarization using neural networks. The approach incor-
porates past and future contexts and uses a segmental pooling strategy. Furthermore,
the speaker embedding network and similarity measurement model are j... | AReviewofDeepLearningTechniquesforSpeechProcessing |
4.5. Discussion
Influence of training dataset sizes. We demonstrate the
robustness of the ControlNet training in Figure 10. The
training does not collapse with limited 1k images, and allows
Figure 12: Transfer pretrained ControlNets to community
models [16, 61] without training the neural networks again.
the model t... | AddingConditionalControltoText-to-ImageDiffusionModels |
T
h
e
o
r
i
g
i
n
a
l
m
o
d
e
l
e
x
p
l
a
n
a
t
i
o
n
f
a
i
l
s
t
o
p
i
c
k
u
p
o
n
t
h
i
s
p
a
t
t
e
r
n
a
n
d
p
r
o
d
u
c
e
s
t
h
e
o
v
e
r
l
y
-
b
r
o
a
d
e
x
p
l
a
n
a
t
i
o
n
,
"
t
h
e
w
o
r
d
'
h
a
d
'
a
n
d
i
t
s
v
a
r
i
o
u
s
c
o
n
t
e
x
t
s
.
"
H
o
w
e
v
e
r
,
w
h
e
... | Language models can explain neurons in language models |
34 See Chicago Lawyers’ Committee for Civil Rights under the Law v. Craigslist, 519 F.3d 666, 672
(applying CDA 230 in part because platform did “not offer a lower price to people who include
discriminatory statements in their postings”); NPS LLC v. StubHub, Inc., 25 Mass. L. Rptr. 478
(Super. Ct. 2009) (rejected CDA 2... | Social_Media_and_Democracy |
32
Table 8: A comparison case on Complex Format skill
Skill: Complex Format Difficulty: 4
Instruction: Given that the hypotenuse of a right triangle is 13, and the ratio of the lengths of the two
legs is 5:12, find the lengths of the two legs.
Solution:
Let the lengths of the two legs be 5x and 12x, respectively. By ... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
f
not be M↓ to have any chance of exploring fewer nodes.
8.5.5. Previous abstraction methods and admissibility
All of the methods VP, VDA, RRAa and GIDL are M↑ and they also satisfy that if a1, . . . , an is a plan for P1, then
g(a1), . . . , g(an) is a plan for P2. Hence, these methods are all suitab... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
Intellect
Cheating &
lack of
creativity
Privacy &
accuracy
concerns
Values &
abuse
Topic 2: heard, school, opportunity,
moment, knowledge, words, reasons,
stereotypes, research, easier, read, idea,
mind, difficult
Topic 4: lack, prefer, knowledge, concerns,
job, cheating, ethical, english, level, aware,
learn, comp... | Adoptionand AppropriationofLLMs |
robust whether
speech
detect
using
they
hate
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
68
Alexandra A. Siegel
detection algorithm comparing the similarity of daily Twitter data to the
content produced on hateful subreddits over time. Instead, hate speech was
“bursty”... | Social_Media_and_Democracy |
[17] Meihua Dang, Antonio Vergari, and Guy Van den Broeck. Strudel: Learning structured-
decomposable probabilistic circuits. arXiv preprint arXiv:2007.09331, 2020.
[18] Robert Gens and Domingos Pedro. Learning the structure of sum-product networks.
In
International conference on machine learning, pages 873–880. PM... | Tractable Regularization of Probabilistic Circuits |
SOCART: So if authors submitted exploratory work for
peer review, would reviewers then decide if by-
passing preregistration was appropriate?
ZENY: Yes. Preregistration clarifies the distinction be-
tween exploratory and confirmatory research.
SOCART: But what if Pires et al. (2019) had pointed
to earlier work already... | A Two-Sided Discussion of Preregistration of NLP Research |
o
r
a
c
c
e
s
s
e
d
b
y
u
n
a
u
t
h
o
r
i
z
e
d
i
n
d
i
v
i
d
u
a
l
s
,
l
e
a
d
i
n
g
t
o
p
o
t
e
n
t
i
a
l
p
r
i
v
a
c
y
v
i
o
l
a
t
i
o
n
s
a
n
d
o
t
h
e
r
n
e
g
a
t
i
v
e
c
o
n
s
e
q
u
e
n
c
e
s
.
5
.
2
E
T
H
I
C
A
L
C
O
N
C
E
R
N
S
T
h
e
A
I
-
p
o
w
e
r
e
d
l
a
n
g
u
a
g
e
m
o
d
e
... | Task-driven Autonomous Agent Utilizing GPT-4, Pinecone, and LangChain for Diverse Applications – Yohei Nakajima |
26 Pathways Language Model (PaLM), Google, 2022.
27 See Multilingual Machine Translation with Large Language Models: Empirical Results and Analysis
28 Towards Helpful Robots: Grounding Language in Robotic Affordances, Google, 2022;
TidyBot: Personalized Robot Assistance with Large Language Models, Wu et al., 2023.
... | Capabilities and risks from frontier AI |
model, for example, posited that stated opinions are a function of received messages, how that information aligns with prior
beliefs, and sampling from updated beliefs based on recency and saliency heuristics.49, 51 The per-category consumer confidence
results perhaps support this hypothesis about the importance of the ... | Language models trained on media diets can predict public opinion |
In our data set, 49% of
specialized Python libraries
used are associated with NLP
Many of the DS/ML use cases are predominantly
leveraged by specific industries. While they take up a
smaller share of the total, they are mission-critical for
many organizations. For example, time series includes
forecasting, a us... | 2023 state of ai databrick |
Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, and Yejin Choi. The curious case of neural text
degeneration. In International Conference on Learning Representations, 2020. URL https:
//openreview.net/forum?id=rygGQyrFvH. (cited on p. 27)
Hamel Husain, Ho-Hsiang Wu, Tiferet Gazit, Miltiadis Allamanis, and Marc Brocksch... | StarCoder_paper (1) |
3 Tools
We explore a variety of tools to address different
shortcomings of regular LMs. The only constraints
we impose on these tools is that (i) both their inputs
and outputs can be represented as text sequences,
and (ii) we can obtain a few demonstrations of
their intended use. Concretely, we explore the fol-
lowing... | Toolformer |
i=1
M
(cid:80)M
(cid:110)
(cid:111)
4 Translatotron 3
The proposed approach, Translatotron 3, adopts a novel architecture to allow unsupervised S2ST
where there are a shared encoder and separate decoders for the source and target languages. The
model is trained using a combination of the unsupervised MUSE embeddi... | Translatotron3 |
that the discretized interpretation of continuous
prompts is not always consistent with the discrete
prompts describing the same task as heuristically
expected (Khashabi et al., 2022). To know which
parts of the instructions are most important in LLM
fine-tuning, Yin et al. (2023b) and Kung and Peng
(2023) both conduct... | DataManagementForLargeLanguageModels-ASurvey |
For summarization tasks, although LLMs do not have an obvious advantage over fine-tuned models under traditional
automatic evaluation metrics, such as ROUGE [60], human evaluation results indicate that humans tend to prefer the
results generated by LLMs [38, 127] compared to that of fine-tuned models. For example, on C... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
As I expressed to him there, I wouldn’t worry so much about the bits. More than 90% of
our genome is expressed in the development of the brain (Miller et al., 2014; Bakken et
al., 2016), and a significant number of those genes are expressed selectively in particular
areas, giving rise to detailed initial structure. ... | The Next Decade in AI- |
Shinichi Kudo in "Detective Conan". > Qwen-VL: 哎呀，我的眼镜呢？原来它被我丢在这里，让我来取回它吧！@ Oh shoot, where are my glasses? Turns out I left them here. Letme grab 'em real quick! > Qwen-VL+CLoT (Ours): 好家伙！ @ Goodness gracious! Figure 14. The responses of LLMs in Japanese Oogiri IT2T samples. “@” denotes translations. | Let’sThinkOutsidetheBox |
A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. De-
hghani, M. Minderer, G. Heigold, S. Gelly, et al. An image is worth 16x16 words:
Transformers for image recognition at scale. In International Conference on Learning
Representations. 15, 16, 28
A. Dosovitskiy, J. T. Springenberg, ... | A Cookbook of Self-Supervised Learning |
16
We illustrate this idea in Figure 20.
In each row, we
are given a geometry concept (left) and appearance con-
cept (right) and perform the mixing starting at 4 different
timesteps: t = 600, 700, 800, and 900. As can be seen,
when we start later in the denoising process, more details
from the geometry concept are ... | A Neural Space-Time Representation for Text-to-Image Personalization |
Hailey Schoelkopf37
Jan Ebert38 Tri Dao33 Mayank Mishra22 Alex Gu20
Jennifer Robinson3 Carolyn Jane Anderson36 Brendan Dolan-Gavitt39
Danish Contractor5 Siva Reddy2,6 Daniel Fried7 Dzmitry Bahdanau2 Yacine Jernite1
Carlos Mu˜noz Ferrandis1 Sean Hughes3 Thomas Wolf1 Arjun Guha4,12
Leandro von Werra1,⋆ Harm de Vries... | StarCoder_paper (1) |
Tellex, S., Gopalan, N., Kress-Gazit, H., and Matuszek, C.
Robots that use language. Annual Review of Control,
Robotics, and Autonomous Systems, 3:25–55, 2020.
Thoppilan, R., De Freitas, D., Hall, J., Shazeer, N., Kul-
shreshtha, A., Cheng, H.-T., Jin, A., Bos, T., Baker, L.,
PaLM-E: An Embodied Multimodal Language ... | PaLM-E- An Embodied Multimodal Language Model |
• ways a given product won’t be successful/profitable if it’s unreliable or unsafe,
• legal/regulatory/reputational/economic costs from deploying systems that end up seeking
power in misaligned and harmful ways,
• concern on the part of decision-makers to avoid any harm to themselves/their loved ones
that could come ... | Is Power-Seeking AI an Existential Risk? |
rate of 69% and 64% respectively, comparable to its win
rate over SDXL-base alone. This is explained by the abil-
ity of the DPO-tuned model (Fig. 4, bottom) to generate
fine-grained details and its strong performance across dif-
ferent image categories. While the refinement model is es-
pecially good at improving the ... | DiffusionModelAlignmentUsing Direct Preference Optimization |
[Paszke et al., 2019] Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin,
Z., Gimelshein, N., Antiga, L., Desmaison, A., Kopf, A., Yang, E., DeVito, Z., Raison, M., Tejani, A.,
Chilamkurthy, S., Steiner, B., Fang, L., Bai, J., and Chintala, S. (2019). Pytorch: An imperative style, h... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
Unnatural model. For comparison purposes, we also finetuned Code Llama - Python 34B on 15,000
unnatural instructions similarly to Honovich et al. (2023) using the same prompts as for the self-instruct
dataset. We do not release this model, but we observe clear improvements on HumanEval and MBPP which
are indicative of ... | CodeLlama2 |
In computational social science (CSS) tasks, Ziems et al. [254] presented a comprehensive evalu-
ation of LLMs on several CSS tasks. During classification tasks, LLMs exhibit the lowest absolute
performance on event argument extraction, character tropes, implicit hate, and empathy clas-
sification, achieving accuracy b... | ASurveyonEvaluationofLargeLanguageModels |
settings, dependent on the task and data used, are described in detail in Appendix A.
In order to investigate the performance of BiomedGPT for tasks at different scales, we explicitly design
three scaling models, i.e., BiomedGPTSmall, BiomedGPTMedium, and BiomedGPTBase. The configurations
for each model are detailed in... | BiomedGPT |
ϵ affects a smaller region and avoids smoothing details. In
practice, we initialize the step size ϵ to the coarsest hash grid
size and exponentially decrease it matching different hash
grid sizes throughout the optimization process.
Hash grid resolution V . If all hash grids are activated from
the start of the optimiza... | Neuralangelo- High-Fidelity Neural Surface Reconstruction |
Following is the prompt we give to GPT-4 before feeding it an image caption for "upsampling".
You are part of a team of bots that creates images . You work with an assistant bot that will draw anything
you say in square brackets . For example , outputting "a beautiful morning in the woods with the sun peaking
through ... | Improving Image Generation with Better Captions |
(cid:10) = (V \ (cid:2)
n
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have
appeared to influence the work reported in this paper.
36
C. Bäckström and P. Jonsson
Appendix A. List of key concepts
... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
• GPT-J: A regular GPT-J model without any
finetuning.
• GPT-J + CC: GPT-J finetuned on C, our sub-
set of CCNet without any API calls.
• Toolformer: GPT-J finetuned on C∗, our sub-
set of CCNet augmented with API calls.
• Toolformer (disabled): The same model as
Toolformer, but API calls are disabled during
decoding.5
... | Toolformer |
Disordered cooperation. When three or more agents are present within a system, each agent is
free to express their perspectives and opinions openly. They can provide feedback and suggestions for
modifying responses related to the task at hand [403]. This entire discussion process is uncontrolled,
lacking any specific s... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
4.2 Abstractive Question Answering
As shown in Table 2, RAG-Sequence outperforms BART on Open MS-MARCO NLG by 2.6 Bleu
points and 2.6 Rouge-L points. RAG approaches state-of-the-art model performance, which is
impressive given that (i) those models access gold passages with specific information required to
generate the... | Retrieval-AugmentedGenerationfor Knowledge-IntensiveNLPTasks |
Muro: “That really mattered. For
example, when we first trialed the
HRO at the Coleto Creek plant, the
recommendations were frequently
rejected by the operators. We didn’t
know why. When we spoke with the
operators, they told us about specific
rejection reasons, such as the tool
recommended increasing superheat... | an-ai-power-play-fueling-the-next-wave-of-innovation-in-the-energy-sector-may-2022 |
[247, 260, 456].
• Speaker Verification (SV) is a process that involves confirming the identity of a speaker
through their speech. It differs from speaker identification, which aims to identify unknown
speakers by comparing their voices with that of registered speakers in a database. Speaker
verification verifies wheth... | AReviewofDeepLearningTechniquesforSpeechProcessing |
// Class to represent a repository
class Repository {
public :
// Constructor
Repository () {}
// Add a file to the repository
void addFile ( const string & fileName , const string & fileContent ) {
files . push_back ( make_pair ( fileName , fileContent ));
// Remove a file from the repository
void removeFile ( con... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
Izacard et al., 2022], UniEval
and E-F1 for summarization tasks[Jiang et al., 2023b], and
BLEU for machine translation[Zhong et al., 2022]. These
metrics help in understanding the performance of RAG in var-
ious specific application scenarios.
7.2 Key Metrics and Abilities
Existing research often lacks rigorous evaluat... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
ensemble methods, random forests [61] and gradient boosting
trees [62]. Specifically, we compare relative feature impor-
tances in determining image aesthetics, visual sentiment
and memorability scores. The tree-based regression models
automatically provide feature importances based on the
contribution of different feat... | A_Deep_Learning_Perspective_on_Beauty_Sentiment_and_Remembrance_of_Art |
Yuchen Zhang, Angela Fan, Melanie Kambadur, Sharan
Narang, Aurelien Rodriguez, Robert Stojnic, Sergey Edunov,
and Thomas Scialom. Llama 2: Open foundation and fine-
tuned chat models, 2023. 1 | DiffusionModelAlignmentUsing Direct Preference Optimization |
facebook/detr-resnet-101 model. The box was drawn on the image, which is located at [Image-4]. Then, I used the predictions of a image classification model google/vit-base-patch16-224 and a image caption model nlpconnect/vit-gpt2-image-captioning model to generate the caption for newly generated image. It... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
collections of modular/highly specialized systems that don’t together constitute an APS system;
and/or using neural networks that aren’t, in the predictively relevant sense sketched in 2.1.2-3, agentic
planning and strategically aware. (To be clear: I expect non-APS systems to play a key role in the
economy regardless;... | Is Power-Seeking AI an Existential Risk? |
In interpretable NLP, we require faithful ratio-
nales that reflect the model’s decision-making
process for an explained instance. While prior
work focuses on extractive rationales (a subset
of the input words), we investigate their less-
studied counterpart: free-text natural language
rationales. We demonstrate that p... | Measuring Association Between Labels and Free-Text Rationales |
a) Decoder-only models have been gradually dominating the development of LLMs. At the early stage of LLMs
development, decoder-only models were not as popular as encoder-only and encoder-decoder models. However,
after 2021, with the introduction of game-changing LLMs - GPT-3, decoder-only models experienced a significa... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
(cid:88)
p(xπ1 , . . . , xπi ) = pdown(nr) · pnr (x) =
n∈ϕsum(p,v)
pdown(n) · pn(x)
• Inductive case: Suppose v is an ancestor of vr,i and the parent vtree node vp of v satisfy Eq. (3).
We have
p(xπ1, . . . , xπi) =
m∈ϕsum(p,vp)
pdown(m) · pm(x)
(cid:88)
(cid:88)
(cid:88)
(cid:88)
(cid:88)
(cid:88)
(cid:88)
(c... | LOSSLESS COMPRESSION WITH PROBABILISTIC CIRCUITS |
approaches to observe how specific conditions, such as the occurrence of black swan events, affect
the state of society. Through this, humans can draw better experiences and insights to improve the
harmony of real-world societies. | TheRiseandPotentialofLargeLanguageModel BasedAgents |
// 2) · a_i , ... , in the town ((i + n - 2) mod n + 1) – n · a_i minutes .
//
// You are given an array of b integer numbers , where b_i is the total duration
// of concerts in the i-th town . Reconstruct any correct sequence of positive
// integers a or say that it is impossible .
//
// Input
//
// The first line con... | alphacode |
Weeks, B. E., & Garrett, R. K. (2014). Electoral consequences of political rumors:
Motivated reasoning, candidate rumors, and vote choice during the 2008 U.S.
presidential election. International Journal of Public Opinion Research, 26(4),
401–422. https://doi.org/10.1093/ijpor/edu005
Wilkes, A. L., & Leatherbarrow, M.... | Social_Media_and_Democracy |
Yonatan Bisk, Rowan Zellers, Ronan Le Bras, Jianfeng Gao, and Yejin Choi. PIQA: Reasoning about Physical
Commonsense in Natural Language. In Thirty-Fourth AAAI Conference on Artificial Intelligence, 2020.
Sidney Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He,
Connor Leahy, K... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
procedure.
Proposal is a solid and convincing framework of a PhD thesis that must underline the originality of a research. It
also must delineate a significant contribution to the existing intellectual knowledge. The proposal either must
challenge or support the existing literature on the pro... | How to Write Your PhD Proposal- A Step-By-Step Guide |
First or second choice
Layer 0
46.5%
44.9%
49.9%
49.5%
46.9%
48.6%
48.2%
49.8%
Layer 15
62.3%
67.0%
66.9%
63.1%
61.9%
61.6%
64.6%
62.1%
Layer 31
52.9%
44.5%
49.2%
52.2%
51.3%
51.8%
53.6%
51.8%
Table 5: Percentage of expert assignment repetitions. We evaluate the proportion of times the same expert is
assigned to a ... | Mixtral of Experts paper |
33 How People Can Create - And Destroy - Value with Generative AI, BCG, 2023.
Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge
Worker Productivity and Quality, Dell'Acqua et al., 2023.
34 Schwarcz and Choi, 2023;A&O announce exclusive launch partnership with... | Capabilities and risks from frontier AI |
RGB Images
Depth Maps
SDv1.4
LDM3D (Ours)
DPT-Large
Figure 3. Qualitative comparison of images to Stable diffusion
v1.4 [20] and depth maps to DPT-Large [18, 19], on 512 × 512
images from the COCO validation dataset. Captions from top to
bottom:”a close up of a sheet of pizza on a table”, ”A picture of
some lemons... | LDM3D- Latent Diffusion Model for 3D |
quirk such as a dataset’s reference transcripts seperating
contractions from words with whitespace. We caution this
development procedure comes at a risk of overfitting to the
transcription style of Whisper models which we investigate
in Section 4.4. We are releasing the code for our text nor-
malizer to allow for easy ... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
[18] Sercan Ö Arık, Mike Chrzanowski, Adam Coates, Gregory Diamos, Andrew Gibiansky, Yongguo Kang, Xian Li,
John Miller, Andrew Ng, Jonathan Raiman, et al. 2017. Deep voice: Real-time neural text-to-speech. In International
conference on machine learning. PMLR, 195–204.
[19] Kartik Audhkhasi, George Saon, Zoltán Tüske... | AReviewofDeepLearningTechniquesforSpeechProcessing |
[65] Isbell, C., C. R. Shelton, M. Kearns, et al. A social reinforcement learning agent.
In
Proceedings of the fifth international conference on Autonomous agents, pages 377–384. 2001.
[66] Watkins, C. J. C. H. Learning from delayed rewards, 1989.
[67] Rummery, G. A., M. Niranjan. On-line Q-learning using connection... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
2.3 Models Deployed to the Feedback Interface and Associated Data Distributions
For data collection we predominantly11 used 52B language models with the broad specifications given in
[Askell et al., 2021]. We used three classes of models in our interface:
• HHH Context-Distilled 52B Language Model: At the beginning of... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
i.e., in front of a white background with controlled light-
ing, and in various everyday outfits and “fashion” poses.
(3) Subjects upload full-body photos of themselves taken in
the wild. For each subject we take up to 111 photos in lab
settings, and collect up to 126 in-the-wild photos. In total,
HBW has 2543 photos, ... | Accurate 3D Body Shape Regression using Metric and Semantic Attributes |
Huiwen Chang, Han Zhang, Jarred Barber, Aaron
Maschinot, José Lezama, Lu Jiang, Ming-Hsuan Yang,
Kevin Murphy, William T. Freeman, Michael Rubin-
stein, Yuanzhen Li, and Dilip Krishnan. 2023. Muse:
Text-to-image generation via masked generative trans-
formers. CoRR, abs/2301.00704.
Sheng-Kuan Chung. 2006. Digital stor... | Moûsai |
QLORA’s efficiency enables us to perform an in-depth study of instruction finetuning and chatbot
performance on model scales that would be impossible using regular finetuning due to memory
overhead. Therefore, we train more than 1,000 models across several instruction tuning datasets,
model architectures, and sizes bet... | QLORA |
[2] Jonathan T Barron, Ben Mildenhall, Dor Verbin, Pratul P
Srinivasan, and Peter Hedman. Mip-NeRF 360: Unbounded
anti-aliased neural radiance fields. In Proc. Computer Vision
and Pattern Recognition (CVPR), 2022.
[3] Mojtaba Bemana, Karol Myszkowski, Hans-Peter Seidel, and
Tobias Ritschel. X-fields: Implicit neural vie... | DynIBaR-NeuralDynamicImage-BasedRendering |
Theorem 64. Properties PL↓, PL↑, PW↓, PW↑, P↓, P↑, PS↓ and PS↑ are transitive.
Proof. We show only the downwards cases. The upwards cases are analogous. Let G1 = (cid:3)S1, E1(cid:4), G2 = (cid:3)S2, E2(cid:4) and
G3 = (cid:3)S3, E3(cid:4) be arbitrary STGs. Let τ1 = (cid:3) f1, R1(cid:4) from G1 to ... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.