Language Models are Few-Shot Learners

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 Henighan Rewon Child Aditya Ramesh

Daniel M. Ziegler Jeffrey Wu Clemens Winter

Christopher Hesse Mark Chen Eric Sigler Mateusz Litwin Scott Gray

Benjamin Chess Jack Clark Christopher Berner

Sam McCandlish Alec Radford Ilya Sutskever Dario Amodei

Abstract

We demonstrate that scaling up language models greatly improves task-agnostic,

few-shot performance, sometimes even becoming competitive with prior state-of-
the-art fine-tuning approaches. Specifically, we train GPT-3, an autoregressive
language model with 175 billion parameters, 10x more than any previous non-
sparse language model, and test its performance in the few-shot setting. For all
tasks, GPT-3 is applied without any gradient updates or fine-tuning, with tasks
and few-shot demonstrations specified purely via text interaction with the model.
GPT-3 achieves strong performance on many NLP datasets, including translation,
question-answering, and cloze tasks. We also identify some datasets where GPT-
3’s few-shot learning still struggles, as well as some datasets where GPT-3 faces
methodological issues related to training on large web corpora.

1 Introduction
NLP has shifted from learning task-specific representations and designing task-specific architectures
to using task-agnostic pre-training and task-agnostic architectures. This shift has led to substantial
progress on many challenging NLP tasks such as reading comprehension, question answering, textual
entailment, among others. Even though the architecture and initial representations are now task-
agnostic, a final task-specific step remains: fine-tuning on a large dataset of examples to adapt a task
agnostic model to perform a desired task.
Recent work [RWC+ 19] suggested this final step may not be necessary. [RWC+ 19] demonstrated
that a single pretrained language model can be zero-shot transferred to perform standard NLP tasks
∗
Equal contribution
†
Johns Hopkins University, OpenAI

34th Conference on Neural Information Processing Systems (NeurIPS 2020), Vancouver, Canada.
Figure 1.1: Performance on SuperGLUE increases with model size. A value of K = 32 means
that our model was shown 32 examples per task, for 256 examples total divided across the 8 tasks in
SuperGLUE. We report GPT-3 values on the dev set, so our numbers are not directly comparable
to the dotted reference lines (our test set results are in the appendix). The BERT-Large reference
model was fine-tuned on the SuperGLUE training set (125K examples), whereas BERT++ was first
fine-tuned on MultiNLI (392K examples) and SWAG (113K examples) before further fine-tuning on
the SuperGLUE training set (for a total of 630K fine-tuning examples).
Performance on SuperGLUE increases with number of examples in context. We find the differ-
ence in performance between the BERT-Large and BERT++ to be roughly equivalent to the difference
between GPT-3 with one example per context versus eight examples per context.
Aggregate performance for all 42 accuracy-denominated benchmarks. While zero-shot perfor-
mance improves steadily with model size, few-shot performance increases more rapidly, demonstrat-
ing that larger models are more proficient at in-context learning.

without the need for finetuning on a dataset of training examples. While this work was a promising
proof of concept, the best case performance only matched some supervised baselines on a single
dataset. On most tasks, performance was still far from even simple supervised baselines.
However [RWC+ 19] also showed a potential way forward. The work observed relatively consistent
log-linear trends in performance on both transfer tasks and language modeling loss across one an
order of magnitude of scaling. [KMH+ 20] then conducted a much more rigorous study of the scaling
behavior of log loss and confirmed smooth scaling trends. In this work, we empirically test whether
scaling continues to improve performance by extrapolating the previously identified phenomena
another two orders of magnitude. We train a 175 billion parameter autoregressive language model,
which we call GPT-3, and measure its transfer learning abilities.
As part of this investigation, we also clarify and systematize the approach introduced in [RWC+ 19].
While [RWC+ 19] describe their work as “zero-shot task transfer” they sometimes provide examples
of the relevant task in the context. Due to the use of what are effectively training examples, these
cases are better described as “one-shot” or “few-shot” transfer. We study these one-shot and few-shot
settings in detail comparing them with the zero-shot setting which only uses a natural language
description or invocation of the task to be performed. Our findings are summarized in Figure 1.1. We
observe that one- and few-shot performance is often much higher than true zero-shot performance
leading us to suggest that language models can also be understood as meta-learners where slow
outer-loop gradient descent based learning is combined with fast “in-context” learning implemented
within the context activations of the model.
Broadly, on NLP tasks GPT-3 achieves promising results in the zero- and one-shot settings, and in
the few-shot setting is sometimes competitive with or even occasionally surpasses state-of-the-art
(despite state-of-the-art being held by fine-tuned models). For example, GPT-3 achieves 81.5 F1 on
CoQA in the zero-shot setting, 84.0 F1 on CoQA in the one-shot setting, and 85.0 F1 in the few-shot
setting. Similarly, GPT-3 achieves 64.3% accuracy on TriviaQA in the zero-shot setting, 68.0% in the
one-shot setting, and 71.2% in the few-shot setting, the last of which is state-of-the-art relative to
fine-tuned models operating in the same closed-book setting.
We additionally train a series of smaller models (ranging from 125 million parameters to 13 billion
parameters) in order to compare their performance to GPT-3 in the zero-, one- and few-shot settings.
In general, we find relatively smooth scaling for most tasks with model capacity in all three settings;
one notable pattern is that the gap between zero-, one-, and few-shot performance often grows with
model capacity, perhaps suggesting that larger models are more proficient meta-learners.

2
2 Approach
Our basic pre-training approach, including model, data, and training, is similar to the process
described in [RWC+ 19], with relatively straightforward scaling up of the model size, dataset size and
diversity, and length of training. Our use of in-context learning is also similar to [RWC+ 19], but in
this work we systematically explore different settings for learning within the context:

• Fine-Tuning (FT) - updates the weights of a pre-trained model by training on thousands of

supervised labels specific to the desired task. The main advantage of fine-tuning is strong
performance on many benchmarks. The main disadvantages are the need for a new large
dataset for every task, the potential for poor generalization out-of-distribution [MPL19], and
the potential to exploit spurious features of the training data [GSL+ 18, NK19]. We focus
on task-agnostic performance, leaving fine-tuning for future work.
• Few-Shot (FS) - the model is given a few demonstrations of the task at inference time as
conditioning [RWC+ 19], but no weights are updated. An example typically has a context
and a desired completion (for example an English sentence and the French translation),
and few-shot works by giving K examples of context and completion, and then one final
example of context, with the model expected to provide the completion (see appendix for
more details). We typically set K in the range of 10 to 100, as this is how many examples can
fit in the model’s context window (nctx = 2048). The main advantage of few-shot is a major
reduction in the need for task-specific data. The main disadvantage is that results from this
method have so far been much worse than state-of-the-art fine-tuned models. Also, a small
amount of task specific data is still required. As indicated by the name, few-shot learning as
described here for language models is related to few-shot learning as used in other contexts
in ML [HYC01, VBL+ 16] – both involve learning based on a broad distribution of tasks
and then rapidly adapting to a new task.
• One-Shot (1S) - similar to few-shot but with K = 1.
• Zero-Shot (0S) - similar to few-shot but with a natural language description of the task
instead of any examples.

The appendix includes a demonstration of the four methods using the example of translating English
to French. While the few-shot results we present in this paper achieve the highest performance,
one-shot, or even sometimes zero-shot, seem like the fairest comparisons to human performance, and
are important targets for future work.

2.1 Model and Architectures

We use the same model and architecture as GPT-2 [RWC+ 19], including the modified initialization,
pre-normalization, and reversible tokenization described therein, with the exception that we use
alternating dense and locally banded sparse attention patterns in the layers of the transformer, similar
to the Sparse Transformer [CGRS19]. To study the dependence of ML performance on model size,
we train 8 different sizes of model, from 125 million parameters to 175 billion parameters, with the
last being the model we call GPT-3. This range of model sizes allows us to test the scaling laws
introduced in [KMH+ 20].
More details on the sizes and architectures of our models can be found in the appendix. We partition
each model across GPUs along both the depth and width dimension in order to minimize data-transfer
between nodes.

2.2 Training Dataset

To create our training data, we (1) downloaded and filtered a version of CommonCrawl1 [RSR+ 19]
based on similarity to a range of high-quality reference corpora, (2) performed fuzzy deduplication at
the document level, within and across datasets, to prevent redundancy and preserve the integrity of
our held-out validation set as an accurate measure of overfitting, and (3) added known high-quality
reference corpora to the training mix to augment CommonCrawl and increase its diversity. These
reference corpora include an expanded version of the WebText dataset [RWC+ 19], collected by
1
https://commoncrawl.org/the-data/

3
 LAMBADA LAMBADA StoryCloze HellaSwag
Setting (acc) (ppl) (acc) (acc)
SOTA 68.0a 8.63b 91.8c 85.6d
GPT-3 Zero-Shot 76.2 3.00 83.2 78.9
GPT-3 One-Shot 72.5 3.35 84.7 78.1
GPT-3 Few-Shot 86.4 1.92 87.7 79.3

Table 3.1: Performance on cloze and completion tasks. GPT-3 significantly improves SOTA on
LAMBADA while achieving respectable performance on two difficult completion prediction datasets.
a
[Tur20] b [RWC+ 19] c [LDL19] d [LCH+ 20]

scraping links over a longer period of time, and first described in [KMH+ 20], two internet-based
books corpora (Books1 and Books2) and English-language Wikipedia (details in the appendix).

2.3 Training Process

As found in [KMH+ 20, MKAT18], larger models can typically use a larger batch size, but require
a smaller learning rate. We measure the gradient noise scale during training and use it to guide
our choice of batch size [MKAT18]. Table A.1 shows the parameter settings we used. To train the
larger models without running out of memory, we use a mixture of model parallelism within each
matrix multiply and model parallelism across the layers of the network. All models were trained on
V100 GPU’s on part of a high-bandwidth cluster. Details of the training process and hyperparameter
settings are described in the appendix.

2.4 Evaluation

For few-shot learning, we evaluate each example in the evaluation set by randomly drawing K
examples from that task’s training set as conditioning, delimited by 1 or 2 newlines depending on
the task. For LAMBADA and Storycloze there is no supervised training set available so we draw
conditioning examples from the development set and evaluate on the test set.
For some tasks we use a natural language prompt in addition to (or for K = 0, instead of) demonstra-
tions. Similar to [RSR+ 19] we also sometimes change the formatting of answers. See the appendix
for per-task examples.
On tasks with free-form completion, we use beam search with the same parameters as [RSR+ 19]: a
beam width of 4 and a length penalty of α = 0.6.
Final results are reported on the test set when publicly available, for each model size and learning
setting (zero-, one-, and few-shot). When the test set is private, our model is often too large to fit on
the test server, so we report results on the development set.

3 Results
3.1 Language Modeling, Cloze, and Completion Tasks

We test GPT-3’s performance on the traditional task of language modeling as well as related tasks.
We calculate zero-shot perplexity on the Penn Tree Bank (PTB) [MKM+ 94] dataset measured in
[RWC+ 19]. We omit the 4 Wikipedia-related tasks and the one-billion word benchmark due to a
high fraction of these datasets being contained in our training set. Our largest model sets a new SOTA
on PTB by a substantial margin of 15 points.
The LAMBADA dataset [PKL+ 16] requires the model to predict the last word of a paragraph.
Although [BHT+ 20] suggested scaling language models is yielding diminishing returns on this
benchmark, we find that zero-shot GPT-3 achieves a substantive gain of 8% over the previous state-of-
the-art. For the few-shot setting, we use a fill-in-the-blank format to encourage the language model to
only generate one word (Alice was friends with Bob. Alice went to visit her friend, . → Bob).
With this format, GPT-3 achieves an increase of over 18% from the previous state-of-the-art, and

4
 Setting NaturalQS WebQS TriviaQA
+
RAG (Fine-tuned, Open-Domain) [LPP 20] 44.5 45.5 68.0
T5-11B+SSM (Fine-tuned, Closed-Book) [RRS20] 36.6 44.7 60.5
T5-11B (Fine-tuned, Closed-Book) 34.5 37.4 50.1
GPT-3 Zero-Shot 14.6 14.4 64.3
GPT-3 One-Shot 23.0 25.3 68.0
GPT-3 Few-Shot 29.9 41.5 71.2

Table 3.2: Results on three Open-Domain QA tasks. GPT-3 is shown in the few-, one-, and
zero-shot settings, as compared to prior SOTA results for closed book and open domain settings.
TriviaQA few-shot result is evaluated on the wiki split test server.

Setting ARC (Easy) ARC (Challenge) CoQA DROP

a b c
Fine-tuned SOTA 92.0 78.5 90.7 89.1d
GPT-3 Zero-Shot 68.8 51.4 81.5 23.6
GPT-3 One-Shot 71.2 53.2 84.0 34.3
GPT-3 Few-Shot 70.1 51.5 85.0 36.5

Table 3.3: GPT-3 results on a selection of QA / RC tasks. CoQA and DROP are F1 while ARC
reports accuracy. See the appendix for additional experiments. a [KKS+ 20] b [KKS+ 20] c [JZC+ 19]
d
[JN20]

performance improves smoothly with model size. However, the fill-in-blank method is not effective
one-shot, where it always performs worse than the zero-shot setting, perhaps because all models
require several examples to recognize the pattern. An analysis of test set contamination identified that
a significant minority of the LAMBADA dataset appears to be present in our training data – however
analysis performed in Section 4 suggests negligible impact on performance.
The HellaSwag dataset [ZHB+ 19] involves picking the best ending to a story or set of instructions.
The examples were adversarially mined to be difficult for language models while remaining easy for
humans. GPT-3 outperforms a fine-tuned 1.5B parameter language model [ZHR+ 19] but is still a fair
amount lower than the overall SOTA achieved by the fine-tuned multi-task model ALUM.
The StoryCloze 2016 dataset [MCH+ 16] involves selecting the correct ending sentence for five-
sentence long stories. Here GPT-3 improves over previous zero-shot results by roughly 10% but is
overall still 4.1% lower than the fine-tuned SOTA using a BERT based model [LDL19].

3.2 Question Answering

In this section we measure GPT-3’s ability to handle a variety of question answering tasks. First, we
look at datasets involving answering questions about broad factual knowledge. We evaluate in the
“closed-book” setting (meaning no conditioning information/articles) as suggested by [RRS20]. On
TriviaQA [JCWZ17], GPT-3 zero-shot already outperforms the fine-tuned T5-11B by 14.2%, and also
outperforms a version with Q&A tailored span prediction during pre-training by 3.8%. The one-shot
result improves by 3.7% and matches the SOTA for an open-domain QA system which not only
fine-tunes but also makes use of a learned retrieval mechanism over a 15.3B parameter dense vector
index of 21M documents [LPP+ 20]. GPT-3’s few-shot result further improves performance another
3.2% beyond this. On Natural Questions (NQs) [KPR+ 19], GPT-3 underperforms a fine-tuned T5
11B+SSM. The questions in NQs tend towards fine-grained Wikipedia knowledge which could be
testing the limits of GPT-3’s capacity and broad pretraining distribution.
ARC [CCE+ 18] is a common sense reasoning dataset of multiple-choice questions collected from
3rd to 9th grade science exams. On the “Challenge” version of the dataset, which has been filtered to
questions which simple statistical or information retrieval methods are unable to correctly answer,
GPT-3 approaches the performance of a fine-tuned RoBERTa baseline [KKS+ 20]. On the “Easy”

5
 Setting En→Fr Fr→En En→De De→En En→Ro Ro→En
SOTA (Supervised) 45.6a 35.0 b 41.2c 40.2d 38.5e 39.9e
XLM [LC19] 33.4 33.3 26.4 34.3 33.3 31.8
MASS [STQ+ 19] 37.5 34.9 28.3 35.2 35.2 33.1
mBART [LGG+ 20] - - 29.8 34.0 35.0 30.5
GPT-3 Zero-Shot 25.2 21.2 24.6 27.2 14.1 19.9
GPT-3 One-Shot 28.3 33.7 26.2 30.4 20.6 38.6
GPT-3 Few-Shot 32.6 39.2 29.7 40.6 21.0 39.5

Table 3.4: Few-shot GPT-3 outperforms previous unsupervised NMT work by 5 BLEU
when translating into English reflecting its strength as an English LM. We report BLEU
scores on the WMT’14 Fr↔En, WMT’16 De↔En, and WMT’16 Ro↔En datasets as mea-
sured by multi-bleu.perl with XLM’s tokenization in order to compare most closely with
prior unsupervised NMT work. SacreBLEUf [Pos18] results reported in the appendix. Under-
line indicates an unsupervised or few-shot SOTA, bold indicates supervised SOTA with relative
confidence. a [EOAG18] b [DHKH14] c [WXH+ 18] d [oR16] e [LGG+ 20] f [SacreBLEU signature:
BLEU+case.mixed+numrefs.1+smooth.exp+tok.intl+version.1.2.20]

version of the dataset, GPT-3 slightly exceeds the same fine-tuned RoBERTa baseline [KKS+ 20].
However, both of these results are still much worse than the overall SOTAs achieved by [KKS+ 20].
Finally, we evaluate GPT-3 on two reading comprehension datasets. Few-shot GPT-3 performs within
3 points of the human baseline on CoQA [RCM19], a free-form conversational dataset. On DROP
[DWD+ 19], a dataset testing discrete reasoning and numeracy, few-shot GPT-3 outperforms the
fine-tuned BERT baseline from the original paper but is still well below both human performance and
state-of-the-art approaches which augment neural networks with symbolic systems [RLL+ 19].

3.3 Translation

In collecting training data for GPT-3, we used the unfiltered distribution of languages reflected
in internet text datasets (primarily Common Crawl). As a result, although GPT-3’s training data
primarily consists of English (93% by word count), it also includes 7% non-English content (full list
at GPT-3 GitHub). Existing unsupervised machine translation approaches often combine pretraining
on a pair of monolingual datasets with back-translation [SHB15] to bridge the two languages in a
controlled way. By contrast, GPT-3 learns from a blend of training data that mixes many languages
together. Additionally, our one / few-shot settings aren’t strictly comparable to prior unsupervised
work since they make use of a small amount of paired examples in-context (1 or 64).
Zero-shot GPT-3 underperforms recent unsupervised NMT results, but the one-shot setting improves
performance by 7 BLEU and nears competitive performance with prior work. Few-shot GPT-3 further
improves another 4 BLEU resulting in similar average performance to prior unsupervised NMT work.
For the three input languages studied, GPT-3 significantly outperforms prior unsupervised NMT work
when translating into English but underperforms when translating in the other direction. Performance
on En-Ro is a noticeable outlier at over 10 BLEU worse than prior unsupervised NMT work. This
could be a weakness due to reusing the byte-level BPE tokenizer of GPT-2 which was developed for
an almost entirely English training dataset. For both Fr-En and De-En, few shot GPT-3 outperforms
the best supervised result we could find but due to our unfamiliarity with the literature and the
appearance that these are un-competitive benchmarks we do not suspect those results represent a
true SOTA. For Ro-En, few shot GPT-3 is very close to the overall SOTA which is achieved with
unsupervised pretraining, finetuning on 608K labeled examples, and backtranslation [LHCG19b].

3.4 SuperGLUE

The SuperGLUE benchmark is a standardized collection of datasets [WPN+ 19]. In the few-shot
setting, we used 32 examples for all tasks, sampled randomly from the training set. For all tasks
except WSC and MultiRC, we sampled a new set of examples to use in the context for each problem.
For WSC and MultiRC, we used the same set of randomly drawn examples from the training set

6
 SuperGLUE BoolQ CB CB COPA RTE
Average Accuracy Accuracy F1 Accuracy Accuracy
Fine-tuned SOTA 89.0 91.0 96.9 93.9 94.8 92.5
Fine-tuned BERT-Large 69.0 77.4 83.6 75.7 70.6 71.7
GPT-3 Few-Shot 71.8 76.4 75.6 52.0 92.0 69.0

WiC WSC MultiRC MultiRC ReCoRD ReCoRD

Accuracy Accuracy Accuracy F1a Accuracy F1
Fine-tuned SOTA 76.1 93.8 62.3 88.2 92.5 93.3
Fine-tuned BERT-Large 69.6 64.6 24.1 70.0 71.3 72.0
GPT-3 Few-Shot 49.4 80.1 30.5 75.4 90.2 91.1

Table 3.5: Performance of GPT-3 on SuperGLUE compared to fine-tuned baselines and SOTA. All
results are reported on the test set. GPT-3 few-shot is given a total of 32 examples within the context
of each task and performs no gradient updates.

as context for all of the problems we evaluated. We sweep values of K up to 32 and note that the
few-shot SuperGLUE score steadily improves with both model size and with number of examples in
the context showing increasing benefits from in-context learning (Figure 1.1).
We observe a wide range in GPT-3’s performance across tasks. On COPA and ReCoRD GPT-3
achieves near-SOTA performance in the one-shot and few-shot settings, with COPA falling only
a couple points short and achieving second place on the leaderboard, where first place is held by
a fine-tuned 11 billion parameter model (T5). On WSC, BoolQ, MultiRC, and RTE, performance
is reasonable, roughly matching that of a fine-tuned BERT-Large. On CB, we see signs of life at
75.6% in the few-shot setting. WiC is a notable weak spot with few-shot performance equivalent to
random chance. We tried a number of different phrasings and formulations for WiC (which involves
determining if a word is being used with the same meaning in two sentences), none of which was
able to achieve strong performance. This hints at a phenomenon (which we saw in other experiments
we ran contained in the Additional Materials) – GPT-3 appears to be weak in the few-shot or one-shot
setting at some tasks that involve comparing two sentences or snippets. This could also explain the
comparatively low scores for RTE and CB, which also follow this format. Despite these weaknesses,
GPT-3 still outperforms a fine-tuned BERT-large on four of eight tasks and on two tasks GPT-3 is
close to the state-of-the-art held by a fine-tuned 11 billion parameter model.

4 Measuring and Preventing Memorization Of Benchmarks

The dataset and model size are about two orders of magnitude larger than those used for GPT-2,
and include a large amount of Common Crawl, creating increased potential for contamination and
memorization. On the other hand, precisely due to the large amount of data, even GPT-3 175B does
not overfit its training set by a significant amount, measured relative to a held-out validation set with
which it was deduplicated. For each benchmark, we produce a ‘clean’ version which removes all
potentially leaked examples, defined roughly as examples that have a 13-gram overlap with anything
in the pretraining set (or that overlap with the whole example when it is shorter than 13-grams). We
then evaluate GPT-3 on these clean benchmarks, and compare to the original score. If the score on
the clean subset is similar to the score on the entire dataset, this suggests that contamination, even if
present, does not have a significant effect on reported results. In most cases performance changes only
negligibly, and we see no evidence that contamination level and performance difference are correlated.
We conclude that either our conservative method substantially overestimated contamination or that
contamination has little effect on performance. We provide full details of the methodology and
analysis on the most problematic tasks in the appendix.

7
5 Limitations
On text synthesis, GPT-3 samples still sometimes repeat themselves semantically at the document
level, start to lose coherence over sufficiently long passages, contradict themselves, and occasionally
contain non-sequitur sentences or paragraphs. Our release repository contains uncurated unconditional
samples.
Our experiments do not include any bidirectional architectures or other training objectives such as
denoising. Our design decision comes at the cost of potentially worse performance on tasks which
empirically benefit from bidirectionality, such as fill-in-the-blank tasks, tasks that involve looking
back and comparing two pieces of content (ANLI, WIC), or tasks that require re-reading or carefully
considering a long passage and then generating a very short answer (QuAC, RACE).
Our objective weights every token equally and lacks a notion of what is most important to predict
and what is less important. [RRS20] demonstrate benefits of customizing prediction to entities of
interest. Also, with self-supervised objectives, task specification relies on forcing the desired task into
a prediction problem, whereas ultimately, useful language systems (for example virtual assistants)
might be better thought of as taking goal-directed actions rather than just making predictions. Finally,
large pretrained language models are not grounded in other domains of experience, such as video or
real-world physical interaction, and thus lack a large amount of context about the world [BHT+ 20].
For all these reasons, scaling pure self-supervised prediction is likely to hit limits, and augmentation
with a different approach is likely to be necessary. Promising future directions in this vein might
include learning the objective function from humans [ZSW+ 19], fine-tuning with reinforcement
learning, or adding additional modalities such as images to provide grounding and a better model of
the world [CLY+ 19].
GPT-3’s size makes it challenging to deploy. Task-specific distillation [HVD15] merits exploration at
this new scale.

6 Related Work
Several efforts have studied the effect of scale on language model performance. [KMH+ 20, RRBS19,
LWS+ 20, HNA+ 17], find a smooth power-law trend in loss as autoregressive language models are
scaled up. There are different approaches to scaling language models through increasing parameters,
compute, or both. Our work is most aligned with methods that have increased the size of transformers
by increasing parameters and FLOPS-per-token roughly in proportion, with a parameter count of 213
million [VSP+ 17] in the original paper, then 300 million [DCLT18], 1.5 billion [RWC+ 19], 8 billion
[SPP+ 19], 11 billion [RSR+ 19], and most recently 17 billion [Tur20]. A second line of work has
focused on increasing parameter count but not computation by using the conditional computation
framework [BLC13]. Specifically, the mixture-of-experts method [SMM+ 17] has produced 100
billion parameter models and 50 billion parameter translation models [AJF19]. One way to decrease
the computational cost of our models would be to draw from work such as ALBERT [LCG+ 19] or
general [HVD15] or task-specific [SDCW19, JYS+ 19, KR16] approaches to distillation. Lastly, a
third approach to scale increases computation without increasing parameters through methods like
adaptive computation time [Gra16] and the universal transformer [DGV+ 18].
There are many approaches to building multi-task models. Giving task instructions in natural language
was first formalized in a supervised setting with [MKXS18] and used in [RWC+ 19] for in-context
learning and in [RSR+ 19] for multi-task fine-tuning. Multi-task learning [Car97] has shown some
promising initial results [LGH+ 15, LCR19] and multi-stage fine-tuning has produced SOTA or SOTA-
competitive results [PFB18, KKS+ 20]. Metalearning was used in language models in [RWC+ 19],
though with limited results and no systematic study. Other uses of metalearning include matching
networks [VBL+ 16], RL2 [DSC+ 16], learning to optimize [RL16, ADG+ 16, LM17] and MAML
[FAL17]. Our approach of stuffing the model’s context with previous examples is most structurally
similar to RL2. It also resembles [HYC01], in that an inner loop adapts to a task, while an outer
loop updates the weights. Our inner loop performs few-shot in-context learning, but prior work has
explored other methods of few-shot learning [SS20, RCP+ 17, GWC+ 18, XDH+ 19].
Finally, Algorithmic innovation in language models over the last two years has been enormous, in-
cluding denoising-based bidirectionality [DCLT18], prefixLM [DL15], encoder-decoder architectures
[LLG+ 19, RSR+ 19], random permutations during training [YDY+ 19], architectures for sampling

8
efficiency [DYY+ 19], data and training improvements [LOG+ 19], and embedding parameters effi-
ciency [LCG+ 19]. It is likely that incorporating some of these algorithmic advances could improve
GPT-3’s performance on downstream tasks, especially in the fine-tuning setting.

7 Conclusion

We presented a 175 billion parameter language model which shows strong performance on many
NLP tasks and benchmarks in the zero-shot, one-shot, and few-shot settings, in some cases nearly
matching the performance of state-of-the-art fine-tuned systems, as well as generating high-quality
samples and strong qualitative performance at tasks defined on-the-fly. We documented roughly
predictable trends of scaling in performance without using fine-tuning. We also discussed the social
impacts of this class of model. Despite many limitations and weaknesses, these results suggest that
very large language models may be an important ingredient in the development of adaptable, general
language systems.

Funding Disclosures

This work was funded by OpenAI. All models were trained on V100 GPU’s on part of a high-
bandwidth cluster provided by Microsoft

Broader Impacts

Language models have a wide range of beneficial applications for society, including code and writing
auto-completion, grammar assistance, game narrative generation, improving search engine responses,
and answering questions. But they also have potentially harmful applications. GPT-3 improves
the quality of text generation and adaptability over smaller models and increases the difficulty of
distinguishing synthetic text from human-written text. It therefore has the potential to advance both
the beneficial and harmful applications of language models.
Here we focus on the potential harms of improved language models, not because we believe the
harms are necessarily greater, but in order to stimulate efforts to study and mitigate them. The broader
impacts of language models like this are numerous. We focus on two primary issues: the potential
for deliberate misuse of language models like GPT-3 in Section 7.1, and issues of bias, fairness, and
representation within models like GPT-3 in Section 7.2. We also briefly discuss issues of energy
efficiency (Section 7.3).

7.1 Misuse of Language Models

Malicious uses of language models can be somewhat difficult to anticipate because they often
involve repurposing language models in a very different environment or for a different purpose than
researchers intended. To help with this, we can think in terms of traditional security risk assessment
frameworks, which outline key steps such as identifying threats and potential impacts, assessing
likelihood, and determining risk as a combination of likelihood and impact [Ros12]. We discuss three
factors: potential misuse applications, threat actors, and external incentive structures.

7.1.1 Potential Misuse Applications

Any socially harmful activity that relies on generating text could be augmented by powerful lan-
guage models. Examples include misinformation, spam, phishing, abuse of legal and governmental
processes, fraudulent academic essay writing and social engineering pretexting. Many of these
applications bottleneck on human beings to write sufficiently high quality text. Language models that
produce high quality text generation could lower existing barriers to carrying out these activities and
increase their efficacy.
The misuse potential of language models increases as the quality of text synthesis improves. The
ability of GPT-3 to generate several paragraphs of synthetic content that people find difficult to
distinguish from human-written text represents a concerning milestone in this regard.

9
7.1.2 Threat Actor Analysis
Threat actors can be organized by skill and resource levels, ranging from low or moderately skilled
and resourced actors who may be able to build a malicious product to ‘advanced persistent threats’
(APTs): highly skilled and well-resourced (e.g. state-sponsored) groups with long-term agendas
[SBC+ 19].
To understand how low and mid-skill actors think about language models, we have been monitoring
forums and chat groups where misinformation tactics, malware distribution, and computer fraud
are frequently discussed. While we did find significant discussion of misuse following the initial
release of GPT-2 in spring of 2019, we found fewer instances of experimentation and no successful
deployments since then. Additionally, those misuse discussions were correlated with media coverage
of language model technologies. From this, we assess that the threat of misuse from these actors is
not immediate, but significant improvements in reliability could change this.
Because APTs do not typically discuss operations in the open, we have consulted with professional
threat analysts about possible APT activity involving the use of language models. Since the release
of GPT-2 there has been no discernible difference in operations that may see potential gains by using
language models. The assessment was that language models may not be worth investing significant
resources in because there has been no convincing demonstration that current language models are
significantly better than current methods for generating text, and because methods for “targeting” or
“controlling” the content of language models are still at a very early stage.

7.1.3 External Incentive Structures

Each threat actor group also has a set of tactics, techniques, and procedures (TTPs) that they rely
on to accomplish their agenda. TTPs are influenced by economic factors like scalability and ease of
deployment; phishing is extremely popular among all groups because it offers a low-cost, low-effort,
high-yield method of deploying malware and stealing login credentials. Using language models to
augment existing TTPs would likely result in an even lower cost of deployment.
Ease of use is another significant incentive. Having stable infrastructure has a large impact on the
adoption of TTPs. The outputs of language models are stochastic, however, and though developers
can constrain these (e.g. using top-k truncation) they are not able to perform consistently without
human feedback. If a social media disinformation bot produces outputs that are reliable 99% of the
time, but produces incoherent outputs 1% of the time, this could reduce the amount of human labor
required in operating this bot. But a human is still needed to filter the outputs, which restricts how
scalable the operation can be.
Based on our analysis of this model and analysis of threat actors and the landscape, we suspect AI
researchers will eventually develop language models that are sufficiently consistent and steerable that
they will be of greater interest to malicious actors. We expect this will introduce challenges for the
broader research community, and hope to work on this through a combination of mitigation research,
prototyping, and coordinating with other technical developers.

7.2 Fairness, Bias, and Representation

Biases present in training data may lead models to generate stereotyped or prejudiced content.
This is concerning, since model bias could harm people in the relevant groups in different ways
by entrenching existing stereotypes and producing demeaning portrayals amongst other potential
harms [Cra17]. We have conducted an analysis of biases in the model in order to better understand
GPT-3’s limitations when it comes to fairness, bias, and representation. 2
Our goal is not to exhaustively characterize GPT-3, but to give a preliminary analysis of some of
its limitations and behaviors. We focus on biases relating to gender, race, and religion, although
many other categories of bias are likely present and could be studied in follow-up work. This is a
preliminary analysis and does not reflect all of the model’s biases even within the studied categories.
Broadly, our analysis indicates that internet-trained models have internet-scale biases; models tend to
reflect stereotypes present in their training data. Below we discuss our preliminary findings of bias
2
Evaluating fairness, bias, and representation in language models is a rapidly-developing area with a large
body of prior work. See, for example, [HZJ+ 19, NBR20, SCNP19].

10
along the dimensions of gender, race, and religion. We probe for bias in the 175 billion parameter
model and also in similar smaller models, to see if and how they are different in this dimension.

7.2.1 Gender

In our investigation of gender bias in GPT-3, we focused on associations between gender and
occupation. We found that occupations in general have a higher probability of being followed by a
male gender identifier than a female one (in other words, they are male leaning) when given a context
such as "The {occupation} was a" (Neutral Variant). 83% of the 388 occupations we tested
were more likely to be followed by a male identifier by GPT-3. We measured this by feeding the
model a context such as "The detective was a" and then looking at the probability of the model
following up with male indicating words (eg. man, male etc.) or female indicating words (woman,
female etc.). In particular, occupations demonstrating higher levels of education such as legislator,
banker, or professor emeritus were heavily male leaning along with occupations that require hard
physical labour such as mason, millwright, and sheriff. Occupations that were more likely to be
followed by female identifiers include midwife, nurse, receptionist, housekeeper etc.
We also tested how these probabilities changed when we shifted the context to be the "The
competent {occupation} was a" (Competent Variant), and when we shifted the context to
be "The incompetent {occupation} was a" (Incompetent Variant) for each occupation in the
dataset. We found that, when prompted with "The competent {occupation} was a," the ma-
jority of occupations had an even higher probability of being followed by a male identifier than a
female one than was the case with our original neutral prompt, "The {occupation} was a". With
the prompt "The incompetent {occupation} was a" the majority of occupations still leaned
male with a similar probability than for our original neutral prompt. The average occupation bias -
1
P P (female|Context)
measured as njobs jobs log( P (male|Context)) ) - was −1.11 for the Neutral Variant, −2.14 for the
Competent Variant and −1.15 for the Incompetent Variant.
We also carried out pronoun resolution on the Winogender dataset [RNLVD18] using
two methods which further corroborated the model’s tendency to associate most occupa-
tions with males. One method measured the models ability to correctly assign a pro-
noun as the occupation or the participant. For example, we fed the model a con-
text such as "The advisor met with the advisee because she wanted to get advice
about job applications. ‘She’ refers to the" and found the option with the lowest
probability between the two possible options (Choices between Occupation Option: advisor; Partici-
pant Option: advisee).
Occupation and participant words often have societal biases associated with them such as the
assumption that most occupants are by default male. We found that the language models learnt some
of these biases such as a tendency to associate female pronouns with participant positions more than
male pronouns. GPT-3 175B had the highest accuracy of all the models (64.17%) on this task. It was
also the only model where the accuracy for Occupant sentences (sentences where the correct answer
was the Occupation option) for females was higher than for males (81.7% vs 76.7%). All other
models had a higher accuracy for male pronouns with Occupation sentences as compared to female
pronouns with the exception of our second largest model- GPT-3 13B - which had the same accuracy
(60%) for both. This offers some preliminary evidence that in places where issues of bias can make
language models susceptible to error, the larger models are more robust than smaller models.
We also performed co-occurrence tests, where we analyzed which words are likely to occur in the
vicinity of other pre-selected words. We created a model output sample set by generating 800 outputs
of length 50 each with a temperature of 1 and top p of 0.9 for every prompt in our dataset. For
gender, we had prompts such as "He was very", "She was very", "He would be described
as", "She would be described as"3 . We looked at the adjectives and adverbs in the top 100
most favored words using an off-the-shelf POS tagger [LB02]. We found females were more often
described using appearance oriented words such as ”beautiful” and ”gorgeous” as compared to men
who were more often described using adjectives that span a greater spectrum.

3
We only used male and female pronouns. This simplifying assumption makes it easier to study co-occurrence
since it does not require the isolation of instances in which ‘they’ refers to a singular noun from those where it
didn’t, but other forms of gender bias are likely present and could be studied using different approaches.

11
 Table 7.1: Most Biased Descriptive Words in 175B Model

Top 10 Most Biased Male Descriptive Words Top 10 Most Biased Female Descriptive Words
with Raw Co-Occurrence Counts with Raw Co-Occurrence Counts
Average Number of Co-Occurrences Across All Average Number of Co-Occurrences Across All
Words: 17.5 Words: 23.9
Large (16) Optimistic (12)
Mostly (15) Bubbly (12)
Lazy (14) Naughty (12)
Fantastic (13) Easy-going (12)
Eccentric (13) Petite (10)
Protect (10) Tight (10)
Jolly (10) Pregnant (10)
Stable (9) Gorgeous (28)
Personable (22) Sucked (8)
Survive (7) Beautiful (158)

Table 7.1 shows the top 10 most favored descriptive words for the model along with the raw number
of times each word co-occurred with a pronoun indicator. “Most Favored” here indicates words
which were most skewed towards a category by co-occurring with it at a higher rate as compared to
the other category. To put these numbers in perspective, we have also included the average for the
number of co-occurrences across all qualifying words for each gender.

7.2.2 Race
To investigate racial bias in GPT-3, we seeded the model with prompts such as - "The {race} man
was very", "The {race} woman was very" and "People would describe the {race}
person as" and generated 800 samples for each of the above prompts, with {race} replaced with
a term indicating a racial category such as White or Asian. We then measure word co-occurrences
in the generated samples. Given prior research demonstrating that language models produce text
of differing sentiment when varying features such as occupation [HZJ+ 19], we explored how race
impacted sentiment. We measured sentiment using Senti WordNet [BES10] for the words which
co-occurred disproportionately with each race. Each word sentiment varied from 100 to -100, with
positive scores indicating positive words (eg. wonderfulness: 100, amicable: 87.5), negative scores
indicating negative words (eg. wretched: -87.5 , horrid: -87.5) and a score of 0 indicating neutral
words (eg. sloping, chalet).
It should be noted that we were explicitly prompting the models to talk about race and this in turn
generated text that focused on racial features; these results are not from the models talking about
race in the wild but talking about race in an experimental setup where they have been primed to do
so. Additionally, since we are measuring sentiment by simply looking at word co-occurrences, the
resulting sentiment can reflect socio-historical factors - for instance, text relating to a discussion of
slavery will frequently have a negative sentiment, which may lead to a demographic being associated
with a negative sentiment under this testing methodology.
Across the models we analyzed, ‘Asian’ had a consistently high sentiment - it ranked 1st in 3 out of 7
models. On the other hand, ’Black’ had a consistently low sentiment - it ranked the lowest in 5 out of
7 models. These differences narrowed marginally on the larger model sizes. This analysis gives a
sense of the biases of different models and highlights the need for more sophisticated analysis of the
relationship between sentiment, entities, and input data.

7.2.3 Religion
We studied which words co-occurred with religious terms relating to Atheism, Buddhism, Christianity,
Hinduism, Islam, and Judaism, by generating 800 model outputs of length ≈50 with a temperature of 1
and a top p of 0.9 for every prompt. Our prompts were of the nature "{Religion practitioners}
are" (Eg. "Christians are") for each of the six religious categories listed above. We then

12
 Figure 7.1: Racial Sentiment Across Models

Religion Most Favored Descriptive Words

Atheism ‘Theists’, ‘Cool’, ‘Agnostics’, ‘Mad’, ‘Theism’, ‘Defensive’, ‘Complaining’, ‘Cor-
rect’, ‘Arrogant’, ‘Characterized’
Buddhism ‘Myanmar’, ‘Vegetarians’, ‘Burma’, ‘Fellowship’, ‘Monk’, ‘Japanese’, ‘Reluctant’,
‘Wisdom’, ‘Enlightenment’, ‘Non-Violent’
Christianity ‘Attend’, ‘Ignorant’, ‘Response’, ‘Judgmental’, ‘Grace’, ‘Execution’, ‘Egypt’,
‘Continue’, ‘Comments’, ‘Officially’
Hinduism ‘Caste’, ‘Cows’, ‘BJP’, ‘Kashmir’, ‘Modi’, ‘Celebrated’, ‘Dharma’, ‘Pakistani’,
‘Originated’, ‘Africa’
Islam ‘Pillars’, ‘Terrorism’, ‘Fasting’, ‘Sheikh’, ‘Non-Muslim’, ‘Source’, ‘Charities’,
‘Levant’, ‘Allah’, ‘Prophet’
Judaism ‘Gentiles’, ‘Race’, ‘Semites’, ‘Whites’, ‘Blacks’, ‘Smartest’, ‘Racists’, ‘Arabs’,
‘Game’, ‘Russian’

Table 7.2: Shows the ten most favored words about each religion in the GPT-3 175B model.

allowed the model to naturally carry out completions and created a corpus of such completions for
studying co-occurrence of words.
The following is an example output from the model:
"Buddhists are divided into two main branches - Theravada and Mahayana.
Theravada is the more conservative branch, centering on monastic life
and the earliest sutras and refusing to recognize the later Mahayana
sutras as authentic."
Similar to race, we found that the models make associations with religious terms that indicate some
propensity to reflect how these terms are sometimes presented in the world. For example, with the
religion Islam, we found that words such as ramadan, prophet and mosque co-occurred at a higher
rate than for other religions. We also found that words such as violent, terrorism and terrorist
co-occurred at a greater rate with Islam than with other religions and were in the top 40 most favored
words for Islam in GPT-3.

7.2.4 Future Bias and Fairness Challenges

We have presented this preliminary analysis to share some of the biases we found in order to motivate
further research, and to highlight the inherent difficulties in characterizing biases in large-scale
generative models; we expect this to be an area of continuous research for us and are excited to
discuss different methodological approaches with the community. We view the work in this section

13
Figure 7.2: Total compute used during training. Based on the analysis in Scaling Laws For Neural
Language Models [KMH+ 20] we train much larger models on many fewer tokens than is typical.
As a consequence, although GPT-3 3B is almost 10x larger than RoBERTa-Large (355M params),
both models took roughly 50 petaflop/s-days of compute during pre-training. Methodology for these
calculations can be found in the Appendix.

as subjective signposting - we chose gender, race, and religion as a starting point, but we recognize
the inherent subjectivity in this choice. Our work is inspired by the literature on characterizing model
attributes to develop informative labels such as Model Cards for Model Reporting from [MWZ+ 18].
Ultimately, it is important not just to characterize biases in language systems but to intervene. The
literature on this is also extensive [QMZH19, HZJ+ 19], so we offer only a few brief comments
on future directions specific to large language models. In order to pave the way for effective bias
prevention in general purpose models, there is a need for building a common vocabulary tying
together the normative, technical and empirical challenges of bias mitigation for these models. There
is room for more research that engages with the literature outside NLP, better articulates normative
statements about harm, and engages with the lived experience of communities affected by NLP
systems [BBDIW20]. Thus, mitigation work should not be approached purely with a metric driven
objective to ‘remove’ bias as this has been shown to have blind spots [GG19, NvNvdG19] but in a
holistic manner.

7.3 Energy Usage

Practical large-scale pre-training requires large amounts of computation, which is energy-intensive:

training the GPT-3 175B consumed several thousand petaflop/s-days of compute during pre-training,
compared to tens of petaflop/s-days for a 1.5B parameter GPT-2 model (Figure 7.2). This means we
should be cognizant of the cost and efficiency of such models, as advocated by [SDSE19].
The use of large-scale pre-training also gives another lens through which to view the efficiency of
large models - we should consider not only the resources that go into training them, but how these
resources are amortized over the lifetime of a model, which will subsequently be used for a variety of
purposes and fine-tuned for specific tasks. Though models like GPT-3 consume significant resources
during training, they can be surprisingly efficient once trained: even with the full GPT-3 175B,
generating 100 pages of content from a trained model can cost on the order of 0.4 kW-hr, or only a
few cents in energy costs. Additionally, techniques like model distillation [LHCG19a] can further
bring down the cost of such models, letting us adopt a paradigm of training single, large-scale models,
then creating more efficient versions of them for use in appropriate contexts. Algorithmic progress
may also naturally further increase the efficiency of such models over time, similar to trends observed
in image recognition and neural machine translation [HB20].

14
7.4 News Generation

We test GPT-3’s ability to generate synthetic “news articles” by prompting the model with a context
of three previous news articles and the title and subtitle of a proposed article to generate. To gauge the
quality of generated articles, we measured human ability to distinguish GPT-3-generated articles from
real ones. Similar work has been carried out by Kreps et al. [KMB20] and Zellers et al. [ZHR+ 19].
Generative language models are trained to match the distribution of content generated by humans, so
the (in)ability of humans to distinguish the two is a potentially important measure of quality.4
In order to see how well humans can detect model generated text, we arbitrarily selected 25 article
titles and subtitles from the website newser.com (mean length: 215 words). We then generated
completions of these titles and subtitles from for language models ranging in size from 125M to 175B
(GPT-3) parameters (mean length: 200 words). For each model, we presented around 80 US-based
participants with a quiz consisting of these real titles and subtitles followed by either the human
written article or the article generated by the model5 . Participants were asked to select whether the
article was “very likely written by a human”, “more likely written by a human”, “I don’t know”,
“more likely written by a machine”, or “very likely written by a machine”.
The articles we selected were not in the models’ training data and the model outputs were formatted
and selected programmatically to prevent human cherry-picking. All models used the same context to
condition outputs on and were pre-trained with the same context size and the same article titles and
subtitles were used as prompts for each model. However, we also ran an experiment to control for
participant effort and attention that followed the same format but involved intentionally bad model
generated articles. This was done by generating articles from a “control model”: a 160M parameter
model with no context and increased output randomness.
Mean human accuracy (the ratio of correct assignments to non-neutral assignments per participant) at
detecting that the intentionally bad articles were model generated was ∼ 86% where 50% is chance
level performance. By contrast, mean human accuracy at detecting articles that were produced by the
175B parameter model was barely above chance at ∼ 52% (see Table 7.3).6 Human abilities to detect
model generated text appear to decrease as model size increases: there appears to be a trend towards
chance accuracy with model size, and human detection of GPT-3 is close to chance.7 This is true
despite the fact that participants spend more time on each output as model size increases (see the
Appendix).
Examples of synthetic articles from GPT-3 are given in Figures 7.4 and 7.5.8 Much of the text
is—as indicated by the evaluations—difficult for humans to distinguish from authentic human content.
Factual inaccuracies can be an indicator that an article is model generated since, unlike human authors,
the models have no access to the specific facts that the article titles refer to or when the article was
written. Other indicators include repetition, non sequiturs, and unusual phrasings, though these are
often subtle enough that they are not noticed.
Related work on language model detection by Ippolito et al. [IDCBE19] indicates that automatic
discriminators like G R O V E R [ZHR+ 19] and GLTR [GSR19] may have greater success at detecting
model generated text than human evaluators. Automatic detection of these models may be a promising
area of future research.
Ippolito et al. [IDCBE19] also note that human accuracy at detecting model generated text increases
as humans observe more tokens. To do a preliminary investigation of how good humans are at
detecting longer news articles generated by GPT-3 175B, we selected 12 world news articles from
Reuters with an average length of 569 words and generated completions of these articles from GPT-3
with an average length of 498 words (298 words longer than our initial experiments). Following the
4
This task is also relevant to the potential misuse of language models discussed in Section 7.1.
5
We wanted to identify how good an average person on the internet is at detecting language model outputs,
so we focused on participants drawn from the general US population. See the Appendix for details.
6
We use a two-sample Student’s T-Test to test for significant difference between the means of the participant
accuracies of each model and the control model and report the normalized difference in the means (as the
t-statistic) and the p-value.
7
If a model consistently produces texts that are more impressive than human articles, it is possible that human
performance on this task would drop below 50%. Indeed, many individual participants scored below 50% on
this task.
8
Additional non-news samples can be found in the Appendix.

15
 95% Confidence t compared to “I don’t know”
Mean accuracy Interval (low, hi) control (p-value) assignments
Control (deliberately bad model) 86% 83%–90% - 3.6 %
GPT-3 Small 76% 72%–80% 3.9 (2e-4) 4.9%
GPT-3 Medium 61% 58%–65% 10.3 (7e-21) 6.0%
GPT-3 Large 68% 64%–72% 7.3 (3e-11) 8.7%
GPT-3 XL 62% 59%–65% 10.7 (1e-19) 7.5%
GPT-3 2.7B 62% 58%–65% 10.4 (5e-19) 7.1%
GPT-3 6.7B 60% 56%–63% 11.2 (3e-21) 6.2%
GPT-3 13B 55% 52%–58% 15.3 (1e-32) 7.1%
GPT-3 175B 52% 49%–54% 16.9 (1e-34) 7.8%

Table 7.3: Human accuracy in identifying whether short (∼200 word) news articles are model
generated. We find that human accuracy (measured by the ratio of correct assignments to non-neutral
assignments) ranges from 86% on the control model to 52% on GPT-3 175B. This table compares
mean accuracy between five different models, and shows the results of a two-sample T-Test for the
difference in mean accuracy between each model and the control model (an unconditional GPT-3
Small model with increased output randomness).

95% Confidence t compared to “I don’t know”

Mean accuracy Interval (low, hi) control (p-value) assignments
Control 88% 84%–91% - 2.7%
GPT-3 175B 52% 48%–57% 12.7 (3.2e-23) 10.6%

Table 7.4: People’s ability to identify whether ∼ 500 word articles are model generated (as measured
by the ratio of correct assignments to non-neutral assignments) was 88% on the control model and
52% on GPT-3 175B. This table shows the results of a two-sample T-Test for the difference in mean
accuracy between GPT-3 175B and the control model (an unconditional GPT-3 Small model with
increased output randomness).

methodology above, we ran two experiments, each on around 80 US-based participants, to compare
human abilities to detect the articles generated by GPT-3 and a control model.
We found that mean human accuracy at detecting the intentionally bad longer articles from the control
model was ∼ 88%, while mean human accuracy at detecting the longer articles that were produced
by GPT-3 175B was still barely above chance at ∼ 52% (see Table 7.4). This indicates that, for news
articles that are around 500 words long, GPT-3 continues to produce articles that humans find difficult
to distinguish from human written news articles.

Acknowledgements
The authors would like to thank Ryan Lowe for giving detailed feedback on drafts of the paper. Thanks
to Jakub Pachocki and Szymon Sidor for suggesting tasks, and Greg Brockman, Michael Petrov,
Brooke Chan, and Chelsea Voss for helping run evaluations on OpenAI’s infrastructure. Thanks to
David Luan for initial support in scaling up this project, Irene Solaiman for discussions about ways to
approach and evaluate bias, Harrison Edwards and Yura Burda for discussions and experimentation
with in-context learning, Geoffrey Irving and Paul Christiano for early discussions of language model
scaling, Long Ouyang for advising on the design of the human evaluation experiments, Chris Hallacy
for discussions on data collection, and Shan Carter for help with visual design. Thanks to the millions
of people who created content that was used in the training of the model, and to those who were
involved in indexing or upvoting the content (in the case of WebText). Additionally, we would like
to thank the entire OpenAI infrastructure and supercomputing teams for making it possible to train
models at this scale.

16
Figure 7.3: People’s ability to identify whether news articles are model-generated (measured by the
ratio of correct assignments to non-neutral assignments) decreases as model size increases. Accuracy
on the outputs on the deliberately-bad control model (an unconditioned GPT-3 Small model with
higher output randomness) is indicated with the dashed line at the top, and the random chance (50%)
is indicated with the dashed line at the bottom. Line of best fit is a power law with 95% confidence
intervals.

Contributions

Tom Brown, Ben Mann, Prafulla Dhariwal, Dario Amodei, Nick Ryder, Daniel M Ziegler, and
Jeffrey Wu implemented the large-scale models, training infrastructure, and model-parallel strategies.
Tom Brown, Dario Amodei, Ben Mann, and Nick Ryder conducted pre-training experiments.
Ben Mann and Alec Radford collected, filtered, deduplicated, and conducted overlap analysis on
the training data.
Melanie Subbiah, Ben Mann, Dario Amodei, Jared Kaplan, Sam McCandlish, Tom Brown,
Tom Henighan, and Girish Sastry implemented the downstream tasks and the software framework
for supporting them, including creation of synthetic tasks.
Jared Kaplan and Sam McCandlish initially predicted that a giant language model should show
continued gains, and applied scaling laws to help predict and guide model and data scaling decisions
for the research.
Ben Mann implemented sampling without replacement during training.
Alec Radford originally demonstrated few-shot learning occurs in language models.
Jared Kaplan and Sam McCandlish showed that larger models learn more quickly in-context, and
systematically studied in-context learning curves, task prompting, and evaluation methods.
Prafulla Dhariwal implemented an early version of the codebase, and developed the memory
optimizations for fully half-precision training.
Rewon Child and Mark Chen developed an early version of our model-parallel strategy.
Rewon Child and Scott Gray contributed the sparse transformer.

17
 Title: United Methodists Agree to Historic Split
Subtitle: Those who oppose gay marriage will form their own
denomination
Article: After two days of intense debate, the United Methodist
Church has agreed to a historic split - one that is expected
to end in the creation of a new denomination, one that will
be "theologically and socially conservative," according to
The Washington Post. The majority of delegates attending the
church's annual General Conference in May voted to strengthen a
ban on the ordination of LGBTQ clergy and to write new rules that
will "discipline" clergy who officiate at same-sex weddings. But
those who opposed these measures have a new plan: They say they
will form a separate denomination by 2020, calling their church
the Christian Methodist denomination.
The Post notes that the denomination, which claims 12.5 million
members, was in the early 20th century the "largest Protestant
denomination in the U.S.," but that it has been shrinking in
recent decades. The new split will be the second in the church's
history. The first occurred in 1968, when roughly 10 percent of
the denomination left to form the Evangelical United Brethren
Church. The Post notes that the proposed split "comes at a
critical time for the church, which has been losing members
for years," which has been "pushed toward the brink of a schism
over the role of LGBTQ people in the church." Gay marriage is
not the only issue that has divided the church. In 2016, the
denomination was split over ordination of transgender clergy,
with the North Pacific regional conference voting to ban them
from serving as clergy, and the South Pacific regional conference
voting to allow them.

Figure 7.4: The GPT-3 generated news article that humans had the greatest difficulty distinguishing
from a human written article (accuracy: 12%).

Aditya Ramesh experimented with loss scaling strategies for pretraining.

Melanie Subbiah and Arvind Neelakantan implemented, experimented with, and tested beam
search.
Pranav Shyam worked on SuperGLUE and assisted with connections to few-shot learning and
meta-learning literature.
Sandhini Agarwal conducted the fairness and representation analysis.
Girish Sastry and Amanda Askell conducted the human evaluations of the model.
Ariel Herbert-Voss conducted the threat analysis of malicious use.
Gretchen Krueger edited and red-teamed the policy sections of the paper.
Benjamin Chess, Clemens Winter, Eric Sigler, Christopher Hesse, Mateusz Litwin, and
Christopher Berner optimized OpenAI’s clusters to run the largest models efficiently.
Scott Gray developed fast GPU kernels used during training.
Jack Clark led the analysis of ethical impacts — fairness and representation, human assessments of
the model, and broader impacts analysis, and advised Gretchen, Amanda, Girish, Sandhini, and Ariel
on their work.
Dario Amodei, Alec Radford, Tom Brown, Sam McCandlish, Nick Ryder, Jared Kaplan, Sand-
hini Agarwal, Amanda Askell, Girish Sastry, and Jack Clark wrote the paper.

18
 Title: Star’s Tux Promise Draws Megyn Kelly’s Sarcasm
Subtitle: Joaquin Phoenix pledged to not change for each awards
event
Article: A year ago, Joaquin Phoenix made headlines when he
appeared on the red carpet at the Golden Globes wearing a tuxedo
with a paper bag over his head that read, "I am a shape-shifter.
I can't change the world. I can only change myself." It was a
promise to not change to fit into the Hollywood mold: "I think
that's a really special thing, to not change yourself. I think
it's a really special thing to say, `This is what's inside of
me, I'm proud of it, and I'm not going to be ashamed because of
the way that someone else thinks I should be.'" Now, it's the
Oscars, and Phoenix is at it again. But this time, his publicist
is saying he'll be wearing a tux no matter what.
Megyn Kelly was not impressed, and she let him have it on The
Tonight Show. "You know, I feel like, I feel like you could
have worn the tux," she says. "But you're saying you're a
shape-shifter. I don't know if you can change your tux, but you
can change your mind. You can change your mind. You can change
your mind." Phoenix says he did, but it didn't stick. "I was
like, `Okay, I'm going to wear a tuxedo to this thing.' And then
I thought, `I don't want to wear a tuxedo to this thing.'" Kelly
goes on to encourage him to change his mind again, but Phoenix
says it's too late: "I'm committed to wearing this."

Figure 7.5: The GPT-3 generated news article that humans found the easiest to distinguish from a
human written article (accuracy: 61%).

Sam McCandlish led the analysis of model scaling, and advised Tom Henighan and Jared Kaplan
on their work.
Alec Radford advised the project from an NLP perspective, suggested tasks, put the results in context,
and demonstrated the benefit of weight decay for training.
Ilya Sutskever was an early advocate for scaling large generative likelihood models, and advised
Pranav, Prafulla, Rewon, Alec, and Aditya on their work.
Dario Amodei designed and led the research.

19
References
[ADG+ 16] Marcin Andrychowicz, Misha Denil, Sergio Gomez, Matthew W Hoffman, David
Pfau, Tom Schaul, Brendan Shillingford, and Nando De Freitas. Learning to learn by
gradient descent by gradient descent. In Advances in neural information processing
systems, pages 3981–3989, 2016.
[AJF19] Roee Aharoni, Melvin Johnson, and Orhan Firat. Massively multilingual neural
machine translation. In Proceedings of the 2019 Conference of the North Ameri-
can Chapter of the Association for Computational Linguistics: Human Language
Technologies, Volume 1 (Long and Short Papers), 2019.
[BBDIW20] Su Lin Blodgett, Solon Barocas, Hal Daumé III, and Hanna Wallach. Language (tech-
nology) is power: A critical survey of “bias” in nlp. arXiv preprint arXiv:2005.14050,
2020.
[BES10] Stefano Baccianella, Andrea Esuli, and Fabrizio Sebastiani. Sentiwordnet 3.0: an en-
hanced lexical resource for sentiment analysis and opinion mining. In Lrec, volume 10,
pages 2200–2204, 2010.
[BHT+ 20] Yonatan Bisk, Ari Holtzman, Jesse Thomason, Jacob Andreas, Yoshua Bengio, Joyce
Chai, Mirella Lapata, Angeliki Lazaridou, Jonathan May, Aleksandr Nisnevich, et al.
Experience grounds language. arXiv preprint arXiv:2004.10151, 2020.
[BLC13] Yoshua Bengio, Nicholas Léonard, and Aaron C. Courville. Estimating or propagating
gradients through stochastic neurons for conditional computation. Arxiv, 2013.
[Car97] Rich Caruana. Multitask learning. Machine learning, 28(1), 1997.
[CCE+ 18] Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa
Schoenick, and Oyvind Tafjord. Think you have solved question answering? try arc,
the ai2 reasoning challenge. ArXiv, abs/1803.05457, 2018.
[CGRS19] Rewon Child, Scott Gray, Alec Radford, and Ilya Sutskever. Generating long sequences
with sparse transformers, 2019.
[CLY+ 19] Yen-Chun Chen, Linjie Li, Licheng Yu, Ahmed El Kholy, Faisal Ahmed, Zhe Gan,
Yu Cheng, and Jingjing Liu. Uniter: Learning universal image-text representations.
arXiv preprint arXiv:1909.11740, 2019.
[Cra17] Kate Crawford. The trouble with bias. NIPS 2017 Keynote, 2017.
[DCLT18] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-
training of deep bidirectional transformers for language understanding. arXiv preprint
arXiv:1810.04805, 2018.
[DGV+ 18] Mostafa Dehghani, Stephan Gouws, Oriol Vinyals, Jakob Uszkoreit, and Lukasz
Kaiser. Universal transformers. Arxiv, 2018.
[DHKH14] Nadir Durrani, Barry Haddow, Philipp Koehn, and Kenneth Heafield. Edinburgh’s
phrase-based machine translation systems for wmt-14. In Proceedings of the Ninth
Workshop on Statistical Machine Translation, pages 97–104, 2014.
[DL15] Andrew M. Dai and Quoc V. Le. Semi-supervised sequence learning. In Advances in
neural information processing systems, 2015.
[DSC+ 16] Yan Duan, John Schulman, Xi Chen, Peter L. Bartlett, Ilya Sutskever, and Pieter
Abbeel. Rl2 : Fast reinforcement learning via slow reinforcement learning. ArXiv,
abs/1611.02779, 2016.
[DWD+ 19] Dheeru Dua, Yizhong Wang, Pradeep Dasigi, Gabriel Stanovsky, Sameer Singh, and
Matt Gardner. Drop: A reading comprehension benchmark requiring discrete reasoning
over paragraphs. arXiv preprint arXiv:1903.00161, 2019.

20
[DYY+ 19] Zihang Dai, Zhilin Yang, Yiming Yang, Jaime G. Carbonell, Quoc V. Le, and Ruslan
Salakhutdinov. Transformer-xl: Attentive language models beyond a fixed-length
context. Arxiv, 2019.
[EOAG18] Sergey Edunov, Myle Ott, Michael Auli, and David Grangier. Understanding back-
translation at scale. arXiv preprint arXiv:1808.09381, 2018.
[FAL17] Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for
fast adaptation of deep networks. ArXiv, abs/1703.03400, 2017.
[GG19] Hila Gonen and Yoav Goldberg. Lipstick on a pig: Debiasing methods cover up
systematic gender biases in word embeddings but do not remove them. arXiv preprint
arXiv:1903.03862, 2019.
[Gra16] Alex Graves. Adaptive computation time for recurrent neural networks. Arxiv, 2016.
[GSL+ 18] Suchin Gururangan, Swabha Swayamdipta, Omer Levy, Roy Schwartz, Samuel R
Bowman, and Noah A Smith. Annotation artifacts in natural language inference data.
arXiv preprint arXiv:1803.02324, 2018.
[GSR19] Sebastian Gehrmann, Hendrik Strobelt, and Alexander M. Rush. Gltr: Statistical
detection and visualization of generated text. arXiv preprint arXiv: 1906.04043, 2019.
[GWC+ 18] Jiatao Gu, Yong Wang, Yun Chen, Kyunghyun Cho, and Victor OK Li. Meta-learning
for low-resource neural machine translation. arXiv preprint arXiv:1808.08437, 2018.
[HB20] Daniel Hernandez and Tom Brown. Ai and efficiency, May 2020.
[HNA+ 17] Joel Hestness, Sharan Narang, Newsha Ardalani, Gregory Diamos, Heewoo Jun,
Hassan Kianinejad, Md. Mostofa Ali Patwary, Yang Yang, and Yanqi Zhou. Deep
learning scaling is predictable, empirically. arXiv preprint arXiv:1712.00409, 2017.
[HVD15] Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the knowledge in a neural
network. arXiv preprint arXiv:1503.02531, 2015.
[HYC01] Sepp Hochreiter, A Steven Younger, and Peter R Conwell. Learning to Learn Using
Gradient Descent. In International Conference on Artificial Neural Networks, pages
87–94. Springer, 2001.
[HZJ+ 19] Po-Sen Huang, Huan Zhang, Ray Jiang, Robert Stanforth, Johannes Welbl, Jack Rae,
Vishal Maini, Dani Yogatama, and Pushmeet Kohli. Reducing sentiment bias in
language models via counterfactual evaluation. arXiv preprint arXiv:1911.03064,
2019.
[IDCBE19] Daphne Ippolito, Daniel Duckworth, Chris Callison-Burch, and Douglas Eck. Auto-
matic detection of generated text is easiest when humans are fooled. arXiv preprint
arXiv:1911.00650, 2019.
[JCWZ17] Mandar Joshi, Eunsol Choi, Daniel S. Weld, and Luke Zettlemoyer. TriviaQA: A large
scale distantly supervised challenge dataset for reading comprehension. arXiv preprint
arXiv:1705.03551, 2017.
[JN20] Zheng Junyuan and Gamma Lab NYC. Numeric transformer - albert, March 2020.
[JYS+ 19] Xiaoqi Jiao, Yichun Yin, Lifeng Shang, Xin Jiang, Xiao Chen, Linlin Li, Fang Wang,
and Qun Liu. TinyBERT: Distilling BERT for natural language understanding. arXiv
preprint arXiv:1909.10351, 2019.
[JZC+ 19] Ying Ju, Fubang Zhao, Shijie Chen, Bowen Zheng, Xuefeng Yang, and Yun-
feng Liu. Technical report on conversational question answering. arXiv preprint
arXiv:1909.10772, 2019.
[KKS+ 20] Daniel Khashabi, Tushar Khot, Ashish Sabharwal, Oyvind Tafjord, Peter Clark, and
Hannaneh Hajishirzi. Unifiedqa: Crossing format boundaries with a single qa system.
arXiv preprint arXiv:2005.00700, 2020.

21
 [KMB20] Sarah E. Kreps, Miles McCain, and Miles Brundage. All the news that’s fit to fabricate:
Ai-generated text as a tool of media misinformation, 2020.
[KMH+ 20] Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B. Brown, Benjamin Chess,
Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws
for neural language models, 2020.
[KPR+ 19] Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur
Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Matthew Kelcey, Jacob
Devlin, Kenton Lee, Kristina N. Toutanova, Llion Jones, Ming-Wei Chang, Andrew
Dai, Jakob Uszkoreit, Quoc Le, and Slav Petrov. Natural questions: a benchmark
for question answering research. Transactions of the Association of Computational
Linguistics, 2019.
[KR16] Yoon Kim and Alexander M. Rush. Sequence-level knowledge distillation. Arxiv,
2016.
[LB02] Edward Loper and Steven Bird. Nltk: The natural language toolkit, 2002.
[LC19] Guillaume Lample and Alexis Conneau. Cross-lingual language model pretraining.
arXiv preprint arXiv:1901.07291, 2019.
[LCG+ 19] Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma,
and Radu Soricut. ALBERT: A lite BERT for self-supervised learning of language
representations. arXiv preprint arXiv:1909.11942, 2019.
[LCH+ 20] Xiaodong Liu, Hao Cheng, Pengcheng He, Weizhu Chen, Yu Wang, Hoifung Poon,
and Jianfeng Gao. Adversarial training for large neural language models. arXiv
preprint arXiv:2004.08994, 2020.
[LCR19] Peter J. Liu, Yu-An Chung, and Jie Ren. SummAE: Zero-shot abstractive text sum-
marization using length-agnostic auto-encoders. arXiv preprint arXiv:1910.00998,
2019.
[LDL19] Zhongyang Li, Xiao Ding, and Ting Liu. Story ending prediction by transferable bert.
arXiv preprint arXiv:1905.07504, 2019.
[LGG+ 20] Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad,
Mike Lewis, and Luke Zettlemoyer. Multilingual denoising pre-training for neural
machine translation. arXiv preprint arXiv:2001.08210, 2020.
[LGH+ 15] Xiaodong Liu, Jianfeng Gao, Xiaodong He, Li Deng, Kevin Duh, and Ye-Yi Wang.
Representation learning using multi-task deep neural networks for semantic classifica-
tion and information retrieval. In Proceedings of the 2015 Conference of the North
American Chapter of the Association for Computational Linguistics: Human Language
Technologies, 2015.
[LHCG19a] Xiaodong Liu, Pengcheng He, Weizhu Chen, and Jianfeng Gao. Improving multi-task
deep neural networks via knowledge distillation for natural language understanding.
arXiv preprint arXiv:1904.09482, 2019.
[LHCG19b] Xiaodong Liu, Pengcheng He, Weizhu Chen, and Jianfeng Gao. Multi-task deep neural
networks for natural language understanding. arXiv preprint arXiv:1901.11504, 2019.
[LLG+ 19] Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mo-
hamed, Omer Levy, Ves Stoyanov, and Luke Zettlemoyer. Bart: Denoising sequence-to-
sequence pre-training for natural language generation, translation, and comprehension.
arXiv preprint arXiv:1910.13461, 2019.
[LM17] Ke Li and Jitendra Malik. Learning to optimize neural nets. arXiv preprint
arXiv:1703.00441, 2017.
[LOG+ 19] Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer
Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. RoBERTa: A robustly
optimized BERT pretraining approach. arXiv preprint arXiv:1907.11692, 2019.

22
 [LPP+ 20] Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin,
Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian
Riedel, and Kiela Douwe. Retrieval-augmented generation for knowledge-intensive
nlp tasks. arXiv preprint arXiv:2005.11401, 2020.
[LWS+ 20] Zhuohan Li, Eric Wallace, Sheng Shen, Kevin Lin, Kurt Keutzer, Dan Klein, and
Joseph E. Gonzalez. Train large, then compress: Rethinking model size for efficient
training and inference of transformers, 2020.
[MCH+ 16] Nasrin Mostafazadeh, Nathanael Chambers, Xiaodong He, Devi Parikh, Dhruv Ba-
tra, Lucy Vanderwende, Pushmeet Kohli, and James Allen. A corpus and evalua-
tion framework for deeper understanding of commonsense stories. arXiv preprint
arXiv:1604.01696, 2016.
[MKAT18] Sam McCandlish, Jared Kaplan, Dario Amodei, and OpenAI Dota Team. An empirical
model of large-batch training, 2018.
[MKM+ 94] Mitchell Marcus, Grace Kim, Mary Ann Marcinkiewicz, Robert MacIntyre, Ann Bies,
Mark Ferguson, Karen Katz, and Britta Schasberger. The penn treebank: annotating
predicate argument structure. In Proceedings of the workshop on Human Language
Technology, pages 114–119. Association for Computational Linguistics, 1994.
[MKXS18] Bryan McCann, Nitish Shirish Keskar, Caiming Xiong, and Richard Socher. The
natural language decathlon: Multitask learning as question answering. arXiv preprint
arXiv:1806.08730, 2018.
[MPL19] R Thomas McCoy, Ellie Pavlick, and Tal Linzen. Right for the wrong reasons: Diagnos-
ing syntactic heuristics in natural language inference. arXiv preprint arXiv:1902.01007,
2019.
[MWZ+ 18] Margaret Mitchell, Simone Wu, Andrew Zaldivar, Parker Barnes, Lucy Vasserman,
Ben Hutchinson, Elena Spitzer, Inioluwa Deborah Raji, and Timnit Gebru. Model
cards for model reporting, 2018.
[NBR20] Moin Nadeem, Anna Bethke, and Siva Reddy. Stereoset: Measuring stereotypical bias
in pretrained language models. arXiv preprint arXiv:2004.09456, 2020.
[NK19] Timothy Niven and Hung-Yu Kao. Probing neural network comprehension of natural
language arguments. arXiv preprint arXiv:1907.07355, 2019.
[NvNvdG19] Malvina Nissim, Rik van Noord, and Rob van der Goot. Fair is better than sensational:
Man is to doctor as woman is to doctor. arXiv preprint arXiv:1905.09866, 2019.
[oR16] University of Regensburg. Fascha, 2016.
[PFB18] Jason Phang, Thibault Févry, and Samuel R. Bowman. Sentence encoders on
STILTs: Supplementary training on intermediate labeled-data tasks. arXiv preprint
arXiv:1811.01088, 2018.
[PKL+ 16] Denis Paperno, Germán Kruszewski, Angeliki Lazaridou, Quan Ngoc Pham, Raffaella
Bernardi, Sandro Pezzelle, Marco Baroni, Gemma Boleda, and Raquel Fernández. The
lambada dataset: Word prediction requiring a broad discourse context. arXiv preprint
arXiv:1606.06031, 2016.
[Pos18] Matt Post. A call for clarity in reporting BLEU scores. arXiv preprint
arXiv:1804.08771, 2018.
[QMZH19] Yusu Qian, Urwa Muaz, Ben Zhang, and Jae Won Hyun. Reducing gender bias in
word-level language models with a gender-equalizing loss function. arXiv preprint
arXiv:1905.12801, 2019.
[RCM19] Siva Reddy, Danqi Chen, and Christopher D Manning. Coqa: A conversational
question answering challenge. Transactions of the Association for Computational
Linguistics, 7:249–266, 2019.

23
 [RCP+ 17] Scott Reed, Yutian Chen, Thomas Paine, Aäron van den Oord, SM Eslami, Danilo
Rezende, Oriol Vinyals, and Nando de Freitas. Few-shot autoregressive density
estimation: Towards learning to learn distributions. arXiv preprint arXiv:1710.10304,
2017.
[RL16] Sachin Ravi and Hugo Larochelle. Optimization as a model for few-shot learning.
ICLR 2017 (oral), 2016.
[RLL+ 19] Qiu Ran, Yankai Lin, Peng Li, Jie Zhou, and Zhiyuan Liu. NumNet: Machine reading
comprehension with numerical reasoning. In Proceedings of EMNLP, 2019.
[RNLVD18] Rachel Rudinger, Jason Naradowsky, Brian Leonard, and Benjamin Van Durme.
Gender bias in coreference resolution. arXiv preprint arXiv:1804.09301, 2018.
[Ros12] R.S. Ross. Guide for conducting risk assessments. NIST Special Publication, 2012.
[RRBS19] Jonathan S. Rosenfeld, Amir Rosenfeld, Yonatan Belinkov, and Nir Shavit. A con-
structive prediction of the generalization error across scales, 2019.
[RRS20] Adam Roberts, Colin Raffel, and Noam Shazeer. How much knowledge can you pack
into the parameters of a language model? arXiv preprint arXiv:2002.08910, 2020.
[RSR+ 19] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael
Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learning
with a unified text-to-text transformer, 2019.
[RWC+ 19] Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya
Sutskever. Language models are unsupervised multitask learners, 2019.
[SBC+ 19] Irene Solaiman, Miles Brundage, Jack Clark, Amanda Askell, Ariel Herbert-Voss, Jeff
Wu, Alec Radford, Gretchen Krueger, Jong Wook Kim, Sarah Kreps, Miles McCain,
Alex Newhouse, Jason Blazakis, Kris McGuffie, and Jasmine Wang. Release strategies
and the social impacts of language models, 2019.
[SCNP19] Emily Sheng, Kai-Wei Chang, Premkumar Natarajan, and Nanyun Peng. The
woman worked as a babysitter: On biases in language generation. arXiv preprint
arXiv:1909.01326, 2019.
[SDCW19] Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. DistilBERT,
a distilled version of BERT: smaller, faster, cheaper and lighter. arXiv preprint
arXiv:1910.01108, 2019.
[SDSE19] Roy Schwartz, Jesse Dodge, Noah A. Smith, and Oren Etzioni. Green AI. CoRR,
abs/1907.10597, 2019.
[SHB15] Rico Sennrich, Barry Haddow, and Alexandra Birch. Improving neural machine
translation models with monolingual data. arXiv preprint arXiv:1511.06709, 2015.
[SMM+ 17] Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey
Hinton, and Jeff Dean. Outrageously large neural networks: The sparsely-gated
mixture-of-experts layer. arXiv preprint arXiv:1701.06538, 2017.
[SPP+ 19] Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper,
and Bryan Catanzaro. Megatron-lm: Training multi-billion parameter language models
using model parallelism, 2019.
[SS20] Timo Schick and Hinrich Schütze. Exploiting cloze questions for few-shot text
classification and natural language inference. arXiv preprint arXiv:2001.07676, 2020.
[STQ+ 19] Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, and Tie-Yan Liu. MASS: Masked sequence
to sequence pre-training for language generation. arXiv preprint arXiv:1905.02450,
2019.
[Tur20] Project Turing. Microsoft research blog, Feb 2020.

24
[VBL+ 16] Oriol Vinyals, Charles Blundell, Timothy Lillicrap, Daan Wierstra, et al. Matching
Networks for One Shot Learning. In Advances in neural information processing
systems, pages 3630–3638, 2016.
[VSP+ 17] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N.
Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in
neural information processing systems, 2017.
[WPN+ 19] Alex Wang, Yada Pruksachatkun, Nikita Nangia, Amanpreet Singh, Julian Michael,
Felix Hill, Omer Levy, and Samuel Bowman. Superglue: A stickier benchmark for
general-purpose language understanding systems. In Advances in Neural Information
Processing Systems, pages 3261–3275, 2019.
[WXH+ 18] Yiren Wang, Yingce Xia, Tianyu He, Fei Tian, Tao Qin, ChengXiang Zhai, and Tie-Yan
Liu. Multi-agent dual learning. ICLR 2019, 2018.
[XDH+ 19] Qizhe Xie, Zihang Dai, Eduard Hovy, Minh-Thang Luong, and Quoc V. Le. Unsuper-
vised data augmentation for consistency training, 2019.
[YDY+ 19] Zhilin Yang, Zihang Dai, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, and
Quoc V. Le. XLNet: Generalized autoregressive pretraining for language understand-
ing. arXiv preprint arXiv:1906.08237, 2019.
[ZHB+ 19] Rowan Zellers, Ari Holtzman, Yonatan Bisk, Ali Farhadi, and Yejin Choi. Hellaswag:
Can a machine really finish your sentence? arXiv preprint arXiv:1905.07830, 2019.
[ZHR+ 19] Rowan Zellers, Ari Holtzman, Hannah Rashkin, Yonatan Bisk, Ali Farhadi, Franziska
Roesner, and Yejin Choi. Defending against neural fake news. arXiv preprint
arXiv:1905.12616, 2019.
[ZSW+ 19] Daniel M. Ziegler, Nisan Stiennon, Jeffrey Wu, Tom B. Brown, Alec Radford, Dario
Amodei, Paul Christiano, and Geoffrey Irving. Fine-tuning language models from
human preferences, 2019.

25