Jared Kaplan

Jared Kaplan works on large neural language models, with a recurring interest in how their behaviour changes as they grow. Kaplan's papers here train families of Transformer models at many sizes and fit the results to simple quantitative trends, typically tracking test loss against parameters, data and compute. The same scaling lens is applied to text, code, images, video and mathematics, and also to how well models perform on new tasks without retraining.

The most cited of these papers is Language Models are Few-Shot Learners, which introduced GPT-3 and showed that a very large autoregressive model can take on many language tasks from instructions and a handful of examples placed in its prompt, an ability the authors called in-context learning. Scaling Laws for Neural Language Models found that loss falls along smooth power laws in model size, dataset size and training compute, as long as none of the three becomes a bottleneck. A companion study, Scaling Laws for Autoregressive Generative Modeling, examined whether those trends carry over to other kinds of data and reported that they do across every domain tested.

The newest work here turns toward programs and transfer. Evaluating Large Language Models Trained on Code presented Codex, a GPT model fine-tuned on public GitHub code. It also released HumanEval, a benchmark that judges generated Python functions by whether they pass unit tests rather than by how closely the text resembles a reference. Scaling Laws for Transfer measured the value of pre-training for tasks with scarce data and expressed that value as an equivalent amount of extra training data.

Kaplan is currently at Anthropic in San Francisco, and earlier bylines list Johns Hopkins University. Frequent co-authors on the papers here include Tom B. Brown, Tom Henighan and Sam McCandlish. The work sits mainly within language models and pretraining, with a further contribution to evaluation and benchmarks.

Jared Kaplan has 5 papers in our library, in 2020–2021, most of them on Language models & pretraining, Evaluation & benchmarks; often with Tom B. Brown, Tom Henighan, Benjamin Mann.

5 papers in our library · 2020–2021

current affiliationAnthropic (United States), San Francisco

OpenAlex · h-index 40 · 82 works · 11 886 citations OpenAlex, 2026-10-14

5 works in our library · 19 286 citations in our library · 7 more works in our library · ⭐ Paper Star rank #108

The profile is the library's match of the name; two people with one name are not always told apart.

Elsewhere: Google Scholar · Semantic Scholar · dblp · ACL Anthology

Affiliations

Anthropic (United States) · San Francisco — current · 2021–2025 on 5 papers
Johns Hopkins University · Baltimore, United States — 2013–2025 on 38 papers
University of Oxford · Oxford, United Kingdom — 2023–2024 on 3 papers
Harvard University · Cambridge, United States — 2006–2022 on 9 papers
Stanford University · Stanford, United States — 2010–2018 on 7 papers
University of California, Davis · Davis, United States — 2018 on 3 papers
SLAC National Accelerator Laboratory · Menlo Park, United States — 2010–2016 on 15 papers
Institute for Advanced Study · Princeton, United States — 2008–2010 on 2 papers

From the bylines of their papers (OpenAlex, 2026-10-14), ORCID and Wikidata; institutions on papers are matched by machine. 7 more institutions on single bylines are not listed.

Topics

Language models & pretraining4 Evaluation & benchmarks1

Works by year

2020s

20212 papers 20203 papers

Papers

The most cited first (OpenAlex and our library).

Language Models are Few-Shot Learners (2020 · 10 782 citations)
The standard approach in NLP has a significant drawback: after pre-training, a model is usually fine-tuned on thousands of labeled examples for each individual task.
Evaluating Large Language Models Trained on Code (2021 · 2 529 citations)
Code generation models were usually scored by how closely their output text matched a reference, for example with BLEU.
Scaling Laws for Neural Language Models (2020 · 1 814 citations)
The authors from OpenAI study how the quality of a language model depends on its scale. They measure quality by the cross-entropy loss on test data. To this end, they trained a large series of Transformer models on the WebText2 corpus, ranging from thousands to roughly a billion parameters, excluding embeddings. In…
Scaling Laws for Autoregressive Generative Modeling (2020 · 115 citations)
The authors from OpenAI examine whether the scaling laws previously found for language models also hold for other data types.
Scaling Laws for Transfer (2021 · 102 citations)
This paper asks how much language-model pre-training is worth when the target task has little data. It answers in units of data. The authors trained transformers across roughly four orders of magnitude in both non-embedding parameters and dataset size, using three curricula:

Also in our library

7 more works of theirs are in our library without a page here yet; they will appear as we publish them.

Co-authors here

12 of 27 co-authors with a page, the most papers together first.

Cited by

Papers of our library cite their works 19 286 times (of the 140 884 papers whose bibliographies we read); 1 035 citing papers with a page here →.

This page in Markdown → https://md.research.elamur.ai/free/en/authors/jared-kaplan/

🧑‍🔬 Jared Kaplan