ABAB Pedia

Jeremy Howard

Jeremy Howard is a data scientist and entrepreneur who co-founded fast.ai with Rachel Thomas to make deep learning more accessible, proposed the language-model transfer-learning method ULMFiT, and founded the AI R&D lab Answer.AI.

Contents26 sections
Key facts

Works in Management Consulting

Co-founds FastMail with Rob Mueller

In 1999 he and his school friend Rob Mueller created the Melbourne-based e-mail service FastMail.FM.⁠[2]

Rob Mueller recalled that Howard built the first prototype because e-mail options for people on the road were limited, and funded early development from his savings; at first the two were the only developers and also handled support and system administration, positioning FastMail as a premium e-mail provider.⁠[3]

Optimal Decisions Group Acquired by ChoicePoint

The Australian company Optimal Decisions Group developed analytics software for insurers, using predictive modelling, multi-year simulation and optimisation to price auto and home insurance. On March 31, 2008, ChoicePoint announced it was acquiring the company; terms were not disclosed.⁠[4]

He was the company’s founding CEO.⁠[5]

FastMail Sold to Opera

In May 2010 the Norwegian browser company Opera announced its acquisition of FastMail.FM; terms were not disclosed, and some staff were to move to Norway.⁠[2]

Invests in Kaggle and Becomes Its President

Leads the New Company Enlitic

In August 2014 he was co-founder and CEO of the new company Enlitic, which trained computers on large sets of anonymised medical images showing conditions such as brain tumours so they could flag them for physicians.⁠[8]

TEDxBrussels Talk

In December 2014 he gave the TEDxBrussels talk The wonderful and terrifying implications of computers that can learn, on advances in deep learning; at the time of writing it had about 2.78 million plays on the TED website.⁠[9]

Launches fast.ai with Rachel Thomas

On October 7, 2016, he and the mathematician Rachel Thomas announced the research lab fast.ai, aiming to let domain experts use deep learning themselves; its first course was offered through the University of San Francisco’s Data Institute.⁠[10]

Publishes the ULMFiT Paper

On January 18, 2018, he and Sebastian Ruder posted a paper on arXiv proposing ULMFiT (Universal Language Model Fine-tuning), a transfer-learning method applicable to any NLP task; the paper found that it outperformed the state of the art on six text classification tasks, reducing error by 18–24% on most datasets.⁠[11]

The paper was accepted at the 56th Annual Meeting of the Association for Computational Linguistics (ACL 2018), held in Melbourne on July 15–20, 2018.⁠[12]

fast.ai Team Tops DAWNBench

In the DAWNBench competition run by Stanford researchers, the top three places for fastest and cheapest training on the CIFAR-10 image task all went to researchers affiliated with fast.ai; he attributed his students’ win to creative thinking, using the super convergence technique devised by Leslie Smith.⁠[13]

Trains an ImageNet Model in 18 Minutes

In August 2018 a fast.ai team trained on ImageNet in 18 minutes using 16 Amazon Web Services instances at a compute cost of about $40; he claimed this was about 40 percent better than GoogleGoogleGoogle is an American technology company founded by Larry Page and Sergey Brin and incorporated in 1998. It became an Alphabet subsidiary in 2015. Its businesses include internet search, advertising, software platforms, cloud computing, and artificial intelligence.Open the full entry’s effort, while acknowledging that comparison was tricky because the hardware differed.⁠[14]

USF Launches WAMRI

In November 2018 the USF Data Institute launched the Wicklow AI and Medical Research Initiative (WAMRI), funded by a gift from Wicklow Capital, to develop practical AI applications in oncology, cardiology and neurology.⁠[15]

He was a Distinguished Research Scientist at the University of San Francisco and the founding chair of WAMRI.⁠[5]

Appears on the Lex Fridman Podcast

Publishes the fastai Paper

In February 2020 he and Sylvain Gugger published the paper Fastai: A Layered API for Deep Learning in the journal Information, volume 11, issue 2.⁠[17]

The paper presented fastai version 2, offering ready-made components for practitioners and modular building blocks for researchers; the authors then expected an official release around July 2020.⁠[18]

Launches #Masks4All

Publishes Deep Learning for Coders with fastai and PyTorch

In July 2020 O’Reilly published Deep Learning for Coders with fastai and PyTorch, co-written with Sylvain Gugger, a 621-page book for programmers comfortable with Python but with little math background.⁠[20]

Mask Evidence Review Published in PNAS

In January 2021 he was first author of a review of the evidence on face masks in Proceedings of the National Academy of Sciences, volume 118, issue 4; it concluded that mask wearing reduces transmissibility per contact and recommended that governments encourage widespread public mask use.⁠[21]

Named Honorary Professor at the University of Queensland

In March 2022 the University of Queensland’s School of Electrical Engineering and Computer Science appointed him an Honorary Professor, with him teaching fast.ai’s seven-week Practical Deep Learning for Coders course there from April; he was then also chief scientist at platform.ai.⁠[22]

Launches Answer.AI

On December 12, 2023, he announced the AI R&D lab Answer.AI, serving as founding CEO, with Lean Startup creator Eric Ries as founding director and a $10 million investment from Decibel VC; he said its mission was the same as fast.ai’s and that the team would be fully remote.⁠[23]

Answer.AI Open-Source Projects and Policy Writing

On March 6, 2024, Answer.AI released its first project, an open-source system combining FSDP and QLoRA that can train a 70-billion-parameter language model on two 24GB consumer gaming GPUs, with collaborators including Tim Dettmers and Hugging Face’s Titus von Koeller.⁠[24]

That year he wrote that California’s SB 1047 bill would reduce safety by restricting open-source AI (April 29), and released the web framework FastHTML (August 3) and the /llms.txt proposal for websites to provide information to large language models (September 3).⁠[25]

fast.ai Joins Answer.AI

On November 7, 2024, he announced that fast.ai was joining Answer.AI and launched the course How To Solve It With Code on the in-house solveit platform, promoting Dialog Engineering, in which humans and AI work in small steps; he later added that the first cohort sold out within 24 hours and that a new course would start on November 3, 2025.⁠[26]

Keynote Speaker at PyTorch Conference

He was listed as a keynote speaker at the Linux Foundation’s PyTorch Conference 2025.⁠[27]

Keynote at AI Engineer Melbourne

He gave the keynote at AI Engineer Melbourne, held at Federation Square in Melbourne.⁠[28]

As of October 2026: Research and Academic Roles

He is an Honorary Professor at the University of Queensland and a Digital Fellow at Stanford UniversityStanford UniversityStanford University is a research university in California founded by Leland and Jane Stanford in 1885 and opened in 1891.Open the full entry, and created the fastai software library; fast.ai states that Enlitic was named one of MIT Technology Review’s 50 smartest companies two years running and that Optimal Decisions Group was later purchased by LexisNexis.⁠[1]

Related Organizations, Websites, and Public Accounts

Sources

  1. About – fast.ai (opens in a new window)
  2. Opera buys Aussie e-mail service FastMail.FM (opens in a new window)
  3. FastMail sale to Opera: the story from one of the owners (opens in a new window)
  4. ChoicePoint Acquires Optimal Decisions Group (opens in a new window)
  5. Jeremy Howard - Stanford Digital Economy Lab (opens in a new window)
  6. Kaggle president on how they did it, and tips on how to tell your start-up’s story to raise millions from investors (opens in a new window)
  7. Aussie data mining firm snaps PayPal co-founder as chairman after $11m funding round (opens in a new window)
  8. A Startup Hopes to Teach Computers to Spot Tumors in Medical Scans (opens in a new window)
  9. The wonderful and terrifying implications of computers that can learn (opens in a new window)
  10. Launching fast.ai (opens in a new window)
  11. Universal Language Model Fine-tuning for Text Classification (opens in a new window)
  12. ACL 2018: Universal Language Model Fine-tuning for Text Classification (opens in a new window)
  13. An AI speed test shows clever coders can still beat tech giants like Google and Intel (opens in a new window)
  14. A small team of student AI coders beats Google’s machine-learning code (opens in a new window)
  15. University of San Francisco unveils AI initiative (opens in a new window)
  16. Jeremy Howard: fast.ai Deep Learning Courses and Research | Lex Fridman Podcast (opens in a new window)
  17. Fastai: A Layered API for Deep Learning (opens in a new window)
  18. fastai A Layered API for Deep Learning (opens in a new window)
  19. USF Data Scientist Launches #Masks4All (opens in a new window)
  20. Deep Learning for Coders with fastai and PyTorch (opens in a new window)
  21. An evidence review of face masks against COVID-19 (opens in a new window)
  22. Renowned AI expert joins UQ to help develop local talent (opens in a new window)
  23. A new old kind of R&D lab – Answer.AI (opens in a new window)
  24. You can now train a 70b language model at home – Answer.AI (opens in a new window)
  25. Answer.AI - Practical AI R&D (opens in a new window)
  26. A New Chapter for fast.ai: How To Solve It With Code – fast.ai (opens in a new window)
  27. Keynote Speakers (opens in a new window)
  28. Jeremy Howard (opens in a new window)