2023 Vertical Snapshot Generative AI Preview
2023 Vertical Snapshot Generative AI Preview
Vertical Snapshot:
Generative AI
VC trends, industry overview, and market landscape
REPORT PREVIEW
2023 The full report is available through
the PitchBook Platform.
Introduction 4 Analysis
Taxonomy 18 Publishing
Market map 23 Report designed by Drew Sanders
Key players 29
Key partnerships 33
Glossary 37
December 1997
Two computer scientists, Sepp December 2013
Hochreiter and Jürgen Schmidhuber, Diederik P. Kingma and Max Welling June 2016
invent Long Short-Term Memory (LSTM) publish a paper on VAEs.6 VAEs Style Transfer, a deep
networks, which improves memory deconstruct input data into code, July 2015 neural network that
capacity in neural networks and thus known as the latent space, and then Google’s DeepDream can transfer the style of
allows for faster and more precise construct new, representative data June 2014 model for image one image to another,
pattern recognition in training data.6 out of the latent space. Ian Goodfellow invents GANs.6 generation is launched. 7 is released.7
1997
2012
2013
2014
2015
2016
2017
2012 January 2014 March 2015 December 2015 June 2017
A seminal year in deep learning research that enables significant Google acquires DeepMind, an AI Researchers Jascha Sohl- OpenAI is founded as a Google researchers release
advances in computer vision, speech recognition, and natural research lab based in the UK. Dickstein, Eric A. Weiss, Niru non-profit, open-source “Attention Is All You Need” paper
language processing.5 These accomplishments are best signified Maheswaranathan, and Surya research lab. on transformer models, a type of
by AlexNet, a convolutional neural network (CNN) trained on Ganguli introduce diffusion neural network architecture that
graphical processing units (GPUs) and developed by researcher models.6 These models apply lends attention to context and
Alex Krizhevksy. AlexNet was the first model to break 75% noise to an image until it is syntax.8 This allows for immense
accuracy, identifying images from a 14-million-image manually incomprehensible, then gradually strides in natural language
tagged database known as ImageNet. reduce the noise in the image and processing and understanding.
thus learn how to generate actual
images from noise.
5: “2012: A Breakthrough Year for Deep Learning,” Medium, Bryan House, July 17, 2019.
6: “The Generative AI Revolution Has Begun—How Did We Get Here?,” Ars Technica, Haomiao Huang, January 30, 2023.
7: “Timeline of Text-To-Image Machine Learning Models,” Fabian Mosele, n.d., accessed March 9, 2023.
8: “Timeline of AI and Language Models,” Life Architect, Alan D. Thompson, n.d., accessed March 9, 2023.
March 2019
June 2018 OpenAI announces July 2019
OpenAI releases GPT that it has moved Microsoft invests November 2020
(Generative Pre- from a nonprofit $1 billion into May 2020 AlphaFold wins the CASP
Trained Transformer), December 2018 to “capped-profit” OpenAI and signs GPT-3 is released competition by a large margin
ushering in a new NVIDIA’s StyleGAN business model in an exclusive with 175 billion and is recognized by organizers
era of large language is able to generate February 2019 order to attract more partnership with the parameters, 100 as a solution to the “protein-
models. hyper-realistic faces. GPT-2 is released. investors.9 research startup. times that of GPT-2. folding problem.”10
2018
2019
2020
November 2018 December 2018 March 2019 July 2019 October 2019 September 2020
Google releases its Google and DeepMind’s Chinese company Baidu Baidu releases Google applies GPT-3 writes a newspaper
Bidirectional Encoder AlphaFold, a protein releases its answer to ERNIE 2.0, which its BERT model column in The Guardian.
Representations for structure prediction Google’s BERT: Enhanced outperforms to searches—
Transformers (BERT). model, places first in the Representation Through Google’s BERT. demonstrating
Critical Assessment of Knowledge Integration semantic search
Protein Structure (CASP) (ERNIE 1.0). capabilities.
competition.
9: “OpenAI Shifts From Nonprofit to ‘Capped-Profit’ to Attract Capital,” TechCrunch, Devin Coldewey, March 11, 2019.
10: “DeepMind’s Protein-Folding AI Has Solved a 50-year-old Grand Challenge of Biology,” MIT Technology Review, Will Douglas Heaven, November 30, 2020.
2022
May 2021 November 2021 March 2022 April 2022 May 2022 July 2022 September 2022
Google announces Language Microsoft announces MidJourney, a text-to- Google releases its Meta releases its Open Research startup Meta announces
Model for Dialogue Applications NÜWA, a multimodal image model offered 540 billion parameter Pretrained Transformer Stability.ai launches Make-A-Video, its
(LaMDA), its transformer-based model for text-to-image through a Discord server, Pathways Language (OPT) model, a totally Stable Diffusion, an text-to-video model.
large language model. and text-to-video. moves into open beta. Model (PaLM), a open-source LLM, to the open-source text-to-
multifunctional public. Details about image model.
language model. the model’s training are
also released. Access is
available upon request,
but the license limits
usage of the model to
research purposes.
February 2023
Microsoft releases a limited beta
upgrade to its Bing search system,
January 2023 with new capabilities such as
October 2022 November 2022 Microsoft announces a “multiyear, summarization and chat based on February 2023
Google launches Stability.ai and Amazon multibillion-dollar” investment in February 2023 ChatGPT. GPT is also integrated into Google announces BARD,
text-to-video model Web Services announce OpenAI and exclusive integration AWS and Hugging Face other Microsoft products including its answer to ChatGPT,
Imagen Video. a partnership. with its Azure cloud system.11 announce partnership. Teams and its Edge browser. for limited beta testing.
2023
November 2022 February 2023 February 2023 March 2023 March 2023
OpenAI announces ChatGPT, a Runway, an AI research lab, Google invests $300 million into Anthropic, a research Google announces OpenAI releases
powerful chatbot tool running announces its Gen-1 model that startup made up of former OpenAI employees. generative tools GPT-4.
on the GPT-3.5 model. It soon allows for style transfer across videos. Anthropic calls Google its “preferred cloud provider.”12 to be integrated
becomes the most quickly into workplace.
adopted tech product in history.
11: “Microsoft Announces New Multibillion-Dollar Investment in ChatGPT-Maker OpenAI,” CNBC, Ashley Capoot, January 23, 2023.
12: “Google Invested $300 Million in AI Firm Founded by Former OpenAI Researchers,” The Verge, James Vincent, February 3, 2023.