Venture capitalists are rushing to invest in artificial intelligence start-ups as growing hype around “generative AI” fills the void left by failing cryptocurrency and blockchain ventures.
The recent leap in developments of sophisticated computer programs that can write scripts and create art in seconds has driven a surge of investor interest, creating a rare bright spot in a start-up landscape dominated by tumbling valuations and job cuts.
OpenAI, a San Francisco-based company in which Microsoft is the largest funder, released the newest form of its GPT-3.5 software to the public last week, which can converse with users through text: answer follow-up questions, admit mistakes and reject inappropriate requests.
In five days, ChatGPT surpassed 1mn users and was praised by billionaire Elon Musk, a co-founder of OpenAI who left the board in 2018, who tweeted: “ChatGPT is scary good. We are not far from dangerously strong AI.”
The freedom to tinker with such a powerful AI has already sparked new start-up ideas for countless investors and entrepreneurs.
“Any organisation can pick these up and start training them and playing with them, see what they can produce,” said Ed Stacey, managing partner at IQ Capital. “It really just doesn’t make sense to sit on the sidelines any more.”
As Silicon Valley investors’ social media feeds fill up with examples of AI-generated imagery and text, pivoting to AI comes as interest in so-called “Web3”, a vision of decentralised virtual worlds built on blockchains, wanes amid the recent cryptocurrency crash.
“There is a huge hype cycle,” said Colin Treseler, co-founder of Supernormal, which uses AI to summarise online meetings. “The Web3 hype ended, and these people needed a place to go.”
Venture capital investment in generative AI has increased 425 per cent since 2020 to $2.1bn this year, according to data from PitchBook, even as the broader technology market declines.
One AI entrepreneur said that after discussing potential fundraising with just three investors, he was inundated with offers from more than 20 others, securing funding after a week of whirlwind meetings around the world.
Two deals in mid-October marked the start of the latest funding frenzy, which saw investments made by Coatue Management and Lightspeed Venture Partners.
Jasper, which describes itself as an “AI copywriter” for marketers, raised $125mn at a $1.5bn valuation, while one of the companies behind image-generation tool Stable Diffusion, London-based Stability AI, raised $101mn, in a move that saw it reach “unicorn” status — a valuation of $1bn. This week, another Stable Diffusion developer, Runway, raised $50mn.
When Cristóbal Valenzuela co-founded Runway four years ago, investors told him, “generative AI is not a thing”. “Everyone thought we were kind of crazy,” he said. Now, investors say this technology could be “as transformative as mobile was 20 years ago”, he added.
“What has changed more recently is that the quality of the models have gotten really good,” said Valenzuela. “It’s not any more about [saying], ‘let’s imagine a future where this might happen’. This is happening now.”
ChatGPT and OpenAI’s art program DALL-E, alongside rival graphics tools Stable Diffusion and Midjourney, are examples of using large language and image models, sometimes called generative AI or foundation models, which can produce content based on previous word sequences or images.
In September, Sequoia Capital partners co-wrote an investment thesis using GPT-3 software, saying that AI would be able to produce final drafts of writing better than the human average, generate code on a commercial scale and make drafts in images and gaming in the next two years.
“Generative AI is well on the way to becoming not just faster and cheaper, but better in some cases than what humans create by hand,” concluded Sequoia’s analysis.
The firm invested in Hugging Face in May in a Series C round that valued the start-up, which has its own large language model Bloom, at $2bn.
The global market for AI-augmented content solutions is set to reach $2.3bn this year and is forecast to grow an overall 17 per cent until 2025, according to research by PitchBook. But the group added that “the technology might not generate high revenues in the short term as professions resist AI solutions and the technology still has to mature”.
Because of the high costs of running the programs and storing vast amounts of data that the AI programs learn from, large language models have previously been the reserve of tech giants like Microsoft, Google and Facebook.
But OpenAI has made its technology available through an application programming interface, offering any company access to its capabilities.
Founded in 2015, OpenAI was created as a non-profit built on the principles of making AI accessible to everybody and developing the technology safely — the brainchild of some of the tech world’s most radical thinkers, including Musk and Peter Thiel.
In 2019, it became a for-profit enterprise, which was shortly followed by a $1bn deal from Microsoft, including the use of its Azure cloud computing platform to conduct experiments. Under the deal, Microsoft gets a first shot at commercialising early results from OpenAI’s research.
Microsoft’s focus on OpenAI came as part of a push to claw back an edge in AI following heavy investment by rival Google in using the technology for search and speech, as well as acquiring London-based AI company DeepMind for around £400mn in 2014.
“One of the historic big advantages corporates had is, is access to large, typically proprietary data sets. They’ve been able to use these data sets to train up large, larger and larger models,” said IQ Capital’s Stacey.
On average, the cost of running ChatGPT is estimated to be a few cents per chat, according to Sam Altman, the chief executive of OpenAI. Asked on Twitter if the tool would be free forever, he replied: “We will have to monetise it somehow at some point; the compute costs are eye-watering.”
we will have to monetize it somehow at some point; the compute costs are eye-watering
— Sam Altman (@sama) December 5, 2022
It has proved so popular in recent days that the platform has limited how many people can use it. The company is in the process of fundraising more capital from investors, according to The Information, and Altman tweeted this week that it was looking to hire more staff.
Bloc Ventures, a UK-based deep tech venture capital firm, has focused its investment on technologies that enable high cloud computing levels, reducing both the costs and energy used in generative AI.
“We are in a world where companies are chasing net zero [carbon emissions], and the luxury of having chatbots we can talk to through AI is burning a hole through the earth in a data centre,” said David Leftley of Bloc Ventures.
New York-based Runway is taking a more ambitious and expensive approach, doing both the primary AI research to build models and turning it into a suite of image generation and collaboration tools already used by companies including Publicis, Google and CBS.
“There’s a lot of companies building on top of existing APIs [but] our bet long-term is you need to own your stack, you need to own your technology, to allow you to more quickly and more easily change if needed,” said Valenzuela.
He insisted AI start-ups like Runway could outmanoeuvre the Big Tech companies: “The field is moving extremely fast. The rate of learning really matters — how you adapt and how you change.”
By making ChatGPT available to the public, OpenAI is able to collect more data to train its large language models and iron out bugs*.
One significant limitation of the technology, and similar AI tools, are so-called “hallucinations”, where the program gives a convincing answer that is factually inaccurate and struggles with simple maths. One example posted to Twitter said ChatGPT wrongly claimed that Angela Merkel and Gerhard Schröder were from the same political party.
“The potential for it spreading misinformation is huge,” said Carissa Véliz, associate professor at the University of Oxford’s institute for ethics and AI. “If you ask it to create a conspiracy theory about Covid, it can make a really convincing case.”
OpenAI admits that GPT “sometimes writes plausible-sounding but incorrect or nonsensical answers”, among other limitations to the technology, leading many to suggest the technology needs human intervention before being embedded in businesses.
“There are a lot of questions around how commercially viable these models and capabilities are,” said Lisa Weaver-Lambert, private equity, data, and AI lead at Microsoft, who said generative AI was in an “experimentation phase”.
She added: “If I were looking to invest in this space, I’d be thinking through, what are the concrete business problems that actually exist that people have been working through [and can AI] come up with a faster, cheaper method for accomplishing the same thing?”
Additional reporting by Madhumita Murgia in London
*This article has been amended since initial publication to correct the description of ChatGPT.
