The Hottest Investment Area of 2022: Generative AI

If I had to choose the hottest investment area of 2022, I would say it has returned to AI. In particular, “AI generation” has sparked even wider discussion this year. The future is here~ ---- SARAH GUO: “From a fundraising perspective, this is the hottest area right now.” Editor's note: The recent rise of AI image-generation tools has ignited venture investors' interest in generative artificial intelligence. This article is translated and adapted. SARAH GUO, founder of VC firm Conviction, organized a gathering in San Francisco that attracted an energetic crowd of engineers, entrepreneurs, and financiers. They all had one thing on their minds: AI's burgeoning creative capabilities. Investors and technologists are excited about the commercial potential of “generative AI.” Guests included AI engineers from major technology companies, fellow investors, and entrepreneurs building businesses on recent advances in text- and image-generating algorithms. One guest was Hugging Face CEO Clement Delangue. His company hosts many open-source generative-AI projects, including one that recently helped ignite the AI frenzy. He answered questions from engineers considering joining the trend and starting their own generative-AI companies. Thanks to advances by Hugging Face and others, social media has recently been flooded with stunning, strange AI-generated images. Related machine-learning techniques allow algorithms to generate large amounts of coherent text on a given subject. Several generative-AI companies have already raised hundreds of millions of dollars, setting investors in pursuit of a new generation of AI unicorns. AI image-tool provider Stability AI announced $101 million in new funding at an astonishing $1 billion valuation. Its gathering attracted technology luminaries including Google cofounder Sergey Brin. Generative-AI enthusiasts predict the technology will take root across industries and do far more than generate images or sentences. Stanford senior David Song has been tracking the excitement and compiled a list of more than 100 generative-AI startups. Their applications include music generation, game development, writing assistants, customer-service bots, coding aids, video editing, and assistants for managing online communities. SARAH GUO invested in a company planning to generate legal contracts from text descriptions—a potentially lucrative application if it works reliably. Song works with Everyprompt, a startup making text generation easier for companies to use. Like many contributors to the field, he says testing generative-AI tools for images, text, or code has made him curious about their possibilities. “I have not used websites and technologies that felt this useful or magical in a long time,” Song says. “Using generative AI feels like using magic.” SARAH GUO sees generative AI as a leap in the technology's potential, similar to a shift that began in 2012 and reshaped the technology industry and its products. Engineers then found that artificial neural networks, a type of machine-learning model, could perform extraordinary new tasks such as recognizing image content or transcribing speech when given enough training data and computing power. In the following years, companies large and small launched an investment wave that spread facial recognition around the world. AI has become an integral part of almost every gadget, application, and service. The race is now on to find generative-AI applications that leave their mark on the world. One early success is Microsoft's Copilot, which writes code for specified tasks for $10 a month. Another is Jasper, which automatically generates text for business blog posts, marketing copy, and emails. Jasper announced $125 million in funding at a $1.5 billion valuation and said it was on track for $75 million in revenue this year. Both Microsoft and Jasper built on OpenAI's services. OpenAI began as a nonprofit funded by Elon Musk and other technology luminaries. Starting in 2019, it pioneered text generation with an algorithm called GPT-2. In late 2021, it made a more powerful successor, GPT-3, available for use. In January 2021, OpenAI also announced DALL-E, a tool that generated rough images from text prompts, greatly increasing interest in AI image generation. DALL-E 2, released in April 2022, rendered more complex images, demonstrating the technology's rapid progress. Many companies, including Stability AI, now offer similar tools. Of course, Silicon Valley hype can run ahead of reality. “There is a lot of FOMO here,” says Air Street Capital investor Nathan Benaich. Adobe's $20 billion acquisition of collaborative design tool Figma, he says, suggests abundant opportunities to reinvent creative tools. He is watching several companies exploring generative AI for protein synthesis. “It is crazy right now. Everyone is talking about it.” Foundation Capital partner Joanne Chen, an early Jasper investor, says turning a generative-AI tool into a valuable company remains difficult. Jasper's founders have devoted most of their effort to fine-tuning the product to meet customers' needs and tastes, but she believes the technology has many possible uses. Chen also says the rise of generative AI means regulators have yet to grapple with some of its potential bad or dangerous uses. She worries about abuse, such as making videos that spread misinformation. “What concerns me most is how people think about safety and fake content,” she says. Other uncertainties raise legal questions. Amir Ghavi, a corporate partner at law firm Fried Frank, says he has recently received a flood of questions from companies wanting to use the technology. They are wrestling with issues such as the legal implications of using models potentially trained on copyrighted data, including images scraped from the web. Some artists complain that AI image tools could undermine human creativity. Image provider Shutterstock announced this week that it would offer an image-generation service powered by OpenAI, while also launching a fund to compensate the original creators of training images. Ghavi says using copyrighted material to train AI models is likely to constitute fair use and thus be exempt from copyright restrictions, but adds that he expects this to be tested in court. Open legal questions and the prospect of malicious uses do not appear to have slowed investor interest. The enthusiasm recalls Silicon Valley's earlier frenzies over social apps and cryptocurrency. And the technology at the heart of this hype cycle could keep the speculative flywheel spinning. VC firm Sequoia Capital demonstrated generative AI's potential in a blog post last month, covering speech synthesis, video editing, biology, and chemistry. A postscript noted that all the images and some text, including future use cases for generative algorithms, had been generated using AI. Translator: Jane #36Kr神譯局 (36Kr Shenyiju)