Since ChatGPT’s launch in November, a new mini-sector has emerged that has defied the general slump experienced by tech companies. Not a week goes by that someone doesn’t make public a “generative” artificial intelligence (AI) based on “foundational” models, the large and complex algorithms that provide their intelligence to ChatGPT and other similar AIs. On February 24, Meta, Facebook’s parent company, unveiled a model called LLaMA. It has been reported this week that Elon Musk, the billionaire boss of Tesla and Twitter, wants to create an AI that is less woke than ChatGPT. The catalog maintained by Ben Tossell, a British tech entrepreneur, and shared via a newsletter has recently grown to include, among others, Ask Seneca (which answers questions based on the writings of the Stoic philosopher), Pickaxe (which analyzes our own documents) and Isaac Editor (which helps students write academic papers).
There may be a lot of talk about ChatGPT and other chatbots like it (there’s talk to them, too: ChatGPT may now have over 100 million users), but Tossell’s newsletter hints that the real action in the generative AI arena is more and more in all kinds of less talkative services provided by foundational models.
Not a week goes by without someone releasing a “generative” AI based on “foundational” models.
Each model is trained with large amounts of text, images, sound files, or any other data set. This allows them to interpret, react to, and create natural language utterances, as well as art, music, and any other type of content that can be found on the Internet. Despite the VC industry suffering from a major hangover from the recent tech crash that ended a couple of years of fizz, entrepreneurs experimenting with generative AI have no trouble attracting investment. In January, it was reported that Microsoft had invested another $10 billion in OpenAI, the company behind ChatGPT, on top of the previous investment of $1 billion. A spreadsheet maintained by Pete Flint of NfX, a venture capital firm, lists 539 generative AI startups. Not counting OpenAI, they have collectively raised more than $11 billion in capital so far (see chart). Mike Volpi, of Index Ventures, another venture capital firm, calls what is happening a “Cambrian explosion.”
There are several factors that feed it. Although the foundational models are some time old, Volpi explains that it has taken a consumer-oriented service like ChatGPT to capture the imagination of the world (and investors). And it turns out that it has appeared just as VC investors, disappointed by the cryptocurrency crash and an empty metaverse, were looking for the next big thing. Moreover, even more than web browsers and smartphones before it, foundational models make it easier to build new services and applications on top of them. “You can open the laptop, create an account and start interacting with the model,” says Steve Loughlin of Accel, another venture capital firm.
The question for venture capitalists is which generative AI platforms will make the most money. At the moment, that is the topic that provokes the most debates in technological circles. “Judging from the available data, it is not clear whether there will be a winner-take-all dynamic in generative AI in the long run,” Martín Casado and colleagues at Andreessen Horowitz, another venture capital firm, wrote in a recent blog post. Many startups offer similar ideas, many of which are more of a feature than a product. Over time, even resource-intensive foundation models could become a low-margin commodity: while proprietary models like OpenAI’s GPT-3.5, which powers ChatGPT, still lead, there are other open source models that don’t. they are lagging behind.
Another source of uncertainty is the legal minefield that generative AI tiptoes through. Foundational models are often wrong. And they can derail. The chatbot that Microsoft is developing for its Bing search engine based on OpenAI models has insulted more than one user and professed its love to at least one other (since then, Sydney, which is what the chatbot is called). from Microsoft, has already been put on the rails). It is possible that generative Internet platforms do not enjoy the same legal protection against liability that protects social networks. Some copyright holders of web content from which existing models are trained willy-nilly, without asking for permission or paying any compensation, have already taken up arms. Getty Images, a photo repository, and various individual artists have already filed lawsuits against AI-based art generators like Stable Diffusion. News organizations whose articles are ransacked for information can do the same. Stable Diffusion states: “We take these matters very seriously. We are reviewing the documents and will respond accordingly.”
One source of uncertainty is the legal minefield that these technologies represent: Microsoft’s chat bot for its Bing search engine has already insulted more than one user
OpenAI is already trying to manage expectations and plays down the planned release later this year of GPT-4, the long-awaited new version of the foundational model behind ChatGPT. That’s unlikely to calm VC appetite for generative AI. For the most risk-averse investors, the safest bet right now is with providers of the extensive processing power needed to train and run foundational models. The share price of Nvidia, which designs chips useful for AI applications, has risen 60% so far this year. Cloud computing services and data center owners are also rubbing their hands. Whichever AI platform ends up taking over, you can’t go wrong selling picks and shovels in a gold rush.
© 2023 The Economist Newspaper Limited. All rights reserved
Translation: Juan Gabriel López Guix
#Investors #crazy #artificial #intelligence #ChatGPT