Sarah Guo: Misjudging the timetable of large technology shifts is a common pitfall in investing. I am all-in on a fundamental bet that this shift will drive substantial value creation, but this is a decade+ transition. In the meantime, areas of mispricing have certainly surfaced. In the private markets, a large cohort of investors is trying to figure out how to gain exposure to this technology, or at least how to think about the risk profile around it. And while they're developing a deeper understanding of the space, the tendency has been to anchor to investments with more obvious heuristics. For example, many investors seem to be assessing startups based on whether the people leading them are former researchers at OpenAI or DeepMind, because that’s a much easier question to answer than whether a particular product or research thesis will be successful. Similarly, because databases are a known and well-understood category of software, vector databases are receiving substantial investor attention.
That said, I am already seeing some investors becoming more skeptical because most enterprises haven’t yet adopted generative AI, but this seems short-sighted. Remember that ChatGPT only launched in November; the average enterprise planning and execution cycle tends to be longer than six months. So, investors will need to be patient. As with the internet, mobile, and cloud, some winners emerged immediately, but others only emerged a decade later; discovering the use cases and building great software takes time and entrepreneurial ingenuity. You wouldn’t have wanted to stop your internet investing with Napster.
Here is a link to the full report.
The question of cognition and intention, as they relate to AI models, are key to the development of artificial general intelligence. At present large language models can provide answers that have correct context but that does not imply understanding.
Additionally, the question of intentionality in an AI system implies desire in a manner similar to how humans describe that term. Anthropomorphization of computer systems should really be confined to science fiction.
Getting to general AI is another example of how the first 90% is the easy part and the last 10% is the big challenge. I’ve spent much of the last week in Phoenix. Waymo, Google’s self-driving car start up, has an autonomous taxi service in the city’s inner suburbs. They are an impressive feat of engineering and great fun to ride around in but that is not yet an example of general AI. Instead it depends on lightning fast computing power and rules based decision making.
Nvidia remains the primary vehicle for investing in AI. The share is very overbought but the sequence of higher reaction lows remains intact.