A bustling cityscape filled with towering skyscrapers and futuristic technology, all powered by the constant chatter of LLMs and AI assistants. DALL·E / OpenAI
Editor's Note: Getting The Disconnect’s inaugural post out the door the week of Thanksgiving here in the US was, unsurprisingly, overly ambitious. But the wheels are turning now! Thanks to all of you who subscribed.
For decades, Hollywood has driven the modern perception of artificial intelligence (AI). In movies like "2001: A Space Odyssey," "Blade Runner," and "The Terminator," AI is typically depicted as either a utopian or a dystopian end-state. This has led many to view the idea of human-like AI as a distant, fictional concept.
But in recent years, the conversation around AI has shifted from science fiction to practical reality. Large language models, or LLMs, have become particularly prominent in the past several years. Large language models are a type of AI typically trained on billions of parameters of text data, allowing them to understand and generate human language. This training enables the models to perform a wide range of language-related tasks, which, as it turns out, is where a lot of practical applications of AI technology lie.
Much of this rapid change in perception has been driven by just one company, OpenAI. Starting with GPT-3, a text-based language model released in mid-2020 that can generate human-like text, and more recently, with the release of DALL·E 2, an AI capable of creating images from text, OpenAI has led the charge in showcasing what's possible with AI to an increasingly mainstream audience.
OpenAI is by no means the only company trying to democratize access to the capabilities offered by LLMs, however. Recently, there's been an explosion of companies putting their own touch on similar concepts, including Google with their LaMDA model and Amazon's AlexaTM.
Interestingly, another tech behemoth, Microsoft, presciently opted to partner directly with OpenAI back in 2019, taking on the role of “preferred partner” when considering whether and how to commercialize new OpenAI technologies as they become available. An example of this partnership can be found in Microsoft’s move to include direct DALL·E integration in its upcoming Canva competitor, Microsoft Designer. A feather in the cap of whoever negotiated that arrangement.
But despite the ongoing language model renaissance, nothing seems to have captured the public's imagination in the way OpenAI's newly announced ChatGPT model has. ChatGPT is a language model that can engage in natural conversation with humans, with the ability to understand and respond to complex questions and statements. In practice, this format results in much more fluent and context-rich exchanges than other, more typically open-ended and ephemeral offerings.
ChatGPT is accessed through a relatively sparse web page reminiscent of those clunky customer service chat windows. There's no API just yet (despite what the model itself will tell you when asked), though if history is any indication, we'll likely see one sooner rather than later. The landing page provides some helpful guidance on what the service can and can't help with, ambiguously emphasizing that "inappropriate requests" may be declined.
Notably, unlike GPT-3 or DALL·E, OpenAI opted to release ChatGPT to the public without a waitlist or private release. Part of this is likely due to the increased confidence in the company’s Moderation API. Still, it may just as well be a calculated bet that the company can iterate faster than the public’s ability to abuse all things AI. This has led to a deluge of interesting and terrifying interactions as people test the boundaries of the model and OpenAI's relatively conservative restrictions on its use.
Many industries would be wise to pay attention to these latest developments. For example, some on Twitter have pointed out that ChatGPT is a natural escalation, even leapfrogging, of Google's "snippets" feature, surfacing insights and content from sources on the search results page directly. Others have voiced concern about what this means for the future of education, particularly given the outsized importance of essays and other virtual homework.
Using it myself over the past couple of days has been a fascinating peek into a future where virtual assistants might not be limited to prewritten scripts and narrow categories of usefulness. I've asked for help curating music playlists, suggesting where to relax while on the Las Vegas strip, and even drafting portions of this newsletter (generated content was edited for tone and consistency but was surprisingly helpful).
ChatGPT is free for now in what OpenAI calls "research preview." Though it will likely adopt a similar billing model to its sister service, GPT-3, when the company feels they've learned enough.
Have you tried out ChatGPT or any of the other OpenAI offerings? Where else might it cause trouble or disrupt existing industries? Let me know what you think at [email protected].