ChatGPT proves AI is finally mainstream — and things are only going to get weirder

A friend of mine texted me earlier this week asking what I thought of ChatGPT. I wasn’t surprised that he was curious. He knows I write about AI and he’s the kind of guy who keeps up to date with all the online trends. We chatted a little, and I asked him: “and what you think ChatGPT? To which he replied, “Well, I wrote a half-decent Excel macro with this morning that saved me a few hours at work” — and my jaw dropped.

For context: this is someone whose job involves quite a bit of database manipulation, but who I wouldn’t describe as particularly tech-savvy. He works in higher education, studied English in college, and never formally learned to code. But there he was, not only playing with an experimental AI chatbot, but use it to do your job faster after only a few days of access.

“I asked him a few questions, asked some more, put it into Excel, then did some debugging,” is how he described the process. “It wasn’t perfect, but it was easier than googling.”

Tools like ChatGPT have made AI publicly accessible like never before

Stories like this have been piling up this week like the first drops of rain before a downpour. On social media, people shared stories of using ChatGPT to write code, write blog posts, write college essays, compile work reports, and even improve their chat game (okay, the latter was definitely done as a joke, but the prospect of an AI-augmented rizz is still tantalizing). As a journalist covering this space, it’s been fundamentally impossible to keep up with everything that’s going on, but there’s one overarching trend that has stuck: AI will main streamand we are only beginning to see the effect it will have on the world.

There’s a concept in AI that I particularly like and that I think helps explain what’s going on. It’s called “overshoot” and refers to the hidden capabilities of AI: skills and abilities latent in systems that researchers haven’t even begun to study. You may have heard that AI models are “black boxes” – that they are so huge and complex that we don’t fully understand how they work or come to specific conclusions. This is largely true and this is what creates this overhang.

“Today’s models are far more capable than we think, and our techniques available to explore [them] are very juvenile,” is how AI policy expert Jack Clark described the concept. in a recent edition of its newsletter. “What about all the abilities we don’t know about because we didn’t think to test them?”

Overflow is a technical term, but it also perfectly describes what is happening right now as AI enters the public domain. For years, researchers have been on a tear, churning out new designs faster than they can get to market. But in 2022, a glut of new apps and programs have suddenly made these skills available to the masses, and in 2023, as we continue to expand into this new territory, things will start to change – fast.

The bottleneck has always been accessibility, as ChatGPT demonstrates. The bones of this program aren’t entirely new (it’s based on GPT-3.5, a large language model that was released by OpenAI this year but is itself an upgrade to GPT-3, from 2020). OpenAI has already sold access to GPT-3 as an API, but the company’s ability to improve the model’s ability to speak in natural dialogue and then publish it to the web for anyone to play with , brought it to a much wider audience. And no matter how imaginative AI researchers are in probing a model’s skills and weaknesses, they’ll never be able to match the massive, chaotic intelligence of the internet as a whole. As a result, the overhang is accessible.

The same dynamic can also be seen in the rise of AI image generators. Again, these systems have been in development for years, but access has been restricted in various ways. This year, however, systems like Midjourney and Stable Diffusion allowed anyone to use the technology for free, and suddenly AI art is everywhere. This is largely due to Stable Diffusion, which offers an open source license that companies can rely on. In fact, it’s an open secret in the AI ​​world that every time a company releases a new AI image feature, there’s a good chance it’ll just be… a refurbished version of Stable Diffusion. This includes all of viral application “magical avatar” Lensa Canva’s AI-powered text-to-image tool for MyHeritage’s “AI Time Machine”. It’s the same technology underneath.

However, as the metaphor suggests, the prospect of a capacity overhang is not necessarily good news. In addition to hidden and emerging capabilities, there are hidden and emerging threats. And those dangers, like our new skills, are almost too numerous to name. How, for example, will colleges adapt to the proliferation of AI-written essays? Will the creative industries be decimated by the dissemination of generative AI? Will machine learning create a tsunami of spam that will ruin the web forever? And what about the inability of AI language models to distinguish fact from fiction or the proven biases of AI image generators that sexualize women and people of color? Some of these issues are known; others are ignored, and still others are just beginning to be noticed. As the excitement of 2022 fades, 2023 is sure to contain some rude awakenings.

Welcome to the AI ​​Overlook. Hold on.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top