If you ever wanted to try out the vaunted machine learning toolset from OpenAI, it’s just got a lot easier. The organization has launched an API that allows developers to call their AI software in on “virtually any function in English.”
Basically, if you have a task that requires words to be understood in English, then OpenAI wants to help automate it. Developers have at their fingertips the various features of the GPT-3 family of natural language processing models, at least if you can get into the private test.
It’s a little difficult to describe what it feels like in practice, as it depends on what you want to do. You might rely, for example, on his ability to search at loads of text at once to answer a question about an article or to find a particular section. In a brief demo video, OpenAI illustrates how this works — asking on the Wikipedia bread entry “why is bread so fluffy” returns the section of the article devoted to the texture of bread and how it’s made.
This is a basic one, but more complicated is the popular AI Dungeon. The AI was fed a bunch of D&D sourcebooks and adventures and, on that basis, improvises a player’s journey based on their inputs. While this had previously to be done through sloppy means, essentially running a model version locally, the inputs can now be easily sent via API.
“We hope the API would dramatically lower the barrier to developing beneficial AI-powered goods, contributing to the hard-to-imagine devices and services of today,” the company wrote in a blog post. To be sure it would have been impossible to conceive a year or two ago AI Dungeon.
Also Read: Microsoft will not offer facial recognition to the police until Congress passes a new privacy rule
It’s also made clear that harmful or abusive uses of the API will be banned immediately, which seems safer to the company than releasing the model into the wild. “Since downstream use cases of our models are difficult to foresee, it seems inherently easier to release them via an API and expand access over time, rather than releasing an open-source model where access cannot be changed if it turns out to have harmful applications,” the post reads.
So far, OpenAI has partnered with roughly a dozen companies to test the API ahead of it being offered more widely. Chatbots, teaching aids, legal research — there is no end to this kind of thing’s applications because language is used to define and record pretty much all we do. Janelle Shane of AI Weirdness has already made a dog-rating Twitter bot that creates and rates dogs eternally, so that idea has already been taken up.
Conclusion:
Basically, this is just a lot faster, clearer way to use GPT-3’s large language comprehension and generation capabilities. It is simply texting in, text out.
