"Crafting a perfect prompt for AI chatbots is often a challenge - so much so that startups are creating roles for prompt engineers. Consumer-facing AI apps are increasingly adding features like suggestion buttons or autogenerated recommendations to nudge customers to use the chatbot more frequently and show them about what the app can do. Hero, a productivity startup founded by former Meta employees, announced a new autocompletion SDK today that will fill in prompts for you based on context."
"Here is how it works: When you are booking a flight, you might write "Book a flight" in the prompt. The autocomplete SDK will start populating fields like "to," "from," "on date," "at time," "airline," and "returning on." You can stop at any point and send the query to the chatbot. This is also useful when you are using an AI-powered image or a video generator, where autocomplete can help you with different parameters like object, style, location, landscape, and camera angle."
"Saharsh Vedi, who worked on this feature, said that usually you usually need a lot of back-and-forth with an AI app to get the results you want. With this autocomplete feature, you can get there with fewer prompts - or even just one. The startup's co-founder, Brad Kowalk, told TechCrunch that the company is using a series of models to predict what you might type into a prompt next."
An autocompletion SDK fills in prompt inputs based on context so users can complete actions with fewer interactions. The SDK populates structured fields such as destinations, dates, times, airlines, and return dates for travel tasks and suggests parameters like object, style, location, landscape, and camera angle for image and video generation. Segmented keyword inputs can also apply to creative tasks such as soundtrack creation with mood, style, and purpose. A series of predictive models estimates likely next inputs, reducing repeated prompts and speeding completion by up to tenfold. The SDK is currently invite-only for developers.
Read at TechCrunch
Unable to calculate read time
Collection
[
|
...
]