OpenAI launches APIs for ChatGPT and Whisper

Share

OpenAI and ChatGPT are in the talks of the world for a long time now and are also trending all around for various reasons.  And now, together with Microsoft, the AI service is on its way to commercialisation.

OpenAI needs to monetize ChatGPT somehow, in order to keep investors off of being too fretful. They announced that they launched an API that will allow any business to build ChatGPT tech into their apps, websites, products and services. OpenAI and Microsoft had already announced in advance that AI would be integrated into Azure and also into their own office products.

Now, they are taking the next step by opening the API of ChatGPT and the audio-to-text model Whisper to developers. This allows AI services to be integrated into apps and products.

Till now, ChatGPT was only available through the AI company’s web interface, and now they are ready to open the API to developers, offering direct access to the ChatGPT model in addition to the GPT-3, GPT-3.5 and Codex models. This, indeed, allows companies to commercialize AI and integrate it into their own services.

“It takes us a while to get these APIs to a certain quality level,” Brockman said.

“I think it’s kind of this, like, just being able to meet the demand and the scale.”

The API service is called gpt-3.5-turbo and costs 0.002 US$ per 1,000 tokens which is ten times cheaper than existing GPT-3.5 models, according to OpenAI, while also being the best existing model outside of chat applications at this current point in time. Early testers migrated from text-davinci-003 to gpt-3.5-turbo with only minor changes to their prompts.

“If you’re building an AI-powered tutor, you never want the tutor to just give an answer to the student. You want it to always explain it and help them learn, that’s an example of the kind of system you should be able to build [with the API],” Brockman said. “We think this is going to be something that will just, like, make the API much more usable and accessible.”

“Grocery shopping can require a big mental load, with a lot of factors at play, like budget, health and nutrition, personal tastes, seasonality, culinary skills, prep time, and recipe inspiration,” Instacart chief architect JJ Zhuang told me via email. “What if AI could take on that mental load, and we could help the household leaders who are commonly responsible for grocery shopping, meal planning, and putting food on the table — and actually make grocery shopping truly fun? Instacart’s AI system, when integrated with OpenAI’s ChatGPT, will enable us to do exactly that, and we’re thrilled to start experimenting with what’s possible in the Instacart app.”

Brockman says that dedicated capacity customers can expect gpt-3.5-turbo models with up to a 16k context window, meaning they can take in four times as many tokens as the standard ChatGPT model. That might let someone paste in pages and pages of tax code and get reasonable answers from the model, say a feat that’s not possible today.

Brockman alluded to a general release in the future, but not anytime soon.

“The context windows are starting to creep up, and part of the reason that we’re dedicated-capacity-customers-only right now is because there’s a lot of performance tradeoffs on our side,” Brockman said. “We might eventually be able to offer an on-demand version of the same thing.”

What is Whisper?

Whisper, which was first introduced by OpenAI in September last year, is an automatic speech recognition (ASR) system trained on 680,000 hours of multilingual and multitask supervised data collected from the web. It also enables transcription in multiple languages and translation from those languages into English.

Companies already using ChatGPT and Whisper APIs

Companies like Snap Inc, the creator of Snapchat, Quizlet, Instacart, Shopify, and Speak are already using OpenAI’s API.

Snapchat, recently introduced My AI for Snapchat+. It offers the users a friendly, customizable chatbot at their fingertips that offers recommendations.

Quizlet, it is a learning platform with more than 60 million students, has worked with OpenAI for the last three years, leveraging GPT-3 across multiple use cases, including vocabulary learning and practice tests. Quizlet is now introducing Q-Chat, a fully-adaptive AI tutor that engages students with adaptive questions.

Whisper API is available through OpenAI’s transcriptions (transcribes in the source language) or translations (transcribes into English) endpoints, and accepts a variety of formats (m4a, mp3, mp4, mpeg, mpga, wav, webm).

Traditionally, GPT models consume unstructured text, represented to the model as a sequence of “tokens.” ChatGPT models instead consume a sequence of messages together with metadata.

A Cheaper Way

The ChatGPT model the OPenAI released, gpt-3.5-turbo, is the same model used in the ChatGPT product. It is priced at $0.002 per 1k tokens, which is 10x cheaper than than their existing GPT-3.5 models.

“Through a series of system-wide optimizations, we’ve achieved 90% cost reduction for ChatGPT since December and we’re now passing through those savings to API users,” company’s Co-Founder Greg Brockman said.

Read more

Recommended For You