After a limited trial period, OpenAI has unleashed its ChatGPT and Whisper models on developers, who can now integrate chatbot interaction and speech-to-text conversion into their own applications via API calls.
ChatGPT is a language model tailored for answering questions and has attracted a lot of attention since its debut on November 30, 2022. Presented with a text prompt, it has a good chance of formulating an acceptable answer if the answer can be found in the training data. Or if prompted with a jailbreak prompt, it may answer in a way that violates security rules.
Whisper is an automatic speech recognition system released last September. It is capable of converting spoken English into text, which can then be fed into ChatGPT or used for any other speech-to-text application, such as interview transcription.
The ChatGPT model family released on Wednesday, gpt-3.5-turbo, is offered at a price of $0.002 per 1,000 tokens (~750 words), which OpenAI says is 10x cheaper than previous GPT-3.5 models. The Whisper large-V2 model costs $0.006/minute. There’s also an open source version of the code, though OpenAI admits it can be difficult to run.
Max Woolf, a data scientist, notes in an online post that API prices are extremely low.
“I have no idea how OpenAI can monetize this,” he said. “This has to be a loss leader to shut out competitors before they even get off the ground.”
Of course, there is no guarantee that these prices will not increase later on.
Accessing ChatGPT with a curl request looks like this:
curl https://api.openai.com/v1/chat/completions
-H "Authorization: Bearer $OPENAI_API_KEY"
-H "Content-Type: application/json"
-d '{
"model": "gpt-3.5-turbo",
"messages": [{"role": "user", "content": "What is the OpenAI mission?"}]
}'
Calling the Whisper endpoint is also straightforward.
In preparation for an expected rush of paying customers, OpenAI has committed to changes based on feedback from early adopters. This includes: no longer using data submitted through the API for service improvements such as model training unless customers choose to; updating the Terms of Service and Usage Policy to clarify “users own the inputs and outputs of the Models”; and other adjustments.
In addition, the company now understands that the stability of its services – something that has not been apparent in recent times – matters.
“Over the past two months, our uptime has fallen short of our own expectations and those of our users,” the company said in its blog post. “The top priority of our engineering team is now the stability of production use cases – we know that ensuring that AI benefits all of humanity requires being a reliable service provider.”
That’s not really a Service Level Agreement, but maybe there will be something like that.
OpenAI’s ambitious plan to ensure that “AI benefits all of humanity” remains vague. But bringing mechanical chatter to the world’s wretched certainly starts with availability.
The possibility that ChatGPT or a more proficient successor could turn the search business upside down has alarmed Google executives and convinced Microsoft that the search service Bing, also run by them, would be better off combining it with a browser sidebar of engaging, if not always accurate conversation.
OpenAI offers early adopters access to these two models via an API. The result is apps like My AI, a chatbot released this week for Snapchat+ subscribers that the messaging company describes as “a personal sidekick for every Snapchatter who chats with it.”
Other apps that integrate the API include: Q-Chat, a tutoring service Quizlet’s ChatGPT-powered tutoring service; Store, ecommerce platform Shopify’s ChatGPT powered shopping assistant; and Speak, a language learning app that relies on Whisper.
These are just examples of what’s possible, and OpenAI — a nonprofit that opened a for-profit branch in 2019 and has since been showered with money from Microsoft — hopes developers will build other apps that use their models in new yet rewarding ways. ®
Leave a Reply