OpenAI opens ChatGPT floodgates with dirt-cheap API

'This has to be a loss-leader to lock out competitors'

After a limited trial 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 through API calls.

ChatGPT is a non-intelligent, sentence-predicting language model tuned for responding to questions, and has attracted considerable attention since its November 30, 2022 debut. Presented with a text prompt, it has a fair chance of formulating an acceptable response if the answer can be found in its training data. Or if queried with a jailbreak prompt, it may reply in a way that violates its safety rules.

Whisper is an automatic speech recognition system released last September. It's capable of turning spoken English into text, which can then be input into ChatGPT or used for any other speech-to-text application, like 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 is priced at $0.006 / minute. There's also an open source version of the code, though OpenAI admits it can be hard to run.

Max Woolf, a data scientist, in an online post, observes that that the API pricing is extraordinarily low.

“I have no idea how OpenAI can make money on this,” he said. “This has to be a loss-leader to lock out competitors before they even get off the ground.”

There is of course no guarantee these prices won’t rise later.

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?"}]
}'

Making a call to the Whisper endpoint is similarly uncomplicated.

In preparation for an anticipated rush of paying customers, OpenAI has committed to changes based on feedback from early adopters. This includes: no longer using data submitted via API for service improvements like model training unless customers opt-in; updating its Terms of Service and Usage Policies to clarify "users own the input and output of the models"; and other adjustments.

Furthermore, the company now understands that the stability of its services – something not evident recently – matters.

"For the past two months our uptime has not met our own expectations nor that of our users," the company said in its blog post. "Our engineering team's top priority is now stability of production use cases – we know that ensuring AI benefits all of humanity requires being a reliable service provider."

That's not quite a Service Level Agreement, but something of the sort may be coming.

OpenAI's aspirational plan to ensure that "AI benefits all of humanity" remains sketchy. But surely bringing mechanical chatter to the wretched of the world begins with availability.

The possibility that ChatGPT or some more capable successor could upend the search business has alarmed Google executives and convinced Microsoft that its also-ran search service Bing could do better wedded to a browser sidebar of engaging if not always accurate conversation.

OpenAI has been providing early adopter companies with access to these two models through 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 to each Snapchatter who chats with it."

Other apps integrating the API include: Q-Chat, a study service Quizlet's ChatGPT-powered tutoring service; Shop, 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 non-profit that opened a for-profit arm in 2019 and has since been showered with money from Microsoft – hopes developers will build other apps that make use of their models in novel but remunerative ways. ®