Shulou(Shulou.com)11/24 Report--

ChatGPT big update in the early morning! Added function calls, 4 times the context length, but also cheaper.

ChatGPT suddenly updated!

Early this morning, OpenAI posted all the update details on its website, including the following:

New function calling capabilities have been added to Chat Completions API

Updated GPT-4 and GPT-3.5-Turbo versions to make it more controllable

Added 16k context length to GPT-3.5-Turbo (previously 4k)

75% reduction in cost of embedded model

25% reduction in input token cost for GPT-3.5-Turbo

The phase-out schedule for the GPT-3.5-Turbo-0301 and GPT-4-0314 models was announced.

Among them, much attention should be paid to the new function calling capability, the upgrade of GPT-4 and GPT-3.5-Turbo models, and various cost reductions.

In response, Twitter user "Baoyu" said that OpenAI's updated function call function essentially allows API to directly support Agent or Plugin.

Function calls developers can now describe functions to GPT-4-0613 and GPT-3.5-turbo-0613, and let the model intelligently choose to output JSON objects that contain parameters for calling these functions.

This is a new way to more reliably connect the functionality of GPT to external tools and API.

These models have been fine-tuned to detect when a function needs to be called (depending on what the user enters) and to respond with a JSON that matches the function's signature.

Function calls allow developers to obtain structured data from the model more reliably.

For example, developers can create a chat robot to answer questions by calling external tools (for example, the ChatGPT plug-in).

Specifically, you can convert a query such as "email Anya to see if she wants coffee next Friday" into a function call such as send_email (to: string, body: string).

Or "what's the weather like in Boston? "convert to get_current_weather (location: string, unit: 'celsius' |' fahrenheit').

In addition, developers can convert natural languages into API calls or database queries.

For example, "who are my top ten clients this month?" "converted to an internal API call, that is: get_customers_by_revenue (start_date: string, end_date: string, limit: int).

Or "Acme, Inc." How many orders were there last month? "converted to a SQL query, which is: sql_query (query: string).

In addition, it allows developers to extract structured data from text.

You only need to define a function called extract_people_data (people: [{name: string, birthday: string, location: string}] to extract all the characters mentioned in Wikipedia articles.

In the past, users needed to add a bunch of descriptions to the Prompt to support Agent, but now it's much easier, as long as a functions parameter is passed in the API request.

The functions parameter is actually similar to the Agent that the user wants to specify.

Note that there can be more than one function,function is an object, and you can also add a name (name), description (description), parameter (parameters), and so on.

The returned result then contains the following structured content:

Function_call: the name of the function called, which is the same as that passed in by the user

The parameter value in arguments:JSON format contains the parameter name and value needed by the user to call the function.

For example:

Function_call: {"name": "get_current_weather", "arguments": "{\" location\ ":\" Boston, MA\ "}"}

Foreign netizens said that the greatest value of function is to solve the problem of structural data returned by GPT, eliminating the need for users to enter complex prompt.

Step 1: OpenAI API

Step 2: third-party API

Step 3: OpenAI API

The main update to the model upgrade GPT-4GPT-4-0613 is the new function calling function mentioned above.

On the other hand, the update of GPT-4-32k-0613 not only contains all the updates of GPT-4-0613, but also adds a longer context length, so that it can better understand the longer text.

More people currently on the waiting list will be able to start using GPT-4 in the coming weeks, OpenAI said.

The update of GPT-3.5 TurboGPT-3.5-turbo-0613 has the same function call function as GPT-4, and provides higher controllability through system information.

These two features will enable developers to guide responses generated by GPT models more effectively.

GPT-3.5-turbo-16k has four times the context length compared to GPT-3.5-turbo, but costs twice as much.

The cost is $0.003 per 1000 token inputs and $0.004 per 1000 output token.

The context length of 16k means that the model can now support about 20 pages of text in a single conversation.

Meanwhile, OpenAI said that the new model will be officially launched on June 27, while the old model will not be accessible again after 9. 13 this year.

The price is lower and finally, many users are moved by the new news of lower prices released by OpenAI.

Text-embedding-ada-002 is the most popular embedded model for OpenAI. Today we reduced costs by 75% to $0.0001 per 1000 token.

As the most popular chat model, GPT-3.5-turbo, its input token fee has been reduced by 25%.

Developers can now use the model at a price of $0.0015 / 1000 input token, at a cost of $0.002 per 1000 output token, equivalent to about 700 pages per dollar.

GPT-3.5-turbo-16k is priced at $0.003 per 1000 input token and $0.004 per 1000 output token.

Reference:

Https://openai.com/blog/function-calling-and-other-api-updates

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.