Source: Finance Association
Author: Shi Zheng Cheng
① A more powerful GPT-4 Turbo model has been introduced, and the API fee has been reduced by 2-3 times compared to GPT-4; ② ChatGPT's product form has also changed; ③ the function of building custom GPT through natural language has been released, and the “GPT App Store” will be launched this month.
On Monday, local time, the leader in the artificial intelligence industryOpenAI's first ever developer conference opens。 In a nearly 45-minute opening speech, OpenAI CEO Sam Altman showed developers and ChatGPT users around the world a series of upcoming product updates from the AI leader.
At the beginning of the press conference, Ultraman briefly reviewed the company's development history over the past year, specifically mentioning “GPT-4, which was released by the company in March of this year, is still the most capable AI model in the world.” As of today,2 million developers are already using OpenAI's API(application program interface), providing a variety of services around the world;92% of Fortune 500 companiesOpenAI is being used to build services, whileThe number of active ChatGPT users also reached 100 million.
Introducing the GPT-4 Turbo model
After that, the new product launch period began. The GPT-4 Turbo model debuted first.
Simply put, compared to GPT-4, which is familiar to global shareholders, the progress of the GPT-4 Turbo is mainly reflected in6 aspects.
1. AI can understand a longer context conversation length (context length). The standard GPT-4 model supports up to 8192 tokens. The previous upgrade raised this to a maximum of 32,000 tokens. whereasGPT-4 Turbo supports up to 128,000 tokens, which is equivalent to the amount of text contained in a standard-sized 300-page paper book; Ultraman also said that the accuracy of the new model in processing long texts has also improved;
2,Giving developers more control。 The new model allows developers to instruct the model to return a valid JSON—JSON pattern in a fixed format. At the same time, developers can access the seed parameter and system_fingerprint response field to achieve “the model gives a definitive output for every request”.
3. GPT-4's knowledge of the real world ends in September 2021.Knowledge of GPT-4 Trubo ends in April 2023.
4. The multi-modal API is here.The Wenshengtu model DALL·E3, GPT-4 Trubo with visual input capabilities, and the new sound synthesis model (TTS) all entered the API today。 OpenAI also released a new speech recognition model, Whisper V3, today, and will provide an API to developers in the near future.
5. Following the opening of GPT 3.5 fine-tuning (Fine-tuning) to global developers,OpenAI Announces It Will Qualify Active Developers for GPT-4 Fine-Tuning。 For developing vertical AI applications in industry segments, fine-tuning is a necessary process. For such developers, OpenAI has also launched a customized model project to help some organizations train customized GPT-4 models for specific fields. Ultraman also said that this thing won't be cheap at first.
6. OpenAI will ownToken rate limit for GPT-4 users has been doubled, developers can apply for a further increase in speed.
Similar to Microsoft and Adobe, OpenAI has also introduced a “copyright shield” mechanism. When ChatGPT enterprise users and API users are subject to copyright lawsuits, the company will defend itself and pay compensation arising from this.
In terms of pricing that the market is extremely concerned about, as an industry-leading large model, the price of the GPT-4 Turbo is far lower than GPT-4. whichThe price of entering a token is only one-thirdwhileThe price of exporting tokens is only half。 In other words, the price for entering 1000 tokens is 1 cent, while the price for exporting 1,000 tokens is 3 cents. At the same time, the price of the GPT-3 Turbo 16K model has also been lowered.
On the sidelines of the new product launch, Microsoft CEO Nadella also came to the scene, boasting about OpenAI and once again emphasizing that “Microsoft loves OpenAI.”
ChatGPT also has an update
Ultraman announced that although today is a developer conference, OpenAI couldn't help but make some updates to ChatGPT.
First, starting today, ChatGPT will use the GPT-4 Trubo model that was just released. In addition, in response to the problem that users have to choose a different mode before each conversation, the GPT-4 Trubo will also welcome an update in product logic. Now the robot can actively adapt the corresponding functions according to the conversation.
Before the update ⬇️
After the update ⬇️
Next, the second important product of this event will be unveiled——gPts。 Users will be able to customize instructions, expand (model) knowledge boundaries, and issue action ordersBuild your own GPT, and can be released to more people around the world for use. More importantly, the entire process of building a “custom GPT” is also shaped through natural language dialogue.
Ultraman also demonstrated live how to set up a GPT through chat. His instructions to GPT Builder were”We want to help entrepreneurs think about business inspiration and provide advice, then 'torture' them why their company isn't growing fast enough.”
Afterwards, ChatGPT quickly set up a business consulting GPT, and also created a logo.
Ultraman then uploaded a copy of his own speech about the startup company on the property page.Additional knowledge was provided to this use case。 At this point, the initial construction of GPT for custom purposes has been completed. Users can save this use case for their own use or publish it publicly.
Since the “public release” was mentioned, OpenAI also followed suit and announced,The “GPT App Store” will be launched later this month。 For those of the hottest GPT,The company will also share a portion of revenue to promote the progress of the GPT application ecosystem.
Assistent API
Finally, there's the new Assistive API for developers. An “assistant API” is a specially built artificial intelligence that has specific instructions, uses additional knowledge, and can call on models and tools to perform tasks. The new helper API provides functions such as code interpreters, retrieval, and function calls to handle many of the heavy lifting that developers previously had to complete on their own.
According to OpenAI, the range of use cases for this API is very flexible, such as natural language-based data analysis applications, programming assistants, AI vacation planners, voice-controlled DJs, intelligent visual canvases, etc.
As an example, OpenAI's chief developer experience officer Romain Huet set up a “know the whole developer conference” use case and called Whisper to achieve voice input.
At the same time, since this API can be connected to the Internet, Romain also used voice instructions to AI on site to randomly select 5 live viewers and deposit 500 US dollars each into their OpenAI account.
As the last surprise part of this event, Romain once again issued instructions to AI, depositing $500 each into the accounts of everyone on site.
edit/new