Author | vanilla
Edit | Li Shuiqing
Zhidong News on May 11, in the early hours of this morning, OpenAI announced that it will broadcast live on the official website at 10 am US time on the 13th (1 am Beijing time on the 14th) to demonstrate the updated content of ChatGPT and GPT-4.
According to foreign media The Information reported today, OpenAI is building an AI voice assistant with audio and visual understanding capabilities, and some of these features have begun to be shown to customers, and may be previewed at next week's launch event.
According to sources familiar with the matter, OpenAI may complete the development of GPT-5 and release it publicly within this year. In addition, OpenAI plans to launch a new pricing model where customers can enjoy up to a 50% discount by pre-paying tokens.
Previously, a number of foreign media and bloggers revealed that OpenAI will release an AI search engine in the near future. Its CEO Altman responded: "It's not GPT-5, it's not a search engine, but we've been trying to develop something new that we think we'll get liked." ”
▲Altman responds to the new product demonstration (source: X)
It is worth noting that its competitor Google's I/O developer conference is scheduled for 10 a.m. U.S. time on the 14th (1 a.m. Beijing time on the 15th), and it is expected that an important update to the Gemini large model may be released. OpenAI's release time was not set on May 9, which was previously exposed by foreign media, but "just happened" to be stuck on the day before the I/O conference, and the smell of gunpowder was strong.
You know, this is not the first time that OpenAI has "stuck" to snipe Google's new products. In February this year, Google released the Gemini 1.5 Pro big move, and within a few hours, OpenAI took out the Wensheng video "King Bomb" model Sora, and stole the limelight from the Gemini 1.5 Pro.
Live broadcast release address: http://openai.com
1. The AI voice assistant still has the problem of "hallucination", or it will be integrated into the free version of ChatGPT
According to The Information, Altman's ultimate goal is to develop a highly responsive virtual assistant similar to the one in the movie "Her" and improve the usability of existing voice assistants such as Apple's Siri.
▲ Stills from the movie "Her".
OpenAI believes that AI voice assistants with visual and audio capabilities have the transformative potential to be like smartphones, and they can theoretically do a range of things that current AI assistants can't, such as acting as an essay or math instructor, or translating traffic signs to help solve car breakdowns.
However, similar technologies currently require too much hardware to run on personal devices, and users can use cloud-based versions to access these capabilities in the short term, such as automated customer service agents.
OpenAI has already launched software with features such as audio transcription and text-to-speech, but these features are based on independent conversational AI models, and the new voice assistant integrates these features. According to people familiar with the matter, the AI voice assistant's audio function can help customer service agents better understand the tone of the other party.
It's unclear when OpenAI will offer these new features to paying customers, but according to people who have tried the voice assistant, OpenAI's eventual plan is to include all of these features in the free version of ChatGPT, with the goal of running at a lower cost than its current state-of-the-art model, GPT-4 Turbo. The person also said that the AI voice assistant answers better than GPT-4 Turbo in some types of questions, but there is still an illusion problem.
An OpenAI spokesperson did not respond to a request for comment.
Second, compete with Google for the iPhone cooperation position, GPT-5 may be released before the end of the year
OpenAI's main goal in launching the AI voice assistant is to compete with Google.
Google's AI model, Gemini, responds to voice commands in real-time and recognizes images, videos, and more. However, these features still require researchers to be accompanied by images and text descriptions, and they do not understand many traditional voice commands or talk to the user in the same way that traditional voice assistants such as Siri and Google Assistant do.
▲Google shows Gemini's multimodal function (Source: Google)
On the other hand, improving the visual and audio capabilities of the model will also help OpenAI reach a cooperation with Apple.
In recent months, Apple has held discussions with OpenAI about how the next-generation iPhone operating system could integrate OpenAI's models. At the same time, however, Apple is in similar talks with Google.
The latest news from this competition is that, as reported by Bloomberg this morning, Apple is close to reaching a deal with OpenAI to use ChatGPT in iOS 18, the next iPhone operating system; No agreement has been reached with Google, but negotiations are still ongoing.
In addition to the AI voice assistant that may be launched next week, OpenAI has been working on building an AI search engine designed to compete with Google. In addition, it is developing an automated software called a "computer agent" to assist with software development and other computer tasks.
Altman is also working with iPhone developer Jony Ive to develop a standalone AI hardware. However, state-of-the-art AI models are bulky, need to run in the cloud, and require an internet connection to work. It can take months or even years to make a complex AI model with visual and auditory capabilities small enough to run on a device.
However, the most concerned thing may be GPT-5. According to The Information, a person close to OpenAI's leadership said it could complete GPT-5 and release it publicly by the end of this year.
OpenAI's work on new products and AI models is in full swing, while some of its previously announced projects are becoming less of a hit. Although OpenAI has promised developers that it will launch the GPT Store developer incentive program in the first quarter of this year, it has not yet materialized.
3. It is planned to launch a prepaid pricing model with a discount of up to 50%.
In terms of pricing, according to people familiar with the matter, OpenAI plans to launch a new pricing model that allows customers to enjoy up to a 50% discount if they pre-order tokens.
This form of discount is more common in the field of cloud computing, Microsoft Azure, Google Cloud, Amazon Web Services (AWS), etc., all offer discounts on pre-booking server capacity to reduce costs.
At present, on-demand pricing provides APIs (Application Programming Interfaces) as the main form of payment for OpenAI's large models, charging developers a fee ranging from a few cents to more than a hundred cents for every 1 million tokens generated, with large customers enjoying volume discounts.
OpenAI has already provided a way to reduce developer costs, launching the Batch API in April, which allows developers to get a lower price if they upload model queries in bulk and accept to wait up to 24 hours.
AI server resellers such as Together AI and Anyscale have said that running open-source models on their software is six times cheaper than using OpenAI's models. With more flexible pricing, OpenAI will have an edge over other model developers. It has been revealed that its revenue this year could reach billions of dollars.
Conclusion: OpenAI Google competes, Apple and Microsoft benefit?
The competition between OpenAI and Google is becoming more and more fierce, and the "business war" between the two is becoming more and more intense. OpenAI once again set the release time of new products near Google's major release time, and seems to be obsessed with stealing the limelight from Google.
On this side, Apple is also negotiating with OpenAI and Google to integrate the model into the iPhone, which has once again intensified the competition between the two companies. Although OpenAI has been revealed to have reached a cooperation with Apple, negotiations on Google's side are still continuing.
On the other hand, Microsoft, as OpenAI's largest financial backer, can prioritize the use of its advanced technology to improve its voice assistant. In addition to actively developing its own models, Microsoft has also invested in OpenAI's strong rival Mistral AI, which has become the second company to offer commercial AI models on Microsoft's Azure cloud platform.
Next week's OpenAI new product launch and Google I/O conference may push this "business war" to a new height. Whose "big move" will be more powerful, we will wait and see.