Author of this article: Li Xiaoyin

Source: Hard AI

The day after OpenAI launched a major new product, Google also “stepped up” to directly compete against GPT-4O.

On Tuesday, May 14, local time, at the annual Google I/O developer conference, Google CEO Sundar Pichai announced a series of new products and features related to AI, including: AI Overview technology generation summary function, Gemini 1.5 Pro context window expansion to 2 million tokens, multi-modal Gemini Nano model, and sixth-generation TPU chip Trillium.

In terms of AI search engines, Google has brought a series of updates. It is worth mentioning that Google released Astra, a multi-modal AI project to process multi-modal input content such as audio and video.

The demo video shows that Astra can identify objects through a mobile phone camera, and can also recognize where they are located.

Whether in terms of positioning or functionality, the arrival of Google's AI assistant clearly poses a threat to GPT-4O.

Chirag Shah, a professor at the University of Washington who specializes in online searches, commented:

“Ultimately, you'll have an agent who really knows you, can do a lot for you, and execute orders across tasks and fields.”

Google also said at the press conference that starting this summer, Gemini will also support real-time voice interaction, and will launch real-time video interaction later this year. In the next few months, Google will also launch a custom AI assistant function similar to GPTS, called Gems, which can be linked to the entire “Google Family Bucket”.

First AI “killer app”?

Judging from OpenAI and Google's press conference, GPT-4O can currently only process still images, but Astra can handle video, which is a significant advantage.

Furthermore, Google also made many updates to the Gemini 1.5 Pro large model at the press conference, so that it can have more natural sound, longer conversations, better understanding of audio and images, more logical reasoning and planning capabilities, and better code generation.

However, the technological innovation behind GPT-4o is also impressive. It is reported that the native multi-modal model can directly receive/generate speech without going through a speech-to-text conversion process, greatly shortening the operating cycle; moreover, the number of parameters required to perform the task is also greatly reduced, thereby increasing the speed of operation and reducing costs.

As far as current progress is concerned, it is difficult to determine which of OpenAI or Google's AI assistants is superior, but there is no doubt that they both place importance on this field.

According to previous media reports, Apple is also considering introducing GPT technology into its mobile phone voice assistant Siri to support AI functions.

Tech giants are making efforts one after another, does it mean that AI assistants will become the next AI “killer app”?

The answer is uncertain.

Some analysts pointed out that while the use cases currently shown by GPT-4O and Astra are interesting, “almost none” help people get the job done. In other words, these two AI assistants seem to be powerful, but their actual utility is still unknown.

According to the analysis, if AI assistants can better understand users' personal preferences in the future, their “agent” attributes may be enhanced to help users actually complete their daily tasks, such as online shopping, reservations, and filling out forms...

What do AI phones need to solve next?

Although OpenAI and Google's AI assistants can be operated directly through voice, video, etc., there are opinions that the two cannot be called an AI assistant.

The reason is that although GPT-4O and Astra can both answer questions and perform search work, they can't actually perform the task.

Wall Street has heard that before, OpenAI's pain points in developing edge AI are: end-side application permissions and system-level permissions. This is probably one of the reasons it is seeking collaboration with Apple.

As of now, as long as AI assistant products are not actually connected to the mobile phone system, the status of voice assistants such as Siri cannot be shaken.

According to some opinions, certainty is more important than AGI (General Artificial Intelligence), and reliability comes first.

According to this opinion, even the best AI systems at present are not prepared enough to actually implement the functions of personal assistants; and although the voice assistants that come with phones aren't that “interesting,” at least they won't go wrong.