projexgame| Big model companies are starting to compete for space at the application level

On the eve of the outbreak of the application layer of the AI model, two more unicorns releasedProjexgameA new product.

In the past week, the Dark side of the month and 010 million things have released Kimi+ and Maanzhi one after another, and both products are positioned as AI professional assistants.

Among them, Kimi+ has built a set of agent store, including office assistant, research report generation, academic search, prompt word design, auxiliary writing and other scenes; Universal knowledge is a set of one-stop AI working platform, covering writing assistant, document interpretation, PPT graphics and text production and other functions.

Since the first half of the year, players on large model tracks such as iFLYTEK, Shangtang Technology and Lanzhou Technology have all iterated over the new functions of their big models at the product level, and almost all of them are aimed at the "workflow" link.

This is the current mainstream product trend in the field of large models, but how to understand it technically and commerciallyProjexgame?

An industry source told the interface journalist that technically speaking, the update of the product direction of "AI Professional Assistant" is not complicated, but more of a normal iteration in the engineering sense. He summed it up as a "mild" agent whose core action is to solidify part of the prompt for users to use directly, which can help users better put forward their requirements to AI.

In his view, these products only have a slight improvement in the user experience and "don't make much sense." Of course, this is mainly compared to the ability breakthroughs of killer applications and underlying large models.

At the first developer Conference in November 2023, OpenA officially released GPTs (Custom GPT), which essentially allows users to combine the capabilities of GPT-4 with the product form of ChatGPT in a more flexible way, so as to quickly create a low-threshold AI Agent (agent) for personal use.

"if you want to go to a deeper intelligence, the action ability of the agent itself will be very important, and this is deeply tied to the model ability." He pointed out that whether it is Kimi+ or omniscience, the model capability behind it is likely to be in GPT-3.ProjexgameBetween .5 and GPT-4.

Based on this point, the change of product form alone is not enough to have a great impact on the C-end (user) market, because it is still the ability of the underlying model that ultimately affects users' perception.

Although it is not difficult to achieve technically, nor is it a product innovation that can stir up a huge splash on the C side, no big model company will refuse to join this trend.

Judging from the iterative path of the application layer alone, the aforementioned industry insiders pointed out that whether it is a generative conversation robot or an AI professional assistant, each lightweight agent is ultimately part of the "super App" and is a necessary option for the iteration of To C products of large model companies. "Super App" means AI 2.ProjexgameKiller applications that integrate a variety of functions and services in the .0 era.

In addition, follow-up AI professional assistant and other functions are also necessary for large model companies to undertake the explosive period of the application layer and obtain dividends.

In China, the predictions about the arrival of the AIGC era or the upcoming outbreak of AI applications have actually reached a consensus between entrepreneurs and investors.

Kaifu Lee, CEO of all things, said bluntly that the decline in the reasoning cost of large models will push China's AI large models into the stage of landing as king, and this year will usher in the "first year of the outbreak of large model applications." Zhu Xiaohu, GSR Ventures's managing partner, has also made it clear many times that the application layer of the AI model will erupt in 2025.

On this basis, another consensus is that the underlying general model may end up with only a small number of players, but the entrepreneurial ecology of the application layer will be relatively prosperous. Therefore, in the case that the AI application industry will not be dominant, each product test at this stage is actually a horse race enclosure on the eve of the outbreak.

No matter how high the ability level of the general model company is, it ultimately depends on the business model of the product, so while intensively promoting the model ability, the team must also build a product moat with cash ability-- this is also the most true interpretation of R & D and application of "two-wheel drive" theory in start-ups.