"There are quite a lot of misunderstandings about the big model outside." Recently, an internal speech by Robin Lee was exposed. Robin Lee believes that the gap between big models may become larger and larger in the future. He further explained that the "ceiling" of the large model is very high, and it is still far from the ideal situation, so the model needs to be constantly iterated, updated, and upgraded quickly; We need to invest continuously for several years or even decades to meet user needs, reduce costs, and increase efficiency.
Robin Lee gave a different view on the industry's statement that "there is no barrier to the ability of big models": "Every time a new model is released, it should be compared with GPT-4o, saying that my score is almost the same as it, and even some individual scores have exceeded it, but this does not mean that there is no gap with the most advanced models."
He said that many models, in order to prove themselves, will go to the leaderboard after release, guessing test questions and answering skills. From the leaderboard, perhaps the models' abilities are already very close, "but in practical applications, there is still a significant gap in strength.
Robin Lee pointed out that the gap between models is multi-dimensional. The industry often focuses more on the gap in understanding, generation, logic, memory, and other abilities, but neglects dimensions such as cost and reasoning speed. Some models can achieve the same effect, but their cost is high and reasoning speed is slow, which is still not as good as advanced models.
Robin Lee also said that "before the era of big model, people were used to open source, which means free and low cost". He explained that, for example, open-source Linux is free to use because computers already exist. But these are not valid in the era of big models. Big model inference is expensive, and open-source models do not provide computing power. You have to buy your own equipment, which cannot achieve efficient utilization of computing power.
Open source models are not effective in terms of efficiency, "he said." Closed source models should be accurately called commercial models, which are machine resources and GPUs used by countless users to share research and development costs and inference. GPU usage efficiency is the highest, with Baidu Wenxin Big Model 3.5 and 4.0 having GPU usage rates of over 90%
Robin Lee believes that the open source model is valuable in teaching and scientific research; But in the business world, when pursuing efficiency, effectiveness, and lowest cost, open source models have no advantages.
At the application level of the big model, Robin Lee believes that Copilot is the first one to assist people; Next is the Agent intelligent agent, which has a certain degree of autonomy and can use tools, reflect, and evolve on its own; If this level of automation continues to develop, it will become an AI worker capable of independently completing various tasks.
He also stated that although "many people are optimistic about the development direction of intelligent agents, so far, intelligent agents are not a consensus, and there are not many companies like Baidu that regard intelligent agents as the most important strategy and development direction for large models.
Robin Lee believes that the threshold for agents is really low. Many people do not know how to turn a large model into an application. Agents are a very direct, efficient and simple way. It is very convenient to build agents on top of models.