After trying Deep Research, this Wall Street analyst: benefits on both ends regarding employment, will struggle in the middle

DeepSeek's large model has become a sensation, showcasing not only the latest advancements in AI technology but also revealing some important trends in the AI field that have yet to be fully recognized.

After trying Deep Research, Deutsche Bank summarized three points in its report on February 12:

• Cognitive work will undergo revolutionary changes, with both the least and most experienced groups likely to benefit, while the middle tier will struggle;
• Demand for high-end chips will continue to rise;
• The self-improvement capability of AI will further enhance;

1. Cognitive work will undergo revolutionary changes

Deutsche Bank stated that with the development of AI deep research technology, the way humans engage in cognitive work will change significantly:

Humans will focus more on asking the right questions to AI and using judgment to evaluate and iterate on AI's responses, with most cognitive processes being transferred to AI systems.

The least and most experienced groups may benefit, while mid-level talent may face challenges in increasing their value.

However, this transformation also brings potential risks: humans may gradually lose the ability to process information and make complex judgments from first principles.

2. Demand for high-end chips will continue to rise

Furthermore, Deutsche Bank indicated that despite the initial misunderstanding in the market regarding reduced demand for high-end chips due to the emergence of DeepSeek, the reality is quite the opposite:

The DeepSeek technology indicates that the demand for computing power will continue to grow. AI tools are developing in two directions: one is small language models that can run on mobile phones, and the other is large language models that require cloud support. The intensive demand for computing power from enterprises and research institutions, coupled with national security considerations, may drive a positive investment cycle in the chip sector.

3. The self-improvement capability of AI will further enhance

Deutsche Bank also pointed out that the DeepSeek model demonstrates advancements in AI systems in combining reasoning capabilities and tool usage:

The new model can break tasks down into steps rather than simply outputting word-for-word; it also has the ability to use tools such as long-term working memory and internet searches. These advancements bring AI closer to the goals of self-improvement and self-coding, paving the way for the potential emergence of Artificial General Intelligence (AGI) in the future.

For example, in Deutsche Bank's testing, DeepSeek took eight minutes to generate a 9,000-word report, citing 22 sources (with links) and assessing the impact of new U.S. tariffs on steel and aluminum. It is not perfect but will improve with iterations, showing significant enhancements in analysis, relevance, and clarity compared to standard ChatGPT responses.

Additionally, the AI industry is launching deep research tools one after another. OpenAI's latest deep research tool performed excellently in benchmark tests, scoring 26.6% in the extremely difficult Human Ultimate Exam (HLE), more than double that of previous products. Reports suggest that this level is close to that of a PhD.

Notably, the open-source community is also responding actively. According to Reuters, researchers at Hugging Face launched an open-source version called Open Deep Research within a day of OpenAI releasing its deep research tool Although the performance has not yet reached the level of OpenAI models, it has shown excellent performance in multi-step tasks.

Deutsche Bank AG stated that although the current deep research tools are still in their early stages, they have significantly surpassed standard ChatGPT responses in terms of analytical capabilities, relevance judgment, and clarity of expression. With the continuous improvement of underlying large language models, these tools are expected to break through the limitations of simply repeating others' ideas in the future and truly achieve the generation of new concepts