OpenAI recently launched Deep Research, an advanced system that uses the internet to summarize information and answer questions effectively. In testing, it excelled on the General AI Assistants benchmark (GAIA), achieving around 67% correct answers in one-shot scenarios. Deep Research integrates a large language model (LLM) with an “agentic framework,” enabling organized multi-step actions like web searching. This innovative approach significantly enhances LLM capabilities, leading to impressive performance on complex tasks. As part of an ongoing effort, the community aims to build an open-source version of Deep Research, inviting collaboration to further improve its functionality and performance.
OpenAI Unveils Revolutionary Deep Research System
Yesterday, OpenAI introduced a groundbreaking system called Deep Research that browses the internet to summarize content and answer questions effectively. This innovative tool has received tremendous praise, especially after we tested its capabilities firsthand.
Deep Research shows a significant improvement in performance on the General AI Assistants benchmark (GAIA). In recent trials, the system achieved nearly 67% correct responses for one-shot questions and 47.6% accuracy on more complex level three questions, which require multi-step reasoning and tool usage.
What sets Deep Research apart is its design, which combines a large language model (LLM) with an internal agentic framework. This unique setup guides the LLM to utilize tools like web searches and organizes its actions in a step-by-step manner. While numerous powerful LLMs are freely available, details about the specific agentic framework behind Deep Research remain limited.
To fully explore this fascinating technology, we embarked on a 24-hour challenge to replicate its results and develop an open-source version along the way.
The GAIA benchmark is particularly challenging for LLM systems, presenting questions that require high-level reasoning and multi-modal capabilities. For instance, one question asks which fruits from a painting were part of a historic breakfast menu, prompting the need for a complex chain of information retrieval and reasoning to provide a correct answer.
So, what happens when we integrate our top LLM into an agentic framework? We’re currently benchmarking our findings against the GAIA challenge to see if we can achieve similar success.
In our initial tests, we’ve seen encouraging progress, improving from around 46% accuracy to over 55% on the GAIA validation set. This accomplishment highlights the advantages of allowing agents to express their actions in code, making tasks significantly more efficient and effective.
As we continue to work on our open Deep Research project, we are welcoming contributions from the community. Our goal is to create a powerful, customizable agentic framework that anyone can use at home with their preferred models.
Overall, OpenAI’s Deep Research represents an exciting advancement in AI technology, and we are eager to explore its possibilities further. Stay tuned for more updates as we continue our journey to unlock the full potential of open-source AI agents!
Tags: OpenAI, Deep Research, GAIA benchmark, AI technology, open-source AI, large language models.
What is Open-source DeepResearch?
Open-source DeepResearch is a project that provides free tools for building and using search agents. These agents help people find information online more easily and quickly by accessing data from different sources.
How can I use Open-source DeepResearch?
You can use Open-source DeepResearch by downloading the tools from their website. Once you have them installed, you can start creating your own search agents to help you search for specific topics or information that interests you.
Is Open-source DeepResearch really free?
Yes, Open-source DeepResearch is completely free. You can use all the features and tools without paying anything. The goal is to make information easier to access for everyone.
Do I need technical skills to use it?
You don’t need to be a tech expert, but some basic understanding of software may help. There are tutorials and guides available to help you learn how to set up and use the tools effectively.
Can I contribute to Open-source DeepResearch?
Absolutely! Open-source DeepResearch encourages contributions from anyone. You can help by reporting issues, suggesting features, or even writing code if you know programming. Your support helps improve the project for everyone.
