“Conduct a comprehensive literature review on the state-of-the-art in Machine Learning and energy consumption. […]”
With this prompt, I tested the brand new Deep Research function, which has been integrated into the OpenAI o3 reasoning model for the reason that end of February — and conducted a state-of-the-art literature review inside 6 minutes.
This function goes beyond a traditional web search (for instance, with ChatGPT 4o): The research query is broken down & structured, the Web is looked for information, which is then evaluated, and eventually, a structured, comprehensive report is created.
Let’s take a better have a look at this.
Table of Content
1. What’s Deep Research from OpenAI and what are you able to do with it?
2. How does deep research work?
3. How are you going to use deep research? — Practical example
4. Challenges and risks of the Deep Research feature
Final Thoughts
Where are you able to proceed learning?
1. What’s Deep Research from OpenAI and what are you able to do with it?
If you could have an OpenAI Plus account (the $20 per thirty days plan), you could have access to Deep Research. This provides you access to 10 queries per thirty days. With the Pro subscription ($200 per thirty days) you could have prolonged access to Deep Research and access to the research preview of GPT-4.5 with 120 queries per thirty days.
OpenAI guarantees that we are able to perform multi-step research using data from the general public web.
Duration: 5 to half-hour, depending on complexity.
Previously, such research often took hours.
It is meant for complex tasks that require a deep search and thoroughness.
What do concrete use cases appear to be?
- Conduct a literature review: Conduct a literature review on state-of-the-art machine learning and energy consumption.
- Market evaluation: Create a comparative report on the very best marketing automation platforms for firms in 2025 based on current market trends and evaluations.
- Technology & software development: Investigate programming languages and frameworks for AI application development with performance and use case evaluation
- Investment & financial evaluation: Conduct research on the impact of AI-powered trading on the financial market based on recent reports and academic studies.
- Legal research: Conduct an outline of information protection laws in Europe in comparison with the US, including relevant rulings and up to date changes.
2. How does Deep Research work?
Deep Research uses various Deep Learning methods to perform a scientific and detailed evaluation of data. The whole process might be divided into 4 predominant phases:
1. Decomposition and structuring of the research query
In step one the tool processes the research query using natural language processing (NLP) methods. It identifies a very powerful key terms, concepts, and sub-questions.
This step ensures that the AI understands the query not only literally, but additionally by way of content.
2. Obtaining relevant information
Once the tool has structured the research query, it searches specifically for information. Deep Research uses a combination of internal databases, scientific publications, APIs, and web scraping. These might be open-access databases comparable to arXiv, PubMed, or Semantic Scholar, for instance, but additionally public web sites or news sites comparable to The Guardian, Latest York Times, or BBC. In the long run, any content that might be accessed online and is publicly available.
3. Evaluation & interpretation of the information
The subsequent step is for the AI model to summarize large amounts of text into compact and comprehensible answers. Transformers & Attention mechanisms make sure that a very powerful information is prioritized. Which means it doesn’t simply create a summary of all of the content found. Also, the standard and credibility of the sources is assessed. And cross-validation methods are normally used to discover incorrect or contradictory information. Here, the AI tool compares several sources with one another. Nonetheless, it will not be publicly known exactly how this is completed in Deep Research or what criteria there are for this.
4. Generation of the ultimate report
Finally, the ultimate report is generated and exhibited to us. This is completed using Natural Language Generation (NLG) in order that we see easily readable texts.
The AI system generates diagrams or tables if requested within the prompt and adapts the response to the user’s style. The first sources used are also listed at the tip of the report.
3. How you should utilize Deep Research: A practical example
In step one, it’s best to make use of one among the usual models to ask how you must optimize the prompt so as to conduct deep research. I actually have done this with the next prompt with ChatGPT 4o:
The 4o model suggested the next prompt for the Deep Research function:
The tool then asked me if I could make clear the scope and focus of the literature review. I actually have, subsequently, provided some additional specifications:
ChatGPT then returned the clarification and commenced the research.
Within the meantime, I could see the progress and the way more sources were steadily added.
After 6 minutes, the state-of-the-art literature review was complete, and the report, including all sources, was available to me.
Deep Research Example.mp4
4. Challenges and risks of the Deep Research feature
Let’s take a have a look at two definitions of research:
“An in depth study of a subject, especially so as to discover recent information or reach a brand new understanding.”
“Research is creative and systematic work undertaken to extend the stock of data. It involves the gathering, organization, and evaluation of evidence to extend understanding of a subject, characterised by a specific attentiveness to controlling sources of bias and error.”
The 2 definitions show that research is an in depth, systematic investigation of a subject — with the aim of discovering recent information or achieving a deeper understanding.
Principally, the deep research function fulfills these definitions to a certain extent: it collects existing information, analyzes it, and presents it in a structured way.
Nonetheless, I feel we also need to concentrate on some challenges and risks:
- Danger of superficiality: Deep Research is primarily designed to efficiently search, summarize, and supply existing information in a structured form (no less than at the present stage). Absolutely great for overview research. But what about digging deeper? Real scientific research goes beyond mere reproduction and takes a critical have a look at the sources. Science also thrives on generating recent knowledge.
- Reinforcement of existing biases in research & publication: Papers are already more prone to be published in the event that they have significant results. “Non-significant” or contradictory results, alternatively, are less prone to be published. This is thought to us as publication bias. If the AI tool now primarily evaluates continuously cited papers, it reinforces this trend. Rare or less widespread but possibly necessary findings are lost. A possible solution here could be to implement a mechanism for weighted source evaluation that also takes into consideration less cited but relevant papers. If the AI methods primarily cite sources which might be quoted continuously, less widespread but necessary findings could also be lost. Presumably, this effect also applies to us humans.
- Quality of research papers: While it is apparent that a bachelor’s, master’s, or doctoral thesis can’t be based solely on AI-generated research, the query I actually have is how universities or scientific institutions cope with this development. Students can get a solid research report with only a single prompt. Presumably, the answer here should be to adapt assessment criteria to present greater weight to in-depth reflection and methodology.
Final thoughts
Along with OpenAI, other firms and platforms have also integrated similar functions (even before OpenAI): For instance, Perplexity AI has introduced a deep research function that independently conducts and analyzes searches. Also Gemini by Google has integrated such a deep research function.
The function gives you an incredibly quick overview of an initial research query. It stays to be seen how reliable the outcomes are. Currently (starting March 2025), OpenAI itself writes as limitations that the feature continues to be at an early stage, can sometimes hallucinate facts into answers or draw false conclusions, and has trouble distinguishing authoritative information from rumors. As well as, it’s currently unable to accurately convey uncertainties.
But it may well be assumed that this function will likely be expanded further and turn into a strong tool for research. If you could have simpler questions, it is best to make use of the usual GPT-4o model (with or without search), where you get a direct answer.
Where are you able to proceed learning?
