OpenAI have a new feature - >Deep Research. It was actually rolled out in early Feb to Plus users (who paid $100 a month), but on 25th Feb UK Pro ($20 a month) users got to explore Deep Research. Although it seems I can only use it 10 times per month.
What is Deep Research?
It is an AI Agent (a system which can perceive environments, make decisions and take actions autonomously - so once it’s set up it can act all by itself) for making AI-assisted research more effective. It takes in lots of information (it’s environment), does multi-step research by making decisions based on this information and taking different actions (e.g. further research, creating a summary, etc).
It uses the OpenAI o3 reasoning model - which is optimised for web searching and data analysis (making it a great research assistant).
Using Deep Research can take 5 - 30 minutes (so it’s very different than just prompting ChatGPT which is more of an immediate back and forth conversation).
How to Use It?
It’s quite easy - there is a button in the prompt box.
ChatGPT's Deep Research Button.
I asked for a report on trends in AI Upskilling. One thing I was impressed with was the chatbot asking a clarifying question before jumping into heavy research (something ChatGPT does not do).
I recently learnt this nifty trick - I can share my ChatGPT conversation with people, here you go.
Deep Research showing a progress bar (different than the immediacy of talking to ChatGPT).
I liked that it showed me the progress and I could see how many sources it had read.
Deep Research showing the steps it is taking.
If you click on the sources you get to see what it is searching, reading, thinking (reasoning). This is Chain of Thought output and can be really helpful to understand how the AI works.
Showing Deep Research's Chain of Thought in the right hand panel.
The research was completed in 17 minutes and used 29 sources (definitely faster than me). I fact checked a few random bits of the report and they all were accurate from the reference provided. The structure and content is logical, but (I have thought this of every one I have read) boring and long. I struggled to read it all to be honest.
Some of the case studies it included (for example about IKEA’s employee upskilling programme) were actually interesting and despite staying up to date with this topic, I hadn’t heard about this.
If you have ever had to write an academic paper or essay, it is very much like a literature review (where you read every possible paper near your topic and talk around it but with absolutely no personality or opinions). But then I wonder if we want our AI researchers to have opinions? This feels like a rabbit hole so I’ll wrap up.
So I guess if you want an extensive introduction or overview of a topic, then this is definitely a time save to get a report to read with references. I can see it being useful for desk research or background research for a blog. I do wonder if we all start to use this will how we write and research change (but then I remember that happened anyway with search engines like Google becoming popular).
Overall - this feels like a useful addition to ChatGPT, and we’re keen to see how they refine it as more people use it.
Stay up to date
To ensure you’re always in the loop with the fast-paced world of AI, sign up for the Taught by Humans newsletter.