🤯 Is Deep Research the biggest thing since ChatGPT?

A couple of weeks ago, OpenAI released their latest model, Deep Research. To use it, you had to pay 200 dollars per month, so I initially passed. But after seeing some user reviews, I decided to give it a try.

It was painful to shell out $200, but damn, it was worth it!

You've actually already read the results. My article about Frankenbabies and what pessimists said about IVF was based on a report that Deep Research compiled.

Reasoning its way to a report

I instructed it to compile how the IVF debate looked before and after the birth of Louise Brown, the first "test tube baby." I specifically asked it to focus on Britain, where IVF development primarily took place, and what famous people, like James Watson, said at the time.

Deep Research got to work and spent nine minutes before producing a ten-page report based on 18 sources. You can follow its work and how it reasons with itself. It began by summarizing the task:

I’m crafting a detailed report on the IVF debate leading up to Louise Brown's birth in 1978, covering media narratives, key opponents, ethical concerns, UK and international perspectives, public reaction, scientific skepticism, and long-term reflections.

Then it did a broad search to form an understanding of the topic:

Searched for 1978 IVF debate ethical concerns.
I’m assessing various materials, including academic articles, Forbes, media coverage, and documentaries, to understand the finer details of IVF's historical context and societal impact.

It decided on the National Library of Medicine where it looked for more information.

This is how it continues to find, evaluate, and then compile information.

Follow-up questions lead to report part 2

The report gave me some ideas for follow-up questions, like what IVF has led to besides just having children. So Deep Research got a new assignment and started again.

It's fascinating to watch it reason. Just like how we sometimes sit and mumble and think out loud. At one point, it encounters problems accessing an article from The Guardian.

"OK, let me see. The Guardian might be blocked."

So it has to find another way, and seems to find the article via National Library of Medicine, but suspects that only part of it is visible there.

"OK, I'm thinking the NCBI aggregator might not show full Guardian piece content, just a tagline. Maybe 'seven ways IVF changed the world Guardian quotes' is worth a try."

After eight minutes I get another report, now based on 29 sources. Together the two reports make up 23 pages or 11,700 words. You can read the whole thing as a pdf here:

Saved me several days of work

If I had done all this research myself, it would have taken days.

I've now done more than thirty reports, on everything from Cambridge Analytica, Jonathan Haidt's book, energy society and DeepSeek to publishing houses and literary agents. I have interest from several publishers so I'm thinking about which one suits the book best. The reports provided deeper knowledge about them. Same with literary agents and how I can approach them.

How's the quality?

I find the quality to be high. But I read about a scientist who used it for their own field. It did a decent job of compiling information and the latest research, but didn't contribute any real depth or new insights for the researcher. Another person asked it to come up with policy proposals, which turned out quite generic.

My own - very early - impression is that it's particularly good at compiling information. If you're already an expert or researcher in the field, it probably doesn't offer much new at all, but if you're not an expert, you get really good help understanding or investigating something.

As almost always when it comes to AI, it's worth remembering the centaurs. It's not mainly about AI doing your job, but how your abilities can be enhanced by AI. The magic happens in the mix of human intelligence and artificial intelligence. Don't get lazy and leave everything to AI. Combine a report with your own research, and use the report to start at the right end and establish a foundation to stand on.

As an expert, it's not what you're best at that you need help with. But maybe your thinking can develop through better knowledge of related areas. To take a completely different example: Let's say you want to write music and can only play guitar. That's possible of course, but if you learn piano and trumpet and actively listen to music genres that are new to you, your general ability as a musician increases. Development is often about being able to connect old knowledge in new ways.

In everyday life, Deep Research will probably mostly be used to absorb relevant information quickly. As a salesperson, you can spend 30 minutes reading a report about a potential customer and get information that would otherwise have taken days to find. Which in practice means you wouldn't have had time to do it at all, and would have settled for some quick Google searches.

Mindblown every other week

I've never been 🤯 so often before as I have since ChatGPT was released. Every other week there's some new AI tool that's incredibly good. That's why it's a bit hard to zoom out and see which of all these are the biggest breakthroughs. My feeling now is that Deep Research belongs among the biggest. Maybe the very biggest since ChatGPT was released.

The reason is the quality (depth) combined with the time it saves. Some of the reports it has generated for me would have taken weeks of work to compile. If you want really deep knowledge in an area, that's exactly what you should do, namely do the work yourself. Read up, search your way forward, write things down. But often you don't need that kind of depth, and don't have time to acquire it either. Now you can get a good way toward that deep knowledge, but in a much shorter time.

Maybe I'm overestimating the usefulness of Deep Research because it's extremely useful for me. I'm constantly gathering information for articles, but can't spend two weeks on a single text. So for me, this help is extremely valuable. At the same time, there are many other jobs where this kind of knowledge and information gathering is important. Not to mention for your private interests.

If you apply Deep Research to all your information and knowledge gathering (where it's practically relevant), you'll simply become much more well-informed and knowledgeable - in the same or less time than you spend on it today.

That's not something to sneeze at.

Mathias Sundin
Angry Optimist