This is Deep Research from OpenAI, an AI technique that is a little research analyst on your chosen topic. What is that good for? Well, for instance, if you wish to buy a new smart tv, a new car, or know more about a potential company before you invest in it, these are all decisions that require very careful consideration these days.
And in tens of minutes it does the research that for us, would take many hours. So they say, so we will find out whether that is true. As usual, I waited a little for the dust to settle and to see how you Fellow Scholars are already using it to provide you better information.
Okay, so how does it differ from just entering a prompt and getting an answer with a list? Because honestly, that’s what this sounds like and it is something that we have been doing for years now. You see, the difference is that it looks up hundreds and hundreds of sources online, creates a full report out of it, and most importantly: it is not always a dry list of things, it often gives you an opinionated report with real conclusions.
It doesn’t just fetch data, it reasons, and it synthesizes too. That sounds fantastic, if it works. I am a little skeptical for now, but let’s see if I get convinced by this idea by the end of this video.
Now, example. We can ask how the retail industry has changed in the last few years, and…here it comes. I find it interesting that their system encourages narrowing down the topics, and in their example video, they show the user doing the exact opposite and asking, you know, just give me everything about retail.
And then, it starts relentlessly looking up sources, and synthesizing. I wonder what the result will be…well, we are not gonna now because they did not show it. Another interesting tidbit from their example video that I don’t quite understand.
Now, their website shows a few examples, I liked the one about user experience design, where you ask for evidence about something, and normally, you get an answer that starts with “research indicates”. Oh boy, that’s not very helpful is it. But with deep research, you get a bunch of hard information from studies, and I looked up quite a few of them, good news: the studies actually exist.
That is important because these systems are known to hallucinate information in their answers and refer to studies that don’t even exist. Now, as good Fellow Scholars, we still check them anyway, of course. I was not a fan of the rest of the examples on the website, but no matter, because we waited a little, Deep Research has been out there, so we can look up how you Fellow Scholars are using it already for incredible things in 3 examples, and then a huge surprise.
Dear Fellow Scholars, this is Two Minute Papers with Dr Károly Zsolnai-Fehér. One. A complex tax situation.
This Fellow Scholar has to exit the US and has a difficult situation with exit taxes, he sat with two trained accountants, which he had to pay for, and didn’t get a satisfactory result. And then, Deep Research really delivered here. He got an exhaustive report that is fully personalized and tailored to his situation.
I say this is a really good one. Two. Now let’s look at a really well presented example from my friend Alex, who wants to know how these new thinking AIs will impact graphics card sales in the near future.
Asking an AI about itself. Loving it. It once again asks you to narrow it down, which he does.
Then, the AI does a ton of work, and after looking at the result, I say that the research is excellent. So far so good. However, conclusion: “The next 24 months will not only determine who leads the Al model race, but also who leads in providing the silicon brains that make these models possible.
” To me, this sounds a bit like those talking heads on television who speak for a long time, but don’t really commit to a solid conclusion. Three, this Fellow Scholar is using it to create a daily briefing of news. But once again, heavily tailored to you.
You can specify where you live, what your interests are, what kinds of biases or media would you like to avoid, and so much more, I recommend pausing the video and looking at these instructions because this is real real good. Why? You see, I always talk about using AI as a shield against people who wish to misdirect you Fellow Scholars, and now you can get this to cross-check information across multiple media sources, eliminate bias, and get high-quality information every day.
I absolutely love how this shield seems to be coming to fruition more and more every single day. Not every year, not even every week, every day, something new and exciting appears. Especially now that folks are already creating open source alternatives only 12 hours after the release of Deep Research.
Yes, that is what I want. Open source, open science, that means the whole world can work together, and create shields for everyone, for free. What a time to be alive!
Now hold on to your papers Fellow Scholars, because what you have seen here, which is, organizing information is just step number one. Step number two is creating new information. That is in a sense what we research scientists do when we write peer reviewed papers.
I am fairly sure that this is coming very soon, perhaps in less than a year. Or so I thought when looking at these results. But, wait a minute, it can already write a paper of some sort, that is…okay, that is something.
But can it create something truly new? Have a look at this. It makes new propositions, and also goes ahead to make a proper formal argument for it.
Findings of this sort might be an indication that AI is transitioning from analysis to genuine innovation. That is a huge and unexpected game changer. Wow!
Now note that the user is a true Fellow Scholar, an associate professor who says that this result is still not that novel, but I can only wonder how long until truly novel things come out. Not long. So if you take away one thing from this video, it should be this: this year might be the year when we start seeing AI techniques create new knowledge, discover new kinds of medicine to cure terrible diseases, and push humanity forward.
This is incredible! So, yes, I came in a little skeptical, and now I am completely stunned by how much of a game changer this is. And I don’t feel that this video would be complete unless I also said a word about the naming of this feature.
You see, Deep Research is a technique that was shipped by Google DeepMind about two months ago, and now, OpenAI releases something with the same name. Interesting. Now, just pushing out a video on something new and shiny super quick gives you a ton of views, but waiting it out for a little deeper analysis with more examples provides a wider perspective for you Fellow Scholars, so that is what we do here.
Hope you will appreciate it. If you do, please press that like button, subscribe and hit the bell icon if you wish to see more like this, and make sure to check out Lambda.
