Why ‘Deep Research’ is My AGI

This article is going to be about why the Deep Research product recently released by OpenAI is what I've truly been waiting for this whole time¹; why, therefore, I'm declaring it my AGI.

So to start, I love the internet. I love using the internet. I love being able to discover things on the internet. I love consuming the acts of creation that people share on the internet. Generally, I love how the internet enables you to learn just about anything you want.

I’m just a generally curious person. I have a lot of questions, and the internet has a lot answers.

Seems simple. Me = has questions. Internet = has answers. Problem = solved. Right? Wrong.

See, the internet does indeed have answers, but they are sometimes so hard to get. The internet has turned into a maze game where you’re constantly in a fight with the creator of the game to define and find the exit. What makes this game even harder is that the creator has asymmetric access to the resources and power that create the maze. How do you get out with what you want in this scenario?

A bit dramatic, yes, but I think there’s something to be said about how when the internet was more nascent, it was easier to at least get what you wanted, and leave. I think as the internet has become more of a dopamine addiction machine with very advanced machine learning algorithms trying to capture every bit of our attention made by people getting paid exuberant amounts of money to optimize to do just that, it has become a lopsided game. In most casual everyday scenarios, I am not as powerful as these algorithms and my attention very easily gets picked to pieces. It’s an attention economy, they say. Well, I’m already fighting an internal battle to regulate my attention every day just by waking up, I certainly don’t want a deeply-funded, researched, and powerful external agent to also split my attention pie².

This exact problem that has been created - being able to access the information on the internet effectively - is why I was so excited about Arc from The Browser Company³. The Browser Company understands that the internet is overwhelming, and they wanted to try and solve it that problem at the browser layer, by building a better browser. A browser that “helped us make sense of it all”? Beautiful. Sign me up. I mean, I did sign up (and I’m such a fan I tell my friends to, too). So for Arc to want to abstract all that maze-game-bullshit away and just let you literally get what you want out of the internet by changing how you interacted with your browser, I was really excited. I continue to be excited. I think the Arc team made a lot of progress towards that goal. My favorite example of this is a feature they implemented where you can simply Command G on your keypad to change the search box from a normal google search to a “search” routed to chatGPT (because in theory, chatGPT has the answers… more on that soon). So instead of the standard answer-retrieval route of searching in google, navigating multiple links and ads and content overload and attention-mining algos - like the normal search bar would - you can send your questions straight to ChatGPT. Which at the time was novel - ChatGPT didn't have an internet search feature, but positioned the LLM model as the answer retrieval machine for your question. Not Google. I love that.

And then ChatGPT eventually came out with the ability to search the web (i.e. get “real-time” information/answers) and has since made a lot of improvements to that feature. I have found that to be very helpful and I used it a lot over the standard search route. But at times, it just really wasn't delivering on what I wanted it to be. I wanted it to be the only place I needed to go on my quest to get answers from the internet. In maybe 20% of my cases, it did that. But for the majority, and specifically for those questions that were more complex, it still felt like I was being fed links with incomplete information and to get the answer I wanted, I still had to do some manual digging.

So, to summarize, ChatGPT's ‘search the internet’ feature gave me direction. It was a good starting place. I didn’t have to go straight to the search bar on Google and try to filter through a bunch of links, getting distracted while trying answer the question I have. It gave me direction to the answer. ‘’Here, try just these 5 links”. What’s clearly still lacking is: the answer. I still had to browse the internet to find that.

Therefore, my core problem - getting answers to my questions - was still not solved. I still couldn't find an answer in the format that I wanted without having to go to the internet and try to out-game all these algorithms and not get distracted along the way.

Along comes Deep Research. And I think, yeah, that's cool. But I'm not paying $200 a month. I’m not breaded like that. So I didn't even consider buying it. I just thought, “Wow, this is great. I like where this is going. But this isn’t for me. $200 isn’t for me”. Yeah, I was wrong.

Two days later, I see a tweet by one of the people I follow on Twitter, and he mentioned how he gave a friend with no technical background the $200 chatGPT subscription, and within a month, he was already generating $1,000 in monthly recurring revenue. And I was like, what the heck? Here I am spending all this time trying to learn, make apps leveraging my baseline technical knowledge, and I haven't even made close to that much money. It was an eye opening moment because why the hell am I spending all this time if someone like that can do what my yearly goal was in a month? And the popular school of thought today is “if you’re a high agency person, now is the best time to be alive” because you can use AI to further accelerate the “doing” part and conquer the world (the world of B2B SaaS that is - kidding). “If you're an ‘ideas’ guy, now's your time.” “If you have agency, now’s your time”. Well, although nuanced, I think have those things. So, why isn’t it converting?

What I do know is that one of the things I struggle with most, and is probably the biggest blocker to getting things done (i.e. making progress), is my decision paralysis. Part of that is because I love to do extensive research. Or, more accurately put, I love the feeling I get after doing extensive research. I love looking at all possible angles and being exhaustive with my search⁴ and simply understanding things - how they work, why they work - and then being confident in making my final decision. As you can see by that run on sentence, that’s really hard to do when you're trying to move fast. And I want to move fast. I want to make decisions quickly (I think now, more than ever, my future life/career depends on it) and I want to make confident decisions, thus requiring me to make them on my level of complete-ish information… which is so damn hard. And what’s making it unnecessarily even harder? Getting that information from the internet.⁵

For some reason after seeing that tweet, it just clicked. I think this is what I need. I think this could give me information I desire, to get the answers I want, to make the decisions I need, to speed things up.

So, impulsively⁶, I bought it. Right after doing so, I hopped in the shower, and it clicked even more; I could just create a single note in Apple Notes with a bullet point list of all my Q2A’s (questions to answer), quickly jot down all the questions I have throughout the day, and whenever I had a moment, turn that bullet point into a slightly more detailed question, put it into ‘Deep Research’, and just walk away.

Battling the all-powerful maze-game has turned into walking away. What? For real?

From what I’ve experienced so far, yes, for real. I’ve been doing just that.

The responses from Deep Research are different. Not only do those responses give me direction, but they actually give me the accompanying details I want. It goes one giant step further to give me everything that I need to know that feels exhaustive - a report that’s up-to-date, intelligent, and even feels like it's speaking to me. It doesn’t read like it’s some summarized report. It actually feels like someone knows the answers to my questions and also knows what I want to know about my question. That’s hard to describe, but the way that it's written, it just works. You can look at it’s score on Humanity’s Last Exam as numerical evidence (it got 26.6% accuracy), but also just read the responses yourselves and you'll be blown away (unfortunately these responses are too long for me to include screenshots here, but thats also the selling point: they’re expertly-written research papers just for you!)

Expertly-written research papers just for me? Actual answers on most all the questions I have? That’s AGI. That’s AGI for me.

You can argue the goal posts for “AGI” keep moving or that there is no singular definition for what AGI is and what it should do. My definition is answering my questions. Free me from having to use the internet as we know it to be in 2024.

And it’s delivered.

Specifically, for the questions I’ve been asking it about coding, it allows me to succinctly understand what’s going on. Sure, “vibe coding” is cool and will only become more of a thing, but I don’t think it helps you understand. If creativity is rooted in association, you must still be able to make associations. How can you connect the dots if you don’t know what dots there are?

So, yeah, I think this is my AGI moment. I think this is what I've been waiting for; I think this Deep Research product is finally hitting the ‘answer machine’ nail on the head. I can't wait to continue to ask questions and actually just get answers, and then make decisions and move forward. That's what's exciting.

I kind of had this thought for a while now that AI is for the optimizers. And to some extent, I still think that’s largely true, AI is for the optimizers. If you’re an optimizer, you’ll naturally understand the AI hype. But more recently, within the past week, I think AI is simply for people that have questions. Because it can give you answers. And I have a lot of questions. So yeah, I’m excited.