Original video in Chinese.
Key Takeaway
- Deep Research is AI’s killer app and can significantly improve workplace efficiency. It will spread rapidly in the future.
- The core of Deep Research is the endgame of AI Search, namely AI Research, achieved by mastering more information and controlling more tools.
- OpenAI’s Deep Research performs best, but it is expensive; Google Gemini’s Deep Research performs well and offers good value; Perplexity’s Deep Search is high-volume but only average in effectiveness.
- Through model collaboration and multiple rounds of analysis, Deep Research can quickly complete in-depth research, generate mind maps, and dig deeply.
- This article emphasizes that once Deep Research becomes widespread, it will have a huge impact on the workplace, and recommends that users learn and use it early.
Deep Research is absolutely AI’s killer app. Everyone who has used it says it’s great. Unfortunately, too few people have used it, because the cost is really high. For example, OpenAI charges $200 a month just to let you use it. So this thing hasn’t gone mainstream yet. If you’re a newcomer in the workplace, or really want to move up, take advantage of this period and start using it and learning it right away—no joke.
Put it this way: any boss with a functioning brain who has used Deep Research will absolutely want to get rid of all the entry-level wage slaves in the company. Those who aren’t quite that ruthless will also force everyone to use it. And then anyone who can’t figure it out can get lost.
Let me demonstrate, and you’ll understand. The request is very simple: help me analyze the possibility of Intel and TSMC forming a joint venture, as well as U.S. stock investment opportunities. This has been a hot topic in recent U.S. stock investing, and I’ve already bought in.
After receiving the request, ChatGPT will proactively carry out a round of communication to clarify the details in the request, and then formally get started. The whole process generally takes about 10 minutes. It will search dozens of webpages and then break everything down carefully. At this point, everyone can switch to another webpage or scroll through their phones.
I’m pretty satisfied with this output from o1 Pro. In just a few minutes, it completed the initial research. Just imagine: if you hired an intern for 1,500 yuan a month to help you gather materials and they could do this well, you’d already be laughing all the way to the bank!
Next, you can have this GPT intern generate a mind map based on the previous content, so you’ll get a more holistic understanding. Then, for certain specific points, you can ask GPT to do deeper digging. I won’t demonstrate that here.
So, as workers, you should weigh this carefully yourselves: once Deep Research becomes fully widespread, how should you position yourselves in the workplace?
Hello everyone, welcome to my channel. Humbly speaking, I’m one of the few bloggers in China who can explain the why and how of AI clearly. What I offer is more valuable than tutorials. Remember to hit follow. If you want to connect with me, come to the newtype community. More than 800 friends have already paid to join!
Back to today’s topic: Deep Research.
At present, the main vendors offering Deep Research products on the market are three: OpenAI, Google, and Perplexity. More vendors will definitely join in next. Why? Remember what I’m about to say:
The end of AI Search is AI Research.
So then, what are Search and Research ultimately for? A couple of days ago, I posted this in the community, and this is what I think:
AI will develop along two lines: mastering more information, and controlling more tools.
Among them, search and deep research are key nodes on the first line. That’s why every company takes them so seriously.
OpenAI, which moves the fastest, has already opened this product to Pro users for quite a while. This tier costs $200 a month, and you can use Deep Research 100 times per month. Plus and free users don’t need to worry either; Sam said that Plus users will get 10 uses per month, while free users will get 2.
To be honest, OpenAI’s Deep Research is currently the best-performing product. It doesn’t just search a pile of webpages and bring the content back for analysis. It uses extensive model collaboration. So the Token consumption is far higher than the others. The reports it produces are also several times denser in information than those from other products. If you don’t believe it, go deploy those open-source so-called alternatives and see for yourself.
But unfortunately, I won’t keep subscribing. It’s not a product problem; it’s a downranking problem.
Many people have run into downranking, and it’s especially disgusting. This problem isn’t unsolvable. For example, you can spend a little more money to get a static IP and disguise yourself. It can be done, but the feeling of being discriminated against really makes me unhappy. Are the dollars I pay not dollars?
And aside from these emotions, I judge that this type of product will definitely spread rapidly within three months and become a standard $20 offering.
Thinking about it this way, I’d still make Gemini my primary tool. Good performance, stable, and not obnoxious.
I subscribe to Gemini Advanced for $20 a month, and that includes access to Deep Research. This usage is also limited, but Google hasn’t said exactly how; I guess it adjusts dynamically based on the user’s usage, and you’ll be prompted when you hit the limit.
Let me show you how Gemini performs. Just like ChatGPT, after receiving a request it doesn’t start immediately. Instead, it first does a round of analysis, drafts a plan, and only gets to work after confirming it with the user. Then, within a few minutes, it searches dozens or even hundreds of webpages until it has enough information. Finally, it outputs the result in report format.
If I remember correctly, Gemini’s Deep Research was launched during the 1.5 stage, so it’s still using the 1.5 Pro model and hasn’t been updated to 2.0 yet. So in terms of results, it definitely can’t match OpenAI’s side. But I believe that once Gemini 2.0 is available, the two sides will definitely be on par. And Gemini will most likely offer more quota. First, because of Google’s accumulated strengths in search; second, because Google has TPUs. For OpenAI, cost is their biggest obstacle.
OK, we’ve finished talking about ChatGPT and Gemini. Let’s finally take a look at Perplexity. They’re the most aggressive: they give Pro users 500 uses per day, and free users 5 per day. Although it’s cheap, the results are also fairly average.
To put it bluntly: this isn’t Deep Research, it’s Deep Search.
Anyway, I use Perplexity frequently in my daily life. Now that there’s a Deep Search that’s high-volume and plentiful, there’s no reason not to use it.
Next, everyone will say their product is Deep Research. But how many can truly reach the level of Research? When you use them, remember to distinguish carefully.
OK, that’s all for this episode. If you want to understand AI, come to our newtype community. See you next time!