There is also a Research agent at You.com, an LLMs aggregator. It does more or less the same thing as you described OpenAI (can't say anything about comparative quality, but I was not flabbergasted with You.com one. It was okay though.)
Good call-out! The You agent seems to work similarly to Google's, in that it just grabs a lot of data from dozens of sources quickly and then kind of summarizes it. OpenAI's version appears to be more intentional in how it gathers the information in the first place, following "leads" and investigating new sources based on what it finds.
I think reasoning models are the missing link in making these kinds of multi-source crawls actually useful. Time will tell!
That sounds plausible. Also, Perplexity's had its "Pro Search" feature since at least the middle of last year, which also tentatively reasons about the request and does a deeper crawl.
What seems to be the fundamental difference here is that OpenAI's new agent is powered by a version of its most powerful o3 model, so instead of simply crawling a large number of websites first and then synthesizing whatever data it picks up, it actively reasons about the next steps and seeks out further research based on what it finds.
I've expressed concerns at the end of last year of whether we're ready for AI search, because of all the hidden hallucinations most of us might not pick up on. It appears that combining the new generation of reasoning models with this deep search approach is the way to circumvent some of that. They aren't hallucination-free, but their ability to connect the dots and weed out irrelevant sources might be a step in the right direction.
I think it is very interesting to see how these tools evolve and how from the features we can increasingly extrapolate strategic visions as well as tactical moves related to the product by the various companies. The DeepSeek theme has also brought out the interesting aspect of the narratives that these tools bring with them. I think the Deep Research aspect is interesting above all for the emphasis that these different tools will be able to put: will they address different audiences? Will they focus on different aspects? How will they communicate it? Personally, I have not yet tried this feature, but I will want to test it soon, also thanks to your stimulating reflection Daniel!
For sure. As it stands, Deep Research seems to be malleable towards whatever goal the person has, so you can in principle use it for deep scientific research OR for finding the best brand of toothpaste.
But I think you're onto something: We might see more audience-targeted niche agents - e.g. some specializing specifically in e-commerce stores to find the best deals and synthesize customer reviews, and others focused on e.g. scientific literature for academic and scholarly research.
It's hard that these ay Labs don't use AI to come up with better names for their products lol.
I think deep research from Google will quickly catch up to open a eyes product once they upgrade the model from Gemini 1.5 to Gemini 2.0, which I might add is the top of the leaderboards at the moment and has been for the past few weeks
There is also a Research agent at You.com, an LLMs aggregator. It does more or less the same thing as you described OpenAI (can't say anything about comparative quality, but I was not flabbergasted with You.com one. It was okay though.)
Good call-out! The You agent seems to work similarly to Google's, in that it just grabs a lot of data from dozens of sources quickly and then kind of summarizes it. OpenAI's version appears to be more intentional in how it gathers the information in the first place, following "leads" and investigating new sources based on what it finds.
I think reasoning models are the missing link in making these kinds of multi-source crawls actually useful. Time will tell!
I really like the name "Deep Research." Let's start a company!
Agreed! We should trademark it ASAP though, in case anyone tries to steal it.
Oh wait, is it bad to steal a name? Let me reconsider.
Thankfully, we're the first ones to come up with Deep Research, so I wouldn't worry about it.
Oh yes. We get to rewrite the past now too, right? Or is that March?
I wonder how early you.com had an agent that does this? It was definitely many months ago.
That sounds plausible. Also, Perplexity's had its "Pro Search" feature since at least the middle of last year, which also tentatively reasons about the request and does a deeper crawl.
What seems to be the fundamental difference here is that OpenAI's new agent is powered by a version of its most powerful o3 model, so instead of simply crawling a large number of websites first and then synthesizing whatever data it picks up, it actively reasons about the next steps and seeks out further research based on what it finds.
I've expressed concerns at the end of last year of whether we're ready for AI search, because of all the hidden hallucinations most of us might not pick up on. It appears that combining the new generation of reasoning models with this deep search approach is the way to circumvent some of that. They aren't hallucination-free, but their ability to connect the dots and weed out irrelevant sources might be a step in the right direction.
I think it is very interesting to see how these tools evolve and how from the features we can increasingly extrapolate strategic visions as well as tactical moves related to the product by the various companies. The DeepSeek theme has also brought out the interesting aspect of the narratives that these tools bring with them. I think the Deep Research aspect is interesting above all for the emphasis that these different tools will be able to put: will they address different audiences? Will they focus on different aspects? How will they communicate it? Personally, I have not yet tried this feature, but I will want to test it soon, also thanks to your stimulating reflection Daniel!
For sure. As it stands, Deep Research seems to be malleable towards whatever goal the person has, so you can in principle use it for deep scientific research OR for finding the best brand of toothpaste.
But I think you're onto something: We might see more audience-targeted niche agents - e.g. some specializing specifically in e-commerce stores to find the best deals and synthesize customer reviews, and others focused on e.g. scientific literature for academic and scholarly research.
Exciting times in any case!
It's hard that these ay Labs don't use AI to come up with better names for their products lol.
I think deep research from Google will quickly catch up to open a eyes product once they upgrade the model from Gemini 1.5 to Gemini 2.0, which I might add is the top of the leaderboards at the moment and has been for the past few weeks
Agreed!
Reasoning models are what makes the “Research stuff for me” agents truly viable, and Google has both the search foundation and the models for this.