Google has been busy over recent weeks with new Gemini models and features availability. One of these that just hit my Android phones a few days ago is Gemini 1.5 Pro with Deep Research (which I’m going to shorten to Gemini Deep Research for the rest of this post). I had already tried out Gemini Deep Research in its web app form, and I was impressed with it. As is often the case for me, having a phone app version lead me to using it more.
Now that I’ve been using it more, I have to say this is an clear AI Hit for Gemini, a very useful tool. Gemini has definitely had its fair share of misses. I’ve mentioned some of those, and also noted multiple occasions when it did not come out well when I rated its responses on the same task, with the same prompt, compared to those from Claude and ChatGPT. So it’s nice to see the new Deep Research model doing solid work, or better than solid really.
Like many of the latest “reasoning” models do, Gemini Deep Research first responds to a prompt with an outline of its plan to do what the prompt asks it to. It also waits for a tap or click on “Start research” before proceeding. My most recent task for it is this one, warts and all here, with not my best prompt ever:
Side Note: There are better definitions out there for reasoning models, but borrowing from a couple of those highlights some of their notable characteristics:
Take extra time to“think” before answering questions, which greatly improves their ability to solve hard problems
Produce internal reasoning tokens before generating a response
Excel at complex tasks
I’ve seen it work its way through a much larger number of websites with a few previous queries, but it still put in some good work here looking at 52 sites:
In the end it cited 23 of those in its response. All 23 are valid links, no hallucinations and/or 404 errors. The quality of the sites chosen is also notable. They include Deloitte, McKinsey, US Career Institute, Bain, BCG, Tripwire, Harvard Business School, and MIT Sloan Management Review.
Finally, the summary - or report - that Gemini Deep Research puts together from those sources, is pretty great. It could definitely be used to brief a colleague, friend, or team, or as a starting point for discussions on this topic.
Another single tap or click at the end of the response will open it as a Google Doc. This is another plus point for Gemini Deep Research - easy integration with some of its Google siblings. Google Docs is a good one, but I think the best one is with NotebookLM. I’m going to write about that soon. In the meantime, here’s a little teaser with the surprisingly ok-to-use NotebookLM Android web app:
Thanks for this Patrick. Are you paying for Gemini Advanced? This seems to be the only way to access it at the moment, right? I find it a bit frustrating that ChatGPT Deep Research is only available to paying customers (200$). I've used Perplexity Deep Research because I got a year Pro for free through my mobile provider. But interestingly, although it did an in-depth research and looked at 45 sources - it could not (or did not) name any sources on request. :(
https://www.perplexity.ai/search/i-want-to-understand-what-stra-i5bHqtUVSP2QhPMdT_J42Q