Google has upgraded its chatbot platform, Gemini, with the ability to “reason” about research questions and create comprehensive reports.
The updated Gemini now offers a feature called “Deep Research,” which Google says uses “advanced inference” and “long context features” to generate research summaries. The summary appears in the Gemini app and can be exported to Google Docs for further editing.
For now, Deep Research is exclusive to Gemini Advanced, the more sophisticated version of Gemini behind the $20/month Google One AI premium plan.
dive deep
Google says Deep Research acts as a kind of research assistant, able to analyze information related to queries from across the web on your behalf. The results are organized into summaries within the overview and combined with links to the original source material.
Here's how it works:
User writes a question. Deep Research creates a “multi-step research plan” for you to modify or approve. Once you accept, Deep Research will spend several minutes refining your analysis. That is, search for interesting information, save it, and start a new search based on that information.
I learned. This process is repeated multiple times, and once completed, Deep Research generates a report of key findings.
Deep Research is only available in English on desktop and mobile web. In the model drop-down menu, the user can select[Gemini 1.5 Pro with Deep Research]You can access it by selecting the option. Google has announced that it will be coming to the Gemini mobile app in early 2025.
Google's new Deep Research feature feels like one of the most aptly “Google-like” uses of AI to date, and it's pretty impressive.
I've visited it a bit and it provides very good initial reports on almost every topic. Paywalls around academic sources impose some restrictions. pic.twitter.com/dwSqr6aKGZ
— Ethan Mollick (@emollick) December 11, 2024
“A new agent system leverages Google's expertise in finding relevant information on the web to direct your browsing and research in Gemini,” Gemini app product director David Citron said in a blog post provided to TechCrunch. “We built it,” he said. “Detailed research can save you hours of time.”
But this feature raises all sorts of thorny ethical questions.
potential harm
Aside from the fact that all AI makes mistakes and hallucinations (anyone remember Glue Pizza?), technologies like deep research could have a profound impact on education.
In a recent op-ed in the New York Times, Jessica Gross wrote about how students are increasingly relying on generative AI to outsource their brainstorming and writing. These students are at risk of losing their ability to think critically and overcome frustration with assignments that are not easy to complete, she said.
At least one study has linked frequent use of ChatGPT among students to procrastination, poor memory, and lower grade point averages.
Deep Research also threatens to harm the source, the publisher, in a financial sense. Deep research collects information from websites and compiles them into summaries, potentially depriving these sites of valuable advertising revenue.
The impact of AI Overview, the AI-generated summaries provided by Google for specific Google search queries, on publishers could be an indication of what will happen when deep research gets into full swing.
One source said publishers have seen a 5% to 10% drop in traffic from search since AI Overviews launched earlier this year. On the revenue side, experts cited by the New York Post estimate that AI-generated summaries could cost publishers more than $2 billion.
Google has often said it works closely with its publisher partners to respect paywalls and allow websites to block AI scraping at the domain level. But outlets often face the dilemma of allowing scraping or losing visibility in Google search (at least for now).
Google claims that Deep Research “allows users to dig deeper and learn more by connecting them to relevant websites they wouldn't otherwise find.” We'll have to see if this feature delivers on its promise of discoverability, or if it distracts views from the broader web.
gemini 2.0 flash
Deep Research isn't the only new feature coming to Gemini.
Starting today, free and paid Gemini users will have access to Gemini 2.0 Flash, Google's latest flagship AI model. To be precise, this is an experimental version of 2.0 Flash optimized for chat, with the full version expected to be released in January.
Google says 2.0 Flash can be selected from the Gemini model dropdown on desktop and mobile web (no mobile app yet). This should result in better performance and faster response across many tasks.
Still, the company warns that some Gemini features are “incompatible.” [the] This is an experimental model. ” There is no mention of specific characteristics.