The most taxing thing about being a journalist, apart from the job insecurity, is all the required reading. Working on a piece means reading through stacks of articles, studies and other reports just to paint a full picture for readers. Even then, you'll invariably miss things that'll immediately be pointed out in the comment section. However, AI can help streamline research and reduce missteps.
As an AI reporter, I'm constantly being sent pitches about the latest AI wares from companies the average person hasn't heard of. Most don't seem particularly useful for the average person, but speaking with Dmitry Shapiro, a former product manager at Google and current CEO of MindStudio, and seeing a video he posted on LinkedIn for his Do Your Research AI agent, made me want to give it a try.
AI is increasingly becoming the go-to tool for reporters, researchers and students. Its ability to synthesize nearly the entire trove of human knowledge and give a bespoke output to any question saves time that would have otherwise been dedicated to cross-referencing material. At the same time, there's a worry that relying too heavily on AI systems atrophies the human mind, making it less capable of problem-solving and critical thinking. Despite this, people, companies and universities are all in on offloading the arduous task of human analysis to neural networks. In some cases, AI can greatly outperform human output, doing 1 billion years of doctoral research in one year.
On the other hand, AI can make grave mistakes, such as telling businesses to break the law. While AI optimists say that the tech will maximize human potential in a fraction of the time, there's worry about the effects of the workforce being supplanted by AI systems and whether society is ready to deal with the potential of mass layoffs. Despite the concerns, AI is here, everyone is using it, and only the most useful tools will survive.
In concept, Do Your Research seems like a godsend for reporting. In practice, it's good overall, but it has some issues that need fixing.
For example, in researching the changes to Twitter's moderation policies after Elon Musk's takeover, Do Your Research did a great job painting a history of all the changes the Tesla CEO made and how it immediately led to an increase in hate speech on the platform. It also highlighted the externalities to Musk's abrupt firings and moderation changes, including an advertiser revolt and an exodus of customers, backed by actual data. The conclusion also gave a position -- I hadn't asked for one, but it shows that the tool can connect all the facts presented.
Per my spot checks, the data that Do Your Research presented was accurate and backed up with correct sources. However, I wish that factoids could be hyperlinked directly to sources, like Wikipedia.
If you're a student, you'll want to be careful with copying and pasting directly from Do Your Research. Upon checking with a plagiarism detection site, Do Your Research's text came up as 20% plagiarized. Unsurprisingly, AI detection tools dinged Do Your Research as 46% AI-generated, which is pretty low considering it's 100% AI-generated.
An example of MindStudio's Do Your Research AI agent looking into Twitter's content moderation.
Screenshot by CNETThe thing I like about Do Your Research over ChatGPT and Gemini is the way it breaks down different points into subheads and tacks on a full list of sources at the bottom. While the other chatbots do this too, Do Your Research lists it out like a detailed bibliography. Given that AI systems can get things wrong, easily being able to go back to the actual source immediately is handy.
Perplexity has a function called Pages that works similarly to Do Your Research. In my tests, the writing read much more like a human and was at a level I'd deem publishable. The sourcing was also well detailed and correctly documented. Granted, Pages was immediately dinged by plagiarism checking tools as more than 93% plagiarized. It's a criticism Perplexity has received in the past. It also explains why Perplexity reads so much better than other AI-generated content.
Do Your Research does need further optimization. A single run of the model can take minutes to compile. In my case, it would fail about 40% of the time. Obviously, these instances are annoying, and it requires more time to rerun the model. After about five successful Do Your Research reports, I ran out of the $5 in token credits allotted to me by MindStudio for my press account. Tokens are essentially the amount of output the AI model can generate before the customer will need to pull out a credit card.
Is Do Your Research worth your time and investment? Yes. It's an incredibly handy tool that does a fantastic job of grabbing various bits of information and collecting it all into an article-like package. The output isn't good enough to be publishable, as its text can read as anodyne and lacking in personality. However, it's a strong jumping-off point to help expand your own research and reporting.
English (US) ·