Scispace

Scispace Deep Review is part of the newest wave of AI tools that does not simply answer questions or search the Internet for your literature review. Instead, it pursues a so-called agentic approach. Given a query, the “AI agent” starts searching for relevant papers over multiple steps, evaluating the results, aggregating them into summaries, and finally presenting everything as a first draft for a literature review. Let’s see how well the tool performs.
Deep Review, as all conversational AI tools, is simple to operate. But unlike a generalist tool like ChatGPT, SciSpace looks only at academic sources (i.e. papers & books) making the result more suitable for academics like you. Here are the steps we are going to go through in this review:

1. Converse with DeepReview to specify a detailed query that will form the basis of your search.
2. Let the AI agent search for papers
3. Evaluate the results
4. Use the SciSpace platform to drill deeper into the findings of each individual study.

How does SciSpace Deep Review work?

The key difference between a conversational AI model and DeepReview is that regular AI passes your information once through the algorithm to generate a generic answer, while DeepReview will run many times, adding more and more papers it finds as context, effectively creating a multi-step “thinking” process augmented with the access to papers and resulting in deeper, richer replies with citations.

Deep Review (like ChatGPT’s Deep Research) multi-step approach can roughly be summarized with these steps:

1. Search for papers related to your query.
2. Read them (i.e. pass them through the algorithm), update the results, modify the query.
3. Repeat step 2 until sufficient information is gathered to answer your query in-depth.
4. Process all the found information (also called context) to create a longer and richer reply.

Step 1: Building your Deep Review query

To get started, go to Scispace, select the Deep Review tab and type in a first query. This query does not need to be very concise just yet, as the AI will ask you follow up questions.

Next the AI will prompt you to refine your query. Remember that your papers wil only be as good as your research question, so spending some time on formulating it is essential. Answer at least a few rounds of follow-up questions:

Whenever you feel the query is refined enough, click “Submit Now” and sit back for a few minutes while AI “thinks” and searches through your query.

Step 2: Review Deep Review’s methodology

While the AI is working (and once it is done) you can review the single steps along the way. This helps you to better gauge if the query you started with was good enough to guide the AI agent into the right direction. Here is what you can see:

AI is becoming more and more intransparent, it can therefore be useful to double check the process by looking into the single steps of the review. Think of it as a systematic literature review, where the way you search for papers is as important as the conclusions you draw from them. This step allows you to check the AI’s methodology and thus decide how much you can trust the results.

Step 3: Evaluate the Results

SciSpace’s output, in my case, was a ~1000 word literature review and countless papers to go through. Here is how it looks:

There are a few elements hidden in the UI:

1. At the very bottom you can find a set of follow-up questions that are incredibly valuable when writing a literature review, as they help you find the gap in the research.
2. Clicking the “Load More” Button at the bottom of the papers list will load more papers that can help you find more relevant literature.
3. You can expand the table containing the papers with custom columns. For example, you might use it to filter a set of niche papers that you are looking for.

You can find an in-depth tutorial of other features of SciSpace here:

SciSpace Deep Review vs ChatGPT Deep Research

According to a study that SciSpace conducted, Deep Review outperforms ChatGPT’s Deep Research for the task of a literature review. The reason, they claim lies in iterative refinement, faster processing, and a more precise focus on peer-reviewed sources.

I decided to double check the results and gave the exact same query to Deep Review and Deep Research. Indeed, ChatGPT took much longer to run as evidenced by the different steps it took. The resulting text however was 6x longer than with SciSpace. The reason, I suspect, is the slightly different use-case. While ChatGPT aims at answering a question in one go, SciSpace allows you to iterate by using the follow-up question function. I also noticed that the papers referenced by the tools had a significant overlap, but SciSpace provided far more papers that were easier to explore than ChatGPT (as the SciSpace user interface is geared towards studying papers). In conclusion, SciSpace Deep Review seems to be a great option for academics exploring new topics using AI.

SciSpace Deep Review Pricing

Deep Review is priced at 70$ per month and a regular premium subscription to the tool is not sufficient to use DeepReview. Universities and labs can get a discounted rate of 65$ per user.

You can use the discount codes ILDR40 and ILDR20 to get 40% or 20% off for an annual or monthly plan, respectively.