As a Computer Engineering major, I don’t write many research papers in my courses. Since it has been quite a while since I wrote a research paper, and I cannot find any of my older ones, I will take a slightly different approach to this assignment. I have come up with a research question related to a course I am currently taking, and I will decide which tool I find more useful if I were to be writing a research paper.

The tools I am exploring for this assignment are Perplexity and Elicit. I will be researching the following topic using both tools:

“How can cache replacement policies be optimized to improve hit rate under different workloads?”

Regardless of your understanding in the topic, I will not discuss the details of the topic but the effectiveness of the AI tools. Also, this is a very well-researched topic and is quite simple, so it should not be too difficult to find good sources using AI.

Elicit

Access my Elicit notebook here.

Access my Elicit research report here.

From a first look at the tool, it seems like a good research tool. It also provides the option to research clinical trials specifically, which makes me think that it may be tailored more for the medical field.

Once I provided just my research question to Elicit, it provided me with a long list of sources, always giving me the option to load more sources. The top sources provided were from IEEE, ACM, and other credible organizations. Since these are organizations which standardize most things in the industry, it was a great indicator of the effectiveness of elicit. I also found it very useful that Elicit provided a lot of data about every source. Not only was were the authors and organizations clearly stated, but a link to the source, the number of citations it has, and a summary were provided for each source. In my experience, AI tools usually provide the link to a source, and maybe the title and authors. All the other features were nice additions as it is exclusively a research tool. You can even access many of the sources directly within Elicit. From my short time exploring Elicit, I can see that the AI chat about the sources always provides citations and is straight to the point. This is different to my usual experience with AI tools as they change what they are saying a lot of the time and provide inconsistent information. Elicit was a very professional tool which had real, academic, credible sources.

Elicit refused to write me a research paper when I asked it directly, however, it did agree to write me a literature review. This does make sense, since it provides sources and summaries so a literature review shouldn’t be too far from that. I found it interesting that it refused to write a research paper, stating “I cannot write a complete research paper as that would require creating substantial new content beyond what’s provided.” I believe this is the most ethical response given that it is a research platform, and writing a paper on its own would make it less credible.

Everything I discussed was only using Elicit’s Finding papers feature. There is also a feature called “Paper chat” which allows you to discuss specific papers with the LLM in more detail. Another feature is “Research report” which provides a detailed research report given a research question. I tried out the research report feature really quickly and was surprised by how good it is. It took a few minutes but it found 50 sources, summarized them, screened them, extracted data, and then generated a report. I was shocked by the quality of the research report. It gave its findings from the research which provided the original sources when clicked on. While I would not directly use this research report, it definitely gives me a good idea of what I would need to research further for my paper, or which provided papers I should read in more detail.

Perplexity

Access my Perplexity chat here.

Since I did not use Perplexity’s Deep Research mode, I did not get the results I would want when conducting research using Perplexity. Rather than assisting with research, Perplexity attempted to answer the research question and provided academic sources as I had explicitly asked it to. I did not think that these were the best sources, in fact one of the links I tried did not even work. However, some of these sources were good, also from credible organizations such as IEEE. While Perplexity did provide sources and summaries, I did not think it would be a great tool when conducting research. When using Elicit it felt a lot more credible as I can see which specific source was used for each specific part and would know exactly what to look for. The version of Perplexity I used functioned just as any LLM such as ChatGPT would.

Reflection

I believe I came across Elicit a few years ago, during RHET 1020. At that point, it was not an AI research tool, but just a scholarly article archive that my Professor had recommended. At that point, ChatGPT had just been released publicly, not even technically available in Egypt yet, and it was nowhere near as powerful as it is today. I believe the AI integration into Elicit might be one of the best integrations of LLMs into a website that did not harm it, but made great improvements. When I took RHET 1020, I think it may have been the first semester after ChatGPT was launched. I admittedly tried using it for my research, and it was very bad. I got fake sources and irrelevant summaries. In comparison to Elicit and Perplexity now, it was useless. These AI tools provided real sources, links, and relevant summaries. Elicit was a much more professional research tool, whereas perplexity, while still a good tool, was not as professional. It seemed more appropriate for quick questions, not academic research. In fact, this was also what it said it was used for, the Deep Research model would be more appropriate for research.

From my short experience with both tools, I would personally use Elicit for research over Perplexity. Elicit seemed like a great research advisor. I will definitely be using it in any future research I do. I think not only would it be great for finding sources and discussing findings, but it could provide feedback on whether my work is supported by the sources I cited, provide more sources, or provide sources with a different perspective. I have not used Perplexity’s Deep Research model, which I am sure would provide a completely different perspective on Perplexity.