Using GenAI for Cybersecurity Research
Searching with a generative AI (GenAI) tool is now my default action when I want to research just about anything in the cybersecurity space. I’m using this option rather than a traditional search engine roughly 90 percent of the time now. I find myself doing this during meetings when there’s an open question or discussion and when I’m searching out details related to projects and tasks. I also get faster answers for simple queries like who owns a domain, what service/application runs on a TCP or UDP port, or what registry key needs to be added or modified to address a misconfiguration vulnerability.
My GenAI tools of choice are Perplexity AI and the Claude-2 large language model within Perplexity. I have tried out, and still use, other chatbots like ChatGPT, Google’s Bard and Pi from Inflection, but over recent weeks the Perplexity and Claude combo has become my go-to. These are some of the reasons that they work well for me:
Perplexity lets me choose which large language model to use - its own, ChatGPT -GPT4, or Claude-2.
Perplexity is “an AI-powered search engine and chatbot that utilizes advanced technologies such as natural language processing (NLP) …”. This is a huge benefit, as (I believe) this is what provides the capability for responses from the models to include citations - a listing of the source links they used. When weighing up the quality of a model’s response, citations are extremely useful.
My experience has been that, despite ChatGPT being more famous, Claude’s responses are more detailed. Here’s a good recent example, when a new version of MITRE ATT&CK was announced recently and one of its features that caught my eye was some new, expanded detection analytics capabilities - and I wanted to know if or how these could be incorporated in threat hunting with some Microsoft tools:
In Perplexity, I can create Collections. All of my queries are saved until I choose to delete them, but being able to group them is very handy.
There are Perplexity Android and iOS apps, and I use the Android app as much or perhaps even a little more than I use the desktop version.
And there’s an (easy to find) option to not allow my queries to be used to educate the models within Perplexity.
I would love to hear about any of these tools you are using and any suggestions you have for getting more out of them.
Great blog! Thanks for mentioning Perplexity AI - I stumbled across this last week. Do you think the answers you get from the gpt-4 engine is the same as the native gpt-4 site? If so, seems like super-good value as well! I really like Claude for coding/syntax, but haven't gone for a subscription just yet.