So, recently in my life I have transitioned to a technical role that involves security stuff.
I quite desire to engage in offensive security for a while, so I thought I'd try my hand at Offensive Security's Pen-200 course and completed the OSCP. After all, it is basically the only penetration testing cert that truly has weight. So I figured it would be my main goal. Then maybe after I will go for a CISSP.
Due to various things occurring in my life, training up has been hard. However this was something I was determined to try for. Nevertheless, I did not have the continous time to study well, and as a result, I failed my first attempt.
That being said, I had fun trying it, and I look forward to my future attempt at it. One aspect that failed me was also my awful notes. My note taking style is thorough but lacks organization and really needs time to put it together in a way that is helpful on a timed exam. This is where I figured LLMs would actually come in handy, as LLMs lack any form of true human style, and the goal is not good prose or humanity, but rather a step by step guidebook for conducting an exam. So I opted to try an LLM for something serious. This is the first time I have truly done so.
I have used it to trim my notes into something that I can actually use on the exam. Basically, it is my main weapon, with my real notes being the backup containing extra knowledge. I am not publishing my notes, for most people they are awful, and basically it's something only I can read.
Regarding LLM use, it is not hard to generate something decent. Though, even using a frontier model, it does require a lot of prodding and a lot of review in order to make sure it is right. I did go through every section manually to make sure it made sense. Still, writing this consumes the entirety of a $20 claude sub for a week. That is not a lot of usage. There is no doubt, the financial side of AI overstates the power. Truly, the finance side of any industry is filled with bullshitters. Honestly, you should stop trusting finance people.
Also it made me kind of realize that while AI companies are subsidizing some of these subscriptions, if one can afford the subscription, then it is a great time to pump out random notebooks that coalesce open information (as they have already scraped that information). And then make them free. AI does not democratize things, but we can use it to make esoteric knowledge more accessible.
That being said, I can see how this technology will bifurcate people into haves and have nots. It can definitely contribute to the existence of a cyberpunk dystopia, and we really need to get into organizing to prevent the further consolidation of capital.
That being said, for those who desire to use LLMs for a lot of things, you can do a lot of good by generating free knowledge runbooks for topics you think would be useful. These do not require the human touch, as they are meant to be dehydrated notes.
I still think LLMs should not be used for real human writing, and this site will continue to never use it for writing or art. And yet, for little shit-work that is nothing but tedium, I see the value. Though I understand there are climate and power consolidation reasons for hating LLMs.
Some people say "what about the LLMs that solve complex math problems, surely we can do more!". Yes, perhaps that happened as LLM companies say they did, but that is the result of operators with intense domain knowledge. Chances are, you are not them. If you are, then wow thanks for spending the time reading this blog. For doing more complex tasks, domain knowledge is king. Instead of trying to build a start-up in a field you might know little about, consider dispersing useful checklists.
So ultimately, I am hoping to learn from my failure, and while I will practice further box-hacking, I would like to publish this runbook for others to use freely.
You can find it in my OSCP Runbook page.
