I just need to Lock In
A disappointing hackathon, learning some data analysis along with dealing with kernel panics. How's your week been?
This is an update I’ve started writing not knowing what to write about.
It’s 6am rn. I just need to lock in.
I find that the way I spend the first two hours of my morning, defines how I spend the rest of the day.
If I’m social with someone in the morning, I feel social all day;
If I do a chore in the morning, I feel like finishing all my chores.
So I’m gonna “lock in” and do things I’ve been procrastinating on early in the morning.
Anyway, here’s what the last few days have been like;
A disappointing CTF
Daniel Gross posted, what I immediately identified as a small CTF on X a few days ago. I came across it just before passing out that night but curiosity got the better of me.
I started with downloading the video. On playing it, I could very audibly hear a hissing noise. That made sense given the first, most definite clue was if “you can hear the music”.
Time to download Audacity.
I ran an ffmpeg command to strip the video from the file;
❯ ffmpeg -i music-1.mp4 -vn -acodec copy music-1.aac
This was my first time using Audacity, so I spent some time playing with it’s different dials and controls.
The first thing I wanted to try out was to isolate the “hiss” and investigate it. I split the stereo audio to mono and then was playing with the different options I had.
I ended up clicking on the “Spectogram” view and I noticed something
I zoomed in, and could very well see the hint;
I started digging around in this spectogram view some more. Zoomed in on different parts of the audio, tried to hypothesize what else could be done.
Even got stuck in a rabbit hole for a loongg time.
I thought maybe there’s some math to it. Some bit manipulation, tweaking the audio file. Isolating the hiss, some place where Planck’s constant would be plugged in to solve it. I think I watched the video like 50+ times easily
After spending half of the following day, I decided to give up. I had no idea where to proceed. No discernable next step. I did know that the number is a hint to the next step but I had no possible sample space to go.
It could be anything at this point.
And probably something I couldn’t see.
If only though;
Imagine my complete and utter disappointment when someone nudged me to lower my IQ and navigate to; https://66260701.com
I felt cheated. It made 0 sense. It hurt my brain.
But I swallowed old hacker pride and carried on.
“Is there anything here?” is what the website said on loading it up. Obviously there was. I loaded up GoBuster and was downloading some fuzzing lists when I realised, the next step is probably “dumber” than what I’m assuming.
I started poking around with dig, nslookup, curl.
Using curl reveals a poem:
"Seek not in words, but in vision's hold,
The painting's artist's name, in colors bold.
A secret whispered through the ages,
Encrypted in cinematic pages.
Through the cinematic lens, a journey beckons,
Past the frames, a clue reckons.
A master's touch, in pigment dressed,
Within his brush, the key rests.
A password hidden, not in text,
But in the art, so don't get vexed!
To find the way, and navigate this lore,
Add a /word this URL, and you shall continue to explore.
So ponder well, the hint conveyed,
In art's embrace, the path is laid.
A quest for eyes, to see and spell,"
Pretty straight-forward. There’s a painting of significance in the move. Identify it’s artist and you’re done.
At 66.26 seconds of the video posted by Daniel Gross, this frame is shown. This is a painting called "Woman with Arms Crossed" by Pablo Picasso. The next step was at “https://66260701/picasso”
Final step, geoguessing;
Just doing a simple Google Image search leads you to the following location: “1389 Bathtub Row Los Alamos, New Mexico” or Oppenheimer’s house during Project Manhattan!
Entering the coordinates solved the last step of the challenge.
State of OpenAI
I love Discourse. It’s used my thousands of places as a forum and for support including OpenAI, and it’s extremely easy to navigate and scrape data from.
Given a forum discussion, here are all the different ways to access it;
Web page: https://community.openai.com/t/need-help-contact-openai-support/106621
Markdown: https://community.openai.com/raw/106621
Paginated JSON by Category: https://community.openai.com/c/api/7.json?page=100
Naturally, I ended up downloading 97,033 posts from the forum as neat little task in dataset curation and data analysis.
Link to the dataset: https://huggingface.co/datasets/julep-ai/openai-community-posts
I also did some sentiment analysis and topic modelling which resulted in a pretty impressive map.
I’m still working on a more detailed blog post about this. Should hopefully get that out soon too!
Good Morning Headache
Yesterday, I started updating my laptop. Mid way through my morning X browsing session, it started to feel jittery. Forgetting that an update was running I force shut it down.
Bad mistake.
The failed system update corrupted my /sbin/init binary which meant that the OS had no idea how to boot itself.
Thus started my 3 hour long ordeal of making a Live USB, booting into it, reading the Arch Wiki.
Except this time, I also had ChatGPT to help me out.
I figured that I still do not trust ChatGPT when it comes to such crucial stuff like this. I’d rather use it for stuff I know would be correct.
But when it asks me to overwrite all conflicting packages forcefully, you know I’m gonna read the Arch Wiki to confirm.
🥂 to Kernel Panics, nothing gets your more sweaty than that.