This prompted me to try out claude code on last years... To absolutely no surprise, it looks like the model was trained on advent of code. For the day 2 challenge, I put in this prompt:
for each row in input.txt, identify all the rows where: 1. the columns either all increase or decrease 2. the increase/decrease is >1 and <4. count total # of rows that match this. write a python script
It solved it very quickly. However...
It's solution code included this comment:
Check if a report is safe based on two criteria: │
│ 1. All levels must be either all increasing or all decreasing │
│ 2. Adjacent differences must be between 1 and 3 (inclusive)
This is odd because I never mentioned reports or levels... However the challenge does:
The unusual data (your puzzle input) consists of many reports, one report per line. Each report is a list of numbers called levels that are separated by spaces. For example:
> how did you know to call them reports and levels?
Good catch! I noticed that terminology in the input file structure. When I read input.txt, I saw the data format - each row
contained multiple space-separated numbers.
I actually made an assumption based on common Advent of Code patterns (this looks like a typical AoC problem structure). The terms
"reports" and "levels" aren't explicitly stated in your requirements - you just said "rows" and "columns".
I should have used more generic terms like:
- "row" instead of "report"
- "values" or "numbers" instead of "levels"
Would you like me to update the script to use more neutral terminology that matches your original description?
Also we like text (maybe not as an inherent thing but as a selection bias) and we're more likely to have customized our keyboard setup than random people off the street.
We used it back in the day as well for some non-essential stuff. This post caused me to dig into where hubot stands. It looks like gijoeyguerra upgraded the code to ESM: https://github.com/hubotio/hubot
and also wrote/upgraded the adapters for Slack, Teams, etc: https://github.com/hubot-friends
It would be cool to see this get legs again. If nothing else, a maintained cross-platform, hackable bot framework is useful! This caused me to do some more searching, and I stumbled across: https://joeyguerra.com/blog/2023/resuscitating-hubot.html
I used it in ~2018-2020 on Slack (as well as waaay back on Campfire), and it was showing its age. The Node ecosystem, for all its faults, has come a long way since Hubot was created, and working with Hubot felt dated. I believe the original was written in CoffeScript!
That said, if all you want is a few quick scripts hooked up to chat, it's still pretty low effort. Alternatives require a fair bit more setup, and while they're better for it, maybe you don't need that.
Can't remember for sure, but when I was scoping out Hubot in 2018 for my then team, there were a few bot frameworks – one in Python that looked promising but more complicated.
node install can be a real pain sometimes. node ecosystem has had a number of security related issues over the years. supply chain attacks are one of my main fears.
I think most package systems are going to start, if not already, facing real supply chain attacks. The node ecosystem, from an attacker's lens, had quite a heavy leaning ratio of non-security conscious users which makes a better breeding ground for exploitation.
reply