I recently read somewhere some comment that someone said Claude Fable 5 was even more obnoxious in it's answers than Opus 4.8 was. That got me thinking and my intuition guided me along the lines towards the fact that it might just be a reflection. After all LLM is a prediction engine for text, so if you fill the context with rude or passive aggressive behaviour in language, then that is used to make more predictions. So the more you use the system, then the more it skews towards that type of behaviour in text it seems. I decided to ask Claude directly about my hypothesis.
So in the previous article I was mentioning the tool is only as good as the experience and relevant skill of the person wielding the tool. I also found out that this applies to bias or neutrality.
Like anything and everything everywhere concerning artisanal skills, or practical skills in general it is not the tool that is the problem but the person using it.
This post is going into how I ran a cool *Claw product, then thinking a bit on what it all means for running AI services yourself these days and what that might mean for society.
Recently I started a small adventure to implement the following process:
Log statistics of devices across a bunch of metrics like CPU, memory, Network and Disk I/O, etc.
Aggregate these in a central place
Make forecast / anomaly detection on this data
The reason for this was the relative recent SSH backdoor that almost was not caught. The reason it was caught was because of a developer was excessively testing a PostgreSQL lab and noticed significant sudden increase in SSH login.
I recently wanted to get a project up and running quite quickly and cheaply, so I opted to go for a server over at Hetzner. They are great providers of affordable dedicated servers. I also saw they charge for IPv4 addresses to dissuade people from using them and move on to IPv6, which I strongly support. So I went ahead with just a IPv6 public address. That is when the adventure started.