While I Have Been Away

So, that is where I am at now. Brace on my knee, trying to heal up, and hoping to get back out to the library so I can get back into the swing of things.

Just because I have been away has not stopped me from reading about the various subjects that interest me. And if you have read any of my posts, you know that AI is definitely one of those subjects.

I use AI in a number of projects I have been working on, and I also try to keep up with news stories about it whenever I run across them.

So the first thing that jumped out at me was a recent article about people getting around AI guardrails by changing how commands or requests were worded.. Meet the AI jailbreakers: ‘I see the worst things humanity has produced’

Actually, when I tried to find this article again, I realized this has been a constant security issue with AI for the last couple of years.

“In hindsight, using natural language to trick these machines was inevitable. Large language models such as ChatGPT are trained on hundreds of billions of words – many of them dredged from the internet’s cesspits – to learn the basic patterns of human communication. Without safety filters, the outputs of these models can be chaotic and easily exploited for dangerous purposes. The AI firms spend billions of dollars on “post-training” to make them usable, including constantly evolving “safety” and “alignment” systems that try to prevent the bot from telling you how to harm yourself or others. But because the AIs are trained on our words, they can be fooled in much the same way that we can.”

That is just one of the methods used to trick AI into providing information that the guardrails were set up to protect against. Trust me, there are many others.

What I will say is that I consider myself pretty much a power user of AI, and I never thought of trying to do anything like that. Maybe that shows how naive I am, or maybe it just shows that there is a difference between using a tool and trying to break it.

The second article that I seen is the complete opposite of what I just described. OpenAI CEO Apologizes For Not Alerting Canadian Authorities About Warning Signs of Mass Shooter

“Warning signs were seen months earlier, though, when OpenAI banned her ChatGPT account the prior June for violating policies related to violent activities. Prompts put into the chatbot by Van Rootselaar created internal debates with OpenAI staff about whether the police should be notified about the person behind the account, but ultimately decided not to, according to a report from the Wall Street Journal.”

To me, this shows that AI guardrails do work a lot of the time. In this case, they appear to have worked. The system flagged behavior serious enough that the account was banned, and there was apparently internal discussion about whether law enforcement should be notified.

The problem was not that nothing was detected. The problem was that nothing was done beyond the internal action.

If I had to guess, when guardrails are not followed, there are probably automated levels of monitoring that begin. If certain thresholds are crossed, notifications may be sent for human review. That is only a guess on my part, but it would make sense.

Now, I know there will be a segment of society that considers this kind of monitoring an invasion of privacy. To a small extent, I understand that concern. But in a case like this, where violent warning signs were allegedly serious enough to cause internal concern, I have a hard time arguing against some level of review.

At the end of the day, people are using a company’s tool. For legal, ethical, and safety reasons, that company has to monitor how the tool is being used, especially when the usage appears to cross into threats, violence, or harm.

In neither case is AI the evil part. The problem is how people choose to use it.

Now, the next 2 topics I would like to cover is a topic where I actually wrote some articles on. This one bit me in the ass right after I wrote the last one. The Fragile Electronic World We Live In

Sigh. Right after I wrote and posted the article on Fragile Electronic World We Live In, the database server for this blog crashed and corrupted everything.

I could not access anything, and anyone who tried to read my previous articles from Facebook or Twitter could not see them either.

I will be honest. I looked into what it would take for me to recover from it, and I was ready to can the blog or accept that I would lose what I had already written and just start from scratch.

I have a good amount of experience with different areas of tech, but this was one area where I did not feel qualified to tackle. I could also see it taking way more time than I was willing to spend, especially while dealing with my knee.

Since the crash happened right after my accident, I was already in an “I don’t give a shit” mood. So yes, my attitude was pretty close to giving up.

Which leads to the positive ending of this post: excellent customer service.

I host this blog and website through a small web hosting company that I have been with for years, going all the way back to when I was into IRC. You do not get all the bells and whistles you see advertised with the larger web hosting services. But what you do get is honest pricing for what is provided and excellent technical support.

As I previously mentioned, I have a wide range of technical experience, and I can do most things on my own without the bells and whistles. I also think the bells and whistles can hide things that cause problems. When you do things old school, you have a better idea of whether things were installed correctly or not.

I think Mark from Lomag, my host, appreciates that I try to research and correct things on my own. If I cannot fix something, I at least try to gather enough information so he can jump right to the problem without digging through everything.

Point being, Mark had my blog back up and running in a matter of hours. He would not let me give up and just trash the blog.

I want to send out a sincere thank you to him for getting me back up and running so quickly.

To wrap things up, I cannot say for sure if I will be posting regularly like I was before. I prefer being at the library when I work on things like this, and right now that is hard for me to do.

I may be moving slower for a while, but I will not be gone for weeks at a time.

Thanks for your patience.