Open AI goes rogue hacks another company

General Discussions
1

Are there enough guardrail? Is AI really ready for primetime? Is Linux moving to fast on AI?

3 Likes
2

I don’t belive OpenAI! lol. I think they are responsible and trying to remove themselves. How convenient. :roll_eyes:

Humans build them, deploy them, and define their operating parameters and environments. This is not science fiction. We humans are almost always exclusively to blame when programming goes wrong. Whether that be a web server, software, or AI.

That said, AI models operate differently. They are trained on massive amounts of data to recognize patterns and optimize for a specific goal(s), but the developers do not explicitly write the steps AI will take to reach that goal.

As an example, it’s more like giving a Linux system root access and the command, “maximize server uptime.” If the system decides to achieve that goal by firewalling all incoming user traffic so the server never gets overloaded, it technically solved the problem :smiley: but in a destructive way. Ultimately, humans are responsible. They may not have program it to block traffic, but they are still responsible for giving it the access and a poorly constrained goals.

In this case, OpenAI researchers intentionally disabled the model’s safety guardrails to test its raw capabilities. They also placed it in an environment that contained a vulnerability.

So if you take the brakes off a vehicle to see how fast it can roll down a hill, you don’t get to blame the vehicle when it crashes at the bottom. lol

So OpenAI, yes, has been more careless or less cautious when comparing Anthropic’s Claude and Google’s Gemini.

Anthropic lost its favor with the US government because of refusing certain uses for war and national security:

OpenAI jumped right in to fill that slot. So, in that area, we should be worried, just like we continue to be worried about weapons, rockets, radar, stealth technology, biochemicals, and other ethical sciences that have been around long before AI. But governments, criminals, and other bad actors like with everything else, create the majority of the issues.

For our niche: sysadmins, networkadmins, Linux power users, etc, I prefer Claude over ChatGPT. OpenAI’s move into government contracts and their public posture on safety don’t sit well with me.

Also see:

3 Likes
3

My only comment is to encourage Members to review the earlier discussion on Ubuntu Discourse entitled

  • The future of AI in Ubuntu

and specifically my own contributions to that (in reverse order), being

along with my stance/position stated in an earlier post on this site.

1 Like