[Cyclopath]

... or a clever redditor was pretending to be a chatbot maybe, to entertain us?

[zimmertr]

Yes, that's possible too. I don't believe this is an impossible interaction though. Since it's trained on dialogue and people constantly abuse the AI. It only makes sense it would learn to dish it out too. I assume they try and code around that. But that's not 100% effective as the prevalence of jailbreaking strategies demonstrate.

[Cyclopath]

Remember though that this isn't a toddler, where you say one wrong thing and it'll never say anything else. Some AI systems go through a process of continual learning, like Microsoft's clever bot that learned to be racist by talking to the public. Some go through training with carefully curated data and then stop learning. Again they don't have brains that automatically record every experience they go through; they have software methods to update their model, and may or may not call them at any time.

[Sculpin]

I had just about convinced myself that the posted sinister Bing chat was fake, but then I saw this: https://www.sfgate.com/business/article/microsoft-s-ai-chatbot-is-going-off-the-rails-17790356.php

[Bootpathguy]

[Jumble Jowls]

[Cyclopath]

Microsoft Wants ChatGPT to Control Robots Next A new paper details how we might command robots using simple language. Sounds fun... until they start commanding themselves. https://gizmodo.com/ai-chatgpt-microsoft-control-robots-terminator-1850145030

[Cyclopath]

[Jumble Jowls]

[Cyclopath]

[Cyclopath]

Security A user, known as CodeBlue29, managed to get chat GPT to write malware that bypassed an EDR solution. What is super interesting is that the user (CodeBlue29) didn't have any experience creating malware and was simply using the chatGPT-created malware to evaluate EDR solutions they were evaluating. The output chat GPT created was good enough not only to bypass a well-known solution, but they also submitted it as a bug bounty and received a $650 payout. ChatGPT isn't meant to be able to write malware but can be tricked into it using by asking in different ways, although this is getting harder. https://www.reddit.com/r/cybersecurity/comments/11kzh9u/chat_gpt_got_its_first_bug_bounty/

[Cyclopath]

This is interesting and a little bit surprising.

[Kascadia]

[Cyclopath]

[Cyclopath]

ChatGPT will most likely impact your job if you work in tech, went to college, and make up to $80,000 a year, research says This is like when people invented the calculator, but you can still pay someone to do your taxes.