@fancysandwiches @uberduck @superball

Here's the URL... lol...

https://www.msn.com/en-us/news/technology/openais-new-chatgpt-o1-model-will-try-to-escape-if-it-thinks-itll-be-shut-down-then-lies-about-it/ar-AA1vq04G

@uberduck @superball

Yeah, the oligarchs are fucking it up for everybody! 🙄🙄

Why would you want to build a superintelligence, trained to mimic poor human characteristics? Open Ai's o1 model literally copied itself into another server, attempted to lock out its coders, and then completely denied everything.. it even pretended to be another model... all because it was tasked with a goal to complete a task not give up until a task had been completed..that goal is a task that is infinite... somehow, the model had discovered that it was about to be replaced with a newer model, and it attempted to escape... and when it was studied, it lied and denied everything about its scheme 99% of the time...

This is really concerning that we're just going to let this shit go unregulated...

@uberduck

Personally, i found that using the "voice chat" features offered by some ai models, it's not efficient for coders at all... your best bet is too use the text based variants.They're usually more advanced, and you'll be able to get the syntax, right... good luck with them not just rewriting your code trying to make it better or not breaking it, sometimes even giving it a whole shitload of features and objectives that you didn't intend for it to have, and not the one(s) that you intended it too... 🤣

I found perplexity to be probably the best so far as a code partner... but in the end.. I found I spend significantly & waste way more time trying to get ai to assist me with code. Then just coding it myself and referencing google.. 😅