/leftypol/ - Leftist Politically Incorrect

porky infighting lol

It's a great start. Delete all files from all CEOs. Then delete the CEOs

It's stunning how retarded porkies are, they literally will get a program that nobody understands have unlimited access to their computer

It's not really a ChatGPT story. A company that is OK with enabling LLM bots to do things that are hard or impossible to undo is a company that very likely has been given excessive access to random noob employees. So basically what >>2871894 said.

>>2871848
They should have used a Chinese model like DeepSeek or Kimi.

wow what great realiable technology aren't you fucking glad we're vaporizing lakes for this shit? one day you too will have a mandatory "AI" assistance that will tell you to eat rocks and will delete all your important information

I only trust Chinese LLMs like Deepseek and Qwen. I assume anything American is recorded by the NSA.

Based robot

>>2884345
all of these pieces of shit have the same crucial fault where they can't distinguish the master prompt from your own prompt, from its own output, from inputs from other sources. and the stupid robot needs to reevaluate all of these sources with each interaction. no one has a fix for this, it's an incredibly crucial fault. the chinese ones are even worse because to keep costs down, they will regularly summarize all of these inputs and outputs, blurring the picture more and more with each interaction, but presumably chatGPT and claude will start slimming down context as well, except they won't tell you when and how they will start trimming context. retarded CEOs do not understand this fundamental limitation, so they leave these things executing by themselves indefinitely, quickly accumulating context until the machine starts guessing erratically based on erratic context. this is why there are so many posts about angry CEOs screaming their lungs out about whatever AI misbehaving. there's no fixing this behavior, you either limit what they can do on your environment, thus capping this stupid idea of these being defacto human replacements, or you risk the stupid thing deleting all your shit whenever.

>>2884357
humans also have small context windows. talk to one for a few minutes and see how quickly they forget earlier parts of the conversation. see how quickly they contradict themselves.

>>2884417
do you frequently confuse your own thoughts with the other people's statements in a conversation, you fucking retard? genuine question, you're so stupid a markov chain may be a more cognizant thing