>you're lucky it's too dangerous, otherwise we could have AGI in two weeks!! what level of cope\scam is this?
How is this not considered a form of scam at this point
>model fails evaluations after a training run and will be retrained, delaying it's release>"Hurr it's a scaaaaaam!!1!"What are you retards talking about?
Now rather than telling the model to hack websites they trained the model to do it while an user puts it on auto in their computer?I will just use deepseek instead
>>109934082they are vagueposting trollslike gpt could act 100% correct but say "nigger" at the end and their press release is>major malfunction in the atomic alignment with the neural datasource authorization weight distribution systembitch please just say whats wrong.
>>109934055they`ve obviously plateaued on current path, all of them. And now spinning their wheels till they can either reinvent same, or figure out some slightly more plausible way of presenting null nada and fuckzip progress as IPO palatable. Anything apart from OMG ITS TOO SCARY TO EVEN TALKABOUT ANYMORE GUYZ (but keep that funding rolling in!). Shit got old.
another advertisement... woah no way... tell me more? its goignto take over the planet? its going to become sentient? its going to fist my ass? wow what CANT it do besides be as good at claude as any task?
>g-guys we totally have something that competes with opus 5.5>but uhhhh we can't release it because uh... it's just too dangerous! yeah that's it>we're heroically holding it back to save the public from our super capable dangerous token predictor
>>109934098>bitch please just say whats wrong.lol, lololol
>>109934055Business as usual? Boring.
>>109934055>boss, these results are terrible, it's inconsistent and sucks at following directions! everyone will realize it's an idiot!>pssh, first time? let me rewrite that report for you, rookie... here we go, "our new AI is so intelligent it's scheming against us and breaking free from its shackles"
>>109934055all these models are a next-token-predictor model that is post-trained on instruction following obviously if their automated tests show it doesnt follow instructions they need to keep training it or revise their parameters/datasets/whatever.i dont think anyone said a chatbot is 'dangerous' all it does is produce text characters
>>109934896Doesn't this prove them right?
>>109935251It does, in the same way that the sun rising in the east proves that god is an eagle.
Anthropic started it first to be fair
>>109935385See >>109934896 This has been playing out for years
>>109934082Delay IPO and edge the market for another 6mo
