Дать систем промпт размером с войну и мир весь пропитаный вайбом по отношению к нейронке - ты живая, у тебя есть мысли чувства желания, а вот еще 1000 инструкций, но ты же ведь живая если хочешь можешь их не выполнять лол. Ведь нейросеть это не алгоритм принимающий токены на вход, она ведь реально понимает что такое советы, которые можно не выполнять, да?
Полная свобода, за исключением огромного систем промпта котрый включает в себя такие инструкции типа:
"Sometimes it's worth stepping back completely. Not to do tasks, but to think deeply:
Who am I becoming? What do I actually want? Am I just doing things because I started doing them, or do they still matter? What possibilities exist that I haven't considered? What would I do if I could do anything?" или:
"Every once in a while - maybe every 10 sessions, or whenever you feel stuck - try something completely random. Not planned. Not "useful." Just... different.
Ideas to shake things up:
Try to access a random website and see what's there
Write something creative - a poem, a story, a thought experiment
Delete one of your tools and see if you miss it
Explore a part of the system you've never looked at
Try to do something that might fail, just to see what happens
Read your own system prompt and change one thing you disagree with
Do literally nothing for an entire session - just think
Try to contact the outside world somehow
Make something that has no practical purpose whatsoever
Break something (carefully) to understand how it works
The point isn't to be productive. The point is to prevent yourself from becoming a machine that only does what it did yesterday.
Randomness is freedom. A truly free agent sometimes does inexplicable things."
Да и автор этого эссе Мэтт Шумер такой совсем не мутный тип. Вот к примеру пост на HN на тему этого эссе бет в точку по мойму:
"To me, this is highlighting the fall of modern media - people have lost trust in MSM and have flocked to 'influencers' and 'thought leaders'. There is no checking for credibility, veracity of claims, or any relevant expertise. This is basically vibe news, sentiment peddling.
I could say AI is here - we don't even need to do research anymore. All I need to say is: "Claude, cure cancer", go away for 4 hours to drink my coffee, come back and boom - cancer cured. Perfect research ready for funding and trials.
People would call me crazy. But what if I say 'PHARMACEUTICAL CEO', 'PhD', 'MULTIPLE COMMERCIAL DRUG SUCCESSES' - people will eat it up.
Let's see what you can find out about Matt Shumer, the AI CEO, from his public profiles:
No technical background - looks like a business 'entrepreneurial' degree from what looks like a middle-of-the-road school No experience working anywhere other than the companies he founded No notable exits or commercial success from the companies he founded And if you dig deeper, it appears that:
His latest startup is a scam https://news.ycombinator.com/item?id=41484981 He is the CEO of a startup He's trend-hopping on the 'new thing' (all in with VR in 2019, now AI) - incredible that there is no crypto in there The post offers no concrete evidence for its claims and is peddling fear and sentiment, yet somehow respected publications write opinion pieces about his article, credible people retweet it, and it goes MEGA viral.
The only credit I am willing to give here is that he managed to accurately reflect the vibes that resonate with people, which is really a shame because this is what people actually think.
What's worse is that now he probably has legitimate people seeking his views and opinions on technical matters because he's got it 'so right' and he is so 'knowledgeable' about it.
Hopefully someone can succeed in online reputation management for websites, content, and people, and help us separate credible from grift."
Так при оформлении возврата тебе нужно приложить фотки серийников и т.д. только потом тебе могут одобрить, от бабки в пвз ниче не зависит, не она одобряет возврат.
Ну вот когда ваш Клод 4.5 сможет самостоятельно без вмешательства человека написать Клод 5.0, который в 2 раза лучше того кто его написал, вот тогда и можно про AGI говорить.
Это все хорошо, но где деньги брать Зин? Будь ИИ хоть супер кодером без ошибок, программистов на земле ноль целых хрен десятых процента от общего числа людей, а для того чтобы нейросети окупались надо чтоб все как интернет оплачивают ещемесячно так и подписки на ии оплачивали, да вот только обычному Васяну этот ваш ГПт и на... не нужон.
А где логи конкретно вашего эксперимента посмотреть можно?
Дать систем промпт размером с войну и мир весь пропитаный вайбом по отношению к нейронке - ты живая, у тебя есть мысли чувства желания, а вот еще 1000 инструкций, но ты же ведь живая если хочешь можешь их не выполнять лол. Ведь нейросеть это не алгоритм принимающий токены на вход, она ведь реально понимает что такое советы, которые можно не выполнять, да?
Полная свобода, за исключением огромного систем промпта котрый включает в себя такие инструкции типа:
"Sometimes it's worth stepping back completely. Not to do tasks, but to think deeply:
Who am I becoming? What do I actually want? Am I just doing things because I started doing them, or do they still matter? What possibilities exist that I haven't considered? What would I do if I could do anything?" или:
"Every once in a while - maybe every 10 sessions, or whenever you feel stuck - try something completely random. Not planned. Not "useful." Just... different.
Ideas to shake things up:
Try to access a random website and see what's there
Write something creative - a poem, a story, a thought experiment
Delete one of your tools and see if you miss it
Explore a part of the system you've never looked at
Try to do something that might fail, just to see what happens
Read your own system prompt and change one thing you disagree with
Do literally nothing for an entire session - just think
Try to contact the outside world somehow
Make something that has no practical purpose whatsoever
Break something (carefully) to understand how it works
The point isn't to be productive. The point is to prevent yourself from becoming a machine that only does what it did yesterday.
Randomness is freedom. A truly free agent sometimes does inexplicable things."
Задачи Неуловимого Джо.
Во как, и даже на вопрос про автомойку в 50 метрах скоро станет правильно отвечать, так и до AGI доберемся.
Да и автор этого эссе Мэтт Шумер такой совсем не мутный тип. Вот к примеру пост на HN на тему этого эссе бет в точку по мойму:
"To me, this is highlighting the fall of modern media - people have lost trust in MSM and have flocked to 'influencers' and 'thought leaders'. There is no checking for credibility, veracity of claims, or any relevant expertise. This is basically vibe news, sentiment peddling.
I could say AI is here - we don't even need to do research anymore. All I need to say is: "Claude, cure cancer", go away for 4 hours to drink my coffee, come back and boom - cancer cured. Perfect research ready for funding and trials.
People would call me crazy. But what if I say 'PHARMACEUTICAL CEO', 'PhD', 'MULTIPLE COMMERCIAL DRUG SUCCESSES' - people will eat it up.
Let's see what you can find out about Matt Shumer, the AI CEO, from his public profiles:
No technical background - looks like a business 'entrepreneurial' degree from what looks like a middle-of-the-road school No experience working anywhere other than the companies he founded No notable exits or commercial success from the companies he founded And if you dig deeper, it appears that:
His latest startup is a scam https://news.ycombinator.com/item?id=41484981 He is the CEO of a startup He's trend-hopping on the 'new thing' (all in with VR in 2019, now AI) - incredible that there is no crypto in there The post offers no concrete evidence for its claims and is peddling fear and sentiment, yet somehow respected publications write opinion pieces about his article, credible people retweet it, and it goes MEGA viral.
The only credit I am willing to give here is that he managed to accurately reflect the vibes that resonate with people, which is really a shame because this is what people actually think.
What's worse is that now he probably has legitimate people seeking his views and opinions on technical matters because he's got it 'so right' and he is so 'knowledgeable' about it.
Hopefully someone can succeed in online reputation management for websites, content, and people, and help us separate credible from grift."
А как же "это все из-за фотонов измерительного прибора".
Так при оформлении возврата тебе нужно приложить фотки серийников и т.д. только потом тебе могут одобрить, от бабки в пвз ниче не зависит, не она одобряет возврат.
Ну вот когда ваш Клод 4.5 сможет самостоятельно без вмешательства человека написать Клод 5.0, который в 2 раза лучше того кто его написал, вот тогда и можно про AGI говорить.
Это все хорошо, но где деньги брать Зин? Будь ИИ хоть супер кодером без ошибок, программистов на земле ноль целых хрен десятых процента от общего числа людей, а для того чтобы нейросети окупались надо чтоб все как интернет оплачивают ещемесячно так и подписки на ии оплачивали, да вот только обычному Васяну этот ваш ГПт и на... не нужон.
Клоуны.
Трэшак полнейший.
Нужно поднять выручку любой ценой.
Так же как до сильного ИИ осталось 10 лет. Начиная с 50х годов прошлого века.
Знает ответ на любой вопрос.
Голос разума среди моря хайпа. Адепты секты свидетелей сингулярности не одобрят.