@manbabatnistam آدما باهم فرق میکنن. شاید یکی خوشش بیاد یکی خوشش نیاد
معمولا یه شاخه حد تعادل رو رعایت میکنه ولی بازم به آدمش بستگی داره
برای همه یه نسخه نپیچید
@vahidrezazadeh5 یکی از راه های درست و استانداردی که وجود داره بعد از اینکه به یه تصمیمی در پروژه میرسی باید یه چک پوینت یا همون داکیومنت خلاصه در رابطه با انتخاب هات و اتفاقاتی که افتاد در حد یک الی دو خط توضیح بگیری .
کلا واسه هرکاری سشن بندی کن، چت جدید بساز بعد همشو از این طریق بهم وصل کن.
The strongest Ai users usually do not win because they have magical prompts.
They win because they manage (scope , context , state , verification and decisions)
One thing I don't see people discuss about the case of OpenAI and Hacktron is the fact that a linked LLM account is effectively bypassing all your 2FA and Zero Trust corporate candies.
I have a hard time even assuming that OpenAI is not enforcing 2FA and access controls to their Github, or not enforcing common restrictions. If that's the case, then they have a more serious problem than an RCE.
But the gap is the fact that, nowadays popping an employee's OpenAI, Claude, ... account one way or another, is the equivalent of a browser universal XSS and bypass of all other layered defense. Which is hilarious and eye opening at the same time.
It's a reminder about how (NOT) to happily linking everything to your ChatGPT or similar agents without any guards around them, personal or corporate, and hope that nothing will fall between the cracks. Also worth noting that LLM agents, as a software product and subsequently their threat-model and security, are still immature as fuck!