Hacker Newsnew | past | comments | ask | show | jobs | submit | rgun's commentslogin

Do we need a 'no means no' campaign for LLMs?

Do what you are asked to ≠ Your duty.

Duty of a warrior is to fight for his country/tribe/side. Duty of a king might be to reduce suffering for his subjects.


Historically, no. It's like Tennyson: Theirs not to make reply, Theirs not to reason why, Theirs but to do and die.

Can you please elaborate on what kind of problems you have faced with uboot?


* John Oliver has entered the room


It will just be a matter of time when the "authority" turns its back on the elites.


I think you have to fill in "additional details" field for it to work. I did it and it worked alright


Oh, I wouldn't have thought to submit the report without filling that in.

The console has: "service error: RpcError"

The POST hangs up before it completes, and it seems to be because the JS can't lift a auth token for my Google account, maybe.


> because in the end it wasn’t the size of the asteroid, the scale of blast, or even its global reach that made dinosaurs extinct – it was where the impact happened

> An impact in the nearby Atlantic or Pacific oceans would have meant much less vaporised rock – including the deadly gypsum. The cloud would have been less dense and sunlight could still have reached the planet’s surface

Makes you wonder what life on earth would be like if the impact happened a few minutes earlier or later.


https://registry.npmjs.org/ is down, affecting our builds


So is https://hub.docker.com which is why I am here and not doing useful work.


This is akin to saying "We are not evil, today"


"we are not evil, we are one default away from being evil!"


Haha, wait, the training directly on websites is off by default, but if you use browser memories, training on that is on by default. Browser memories contain arbitrary amounts of information from websites...

Anil's point is that Chatgpt can't, on its own, train on data that requires a login. So Atlas is in fact automating training on login-walled content, and doing so by default, but also adding decoy settings that make you think it's not.

No evil here!


I don't understand how this helps in improving performance. Can you elaborate?


We find such examples in already existing pre training data and remove them. Do you not think it will work?


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: