Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Low-Background Tokens (latent.space)
17 points by swyx on Dec 11, 2023 | hide | past | favorite | 6 comments


How are we running out of them if more are produced everyday? Wikipedia says that radiation in steal has almost returned to the levels it was before the nuclear bomb so that doesn't seem to be an issue anymore.


All the spam skirpts running for the last 20 years are fine for inputs to you models, tho....


unironically fine because they mostly get sent to email which is private by default so not in the datasets


digital grey goo.

this is far worse than dirty metal, not that that’s minor. (it is a fantastic analogy.)


thanks! cant take credit for it, as footnoted Huber, Karpathy and Carmack all made similar observations. however once i heard the concept i had to share it.


Honestly: good.

I also look forward to the efforts to deliberately feed disruptive garbage into the AI training process (I understand some fan fiction communities are already doing that).




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

Search: