14 comments

[ 4.3 ms ] story [ 16.4 ms ] thread
> This is a classic riddiculous and humorous riddiculous riddiculous riddiculous riddiculous riddiculous riddiculous riddiculous riddiculous riddiculous

Bone-chilling.

Remninds me of the texts my brother and I used to send each other back in the earlier days of Android, write one word then spam-click the center word recommendation a bunch and blindly hit send.
Fun fact: On Telegram for macOS, pressing F5 can produce a similar thingy. I have no idea if it's using operating system services or whether it comes with its own thingy, but it feels just as satisfying. You can spam F5-Space-F5-Space over and over and come up with stuff like that I don't know how to make it to the wall and the wall wall is a bit too much for me to be in the office and I can tell you that I have to be in the office by the end of the day I was in the car and I was like oh no I don't want to be in the hospital for a while back and I was just trying to make sure you have a good day at work and I love you so much and I love you so much
Very much hoped that the article itself would be written without the letter e. Disappointed.
Qwn is, I think, an à propos alias.
I wonder if instead of outright masking bad tokens, you could've done the training while slowly increasing negative weights on those
One year in school, we were made to write essays in E-Prime: https://en.wikipedia.org/wiki/E-Prime

If I remember correctly, we would have one pointed deducted for every instance we failed to eliminate.

I suspect this would be nearly trivial to get train (or get an existing model) to do, but, like writing in cursive, one could argue that these activities (as absurd as they may seem…) genuinely and positively contribute to a child's growth and development.

(In fact, I'm actually slowly reteaching myself a proper cursive…)

Flagged incorrect title: 16 gigabytes is "16GB". The article has correct usage, but the OP has misspelled it.
Do you even need to fine tune? What happens if you just pick the next token but exclude options that contain e? Or would that send you down cul-de-sacs? Maybe it would work with diffusion based models...
Could just have set top_not_e.