Alien Top
  • Communities
  • Create Post
  • heart
    Support Lemmy
  • search
    Search
  • Login
  • Sign Up
ninjasaid13B to LocalLLaMA@poweruser.forumEnglish · 2 years ago

LoRA Fine-tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B

arxiv.org

external-link
message-square
13
link
fedilink
1
external-link

LoRA Fine-tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B

arxiv.org

ninjasaid13B to LocalLLaMA@poweruser.forumEnglish · 2 years ago
message-square
13
link
fedilink
alert-triangle
You must log in or register to comment.
  • Moist_Influence1022B
    link
    fedilink
    English
    arrow-up
    1
    ·
    2 years ago

    “Hey psshhh, AI is Bad and Evil so please regulate the fuck out if it, so we, Big Tech Corps can gain as much power as possible”

  • ambient_temp_xenoB
    link
    fedilink
    English
    arrow-up
    1
    ·
    2 years ago

    Just a handful of miserable, doom-laden short stories killed all positivity bias dead in my amateur tests.

  • a_beautiful_rhindB
    link
    fedilink
    English
    arrow-up
    1
    ·
    2 years ago

    Yea, no shit. I did it to vicuna using proxy logs. The LLM attacks are waaaay more effective once you find the proper string.

    I’d run the now working 4-bit version on more models, it’s just that I tend to boycott censored weights instead.

  • ProperShape5918B
    link
    fedilink
    English
    arrow-up
    1
    ·
    2 years ago

    “Beware he who would deny you access to information, for in his heart, he dreams himself your master.” - Commisioner Pravin Lal

  • FPhamB
    link
    fedilink
    arrow-up
    1
    ·
    2 years ago

    ​

    https://preview.redd.it/1vwlmplq6txb1.jpeg?width=1024&format=pjpg&auto=webp&s=9add9b120b6aa5a378ca79d66b9739883cf48eee

LocalLLaMA@poweruser.forum

localllama@poweruser.forum

Subscribe from Remote Instance

Create a post
You are not logged in. However you can subscribe from another Fediverse account, for example Lemmy or Mastodon. To do this, paste the following into the search field of your instance: !localllama@poweruser.forum

Community to discuss about Llama, the family of large language models created by Meta AI.

Visibility: Public
globe

This community can be federated to other instances and be posted/commented in by their users.

  • 4 users / day
  • 4 users / week
  • 4 users / month
  • 4 users / 6 months
  • 3 local subscribers
  • 11 subscribers
  • 1.03K Posts
  • 5.96K Comments
  • Modlog
  • mods:
  • communick@poweruser.forum
  • BE: 0.19.11
  • Modlog
  • Instances
  • Docs
  • Code
  • join-lemmy.org