Lars Marowsky-Brée 😷 on Nostr: This IBM project amuses me. An LLM checking the prompt/outputs of other LLMs as an ...
This IBM project amuses me.
An LLM checking the prompt/outputs of other LLMs as an additional layer. Ironic, because apparently, LLMs can't be made safe by themselves.
A great way to sell even more hardware.
This project will always be underfunded compared to the big models. And it enables adversarial training.
Quis custodiet ipsos custodes?
Is this how we get true Byzantine Generals?
https://github.com/ibm-granite/granite-guardian
#ibm #LLM #AI #Chatbot #Safety
An LLM checking the prompt/outputs of other LLMs as an additional layer. Ironic, because apparently, LLMs can't be made safe by themselves.
A great way to sell even more hardware.
This project will always be underfunded compared to the big models. And it enables adversarial training.
Quis custodiet ipsos custodes?
Is this how we get true Byzantine Generals?
https://github.com/ibm-granite/granite-guardian
#ibm #LLM #AI #Chatbot #Safety