Latent Space Guardrails That Reduce Hallucinations by 43 Percent Now Open Source
Heyah,
This is Lukasz. I am running Wisent, a representation engineering company. I created guardrails that allow you to block certain patterns of LLM activations on the latent space level. It is now fully self-hosted and open source. Think stopping hallucination, harmful thoughts of the LLM or bad code generation. Let me know how it works for your use case- happy to help you generate the most value from it.
Check out more at https://www.wisent.ai/ or https://www.lukaszbartoszcze.com/
What's a Latent Space again (in LLMs?)
[dead]
[flagged]