Latent Space Guardrails That Reduce Hallucinations by 43 Percent Now Open Source

10 points by warnke 19 hours ago

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/

blatantly 14 hours ago

What's a Latent Space again (in LLMs?)

ongi 18 hours ago

[dead]