Improving response quality and reducing bias
As it is, Lumo doesn't work very well.
It hallucinates facts more often than other modern models that also don't access the web. When asked about privacy-related questions, it almost always promotes Proton's services (including ones that don't exist such as "Proton Chat"). I understand that there is an incentive to promote the company's own services, but most people interested in a privacy-respecting LLM, and especially those willing to pay, don't want it to act as a shill for Proton.
3
votes