3 ms·
I'm not saying you necessarily should do it, but if you really want a sexy chatbot like Replika's that you can run on your own hardware and not worry about what
by jhbadger 4y ago
I'm not saying you necessarily should do it, but if you really want a sexy chatbot like Replika's that you can run on your own hardware and not worry about what a company is going to do with the model or what you say to it, you can use a local LLM tool like llama.cpp with an uncensored model like the uncensored version of vicuna.
- EvanAnderson 4y agoI'd argue you should self-host all software that you make any kind of "emotional" commitment to. I won't use so many services and devices because I know the white hot rage I'll feel when it's inevitably taken away from me.
- ashleyn 4y agopygmalion-6B may work better for that specific purpose. however, it may sound less like an actual person and more like a character in a grimey ao3 story. my personal opinion is that anyone using AI for this purpose should treat it more like enhanced fiction and less like it's a human substitute.
- MagicMoonlight 4y ago>just make your own <financial infrastructure/cutting edge neuroscience/road> bro The companies don’t let you get access to the latest research… we get terrible 6b models while they get the best. Trained on our stolen data.
- jhbadger 4y agoEh, the local LLM models (and not just 6B ones but 13B and 30B) out there are actually pretty good these days for chatbots and AI Dungeon-like uses. Maybe not ChatGPT4 quality, but certainly on the level of ChatGPT3 and on normal consumer hardware. Sure, I'm sure there's private cutting edge stuff out there better than this, but most of the benefit isn't that they have better models but that they have large compute farms.