41–50 of 108 posts
Classic bookmarks have failed because mnemonic organization doesn't scale. This kind of interface does, and can replace it entirely if done right. Thinking of it, something like this can be used for all your local files as well, acting as a better version of the old filesystem-as-a-database idea. Or for a specific knowledge base (think LLM-powered Zotero).
https://memorycache.ai/developer-blog/2023/11/30/we-have-a-w...
links to https://github.com/misslivirose/Memory-Cache
but did you mean https://github.com/Mozilla-Ocho/Memory-Cache
Earlier quoted context omitted.
I assume that training LLMs locally require high-end hardware. Even running a model requires a decent CPU or, even better, a high end GPU, but it is not so expensive as training a model. And usually you have to use hardware that is available on the cloud, so not much of privacy here.
You don't need to train the model on your data: you can use retrieval augmented generation to add the relevant documents to your prompt at query time.
My usage with browsing is not relevant for this. I don't want to "chat" with my browsin g history. I would simply love my browser would index my bookmarks on my OS so I could search the actual content of those bookmarks. The feedback loop coming gained from chatgtp will I assume always be way better than my local gpt equivalent. But often I bookmark pages where I know the information on there are important enough for…
Very misleading name. The word "Memory" has a distinct meaning in relation to computing, but this is more about human memories.
Earlier quoted context omitted.
Teach it to press the skip ad button
Or it could click "hide" on cookie banners for me!
https://support.mozilla.org/en-US/kb/cookie-banner-reduction
Earlier quoted context omitted.
I assume that training LLMs locally require high-end hardware. Even running a model requires a decent CPU or, even better, a high end GPU, but it is not so expensive as training a model. And usually you have to use hardware that is available on the cloud, so not much of privacy here.
You don't need to train the model on your data: you can use retrieval augmented generation to add the relevant documents to your prompt at query time.
If you're passing in smaller documents then it works pretty good for real-time feedback.
I use the offline translator built into FF regularly and It's magic. I would've never thought something like that can run locally, without a server park worth of hardware thrown at it.
Here's hoping this experiment turns out the same way.