- cross-posted to:
- hackernews@lemmy.smeargle.fans
- cross-posted to:
- hackernews@lemmy.smeargle.fans
You must log in or # to comment.
OSS, local llm, SearXNG. I likey, is there a demo ? SearXNG via VPN has helped unshittifying my search, but GIGO still applies.
Dumb question, why do you need VPN to use SearxNG?
You don’t, I like it coz it minimizes profiling for the component search engines, and gluetun is right there, just point SearxNG at the proxy. I still get reasonable localized results by chosing a nearby exit node.
deleted by creator
Ok, now I understand what OP meant.
However, I use my own SearxNG instance, so I guess I never thought about it that way.
Google using ai, everyone hates it
Some upstart uses ai search, everyone is like woooowowow???
LLMs are not necessarily evil. This project seems to be free and open source, and it allows you to run everything locally. Obviously this doesn’t solve everything (e.g., the environmental impact of training, systemic bias learned from datasets, usually the weights themselves are derived from questionably collected datasets), but it seems like it’s worth keeping an eye on.
Google using ai, everyone hates it
Because Google has a long history of doing the worst shit imaginable with technology immediately. Google (and other corporations) must be viewed with extra suspicion compared to any other group or individual because they are known to be the worst and most likely people to abuse technology.
Literally if Google does literally anything, it sucks by default and it’s going to take a lot more proof to convince me otherwise for a given Google product. Same goes for Meta, Apple, and any other corporations.
The main complaints towards Google was LLM maturity, bias and other factors. The same things will be true for any LLM
Better not tell perplexity about this.
So this is better huh?
I haven’t tried it, but having tried Perplexity, I can say that it’s difficult to have something that’s worse than it!
Has anyone used this?
I’ve used it, it’s pretty rough and unfinished, the current main branch doesn’t build without help and you’ll need ollama or openai keys.
The results however are impressive, even with a small model like phi3 mini through ollama. They got some good prompts behind it and the results name the sources + have some good followup questions.
I haven’t no
You mean chatgpt or real ai?
It can use ChatGPT I believe, or you could use a local GPT or several other LLM architectures.
GPTs are trained by “trying to fill in the next word”, or more simply could be described as a “spicy autocomplete”, whereas BERTs try to “fill in the blanks”. So it might be worth looking into other LLM architectures if you’re not in the market for an autocomplete.
Personally, I’m going to look into this. Also it would furnish a good excuse to learn about Docker and how SearXNG works.
Super nice!
Seems broken, couldn’t get the yarn to build. I’ll try again another day
