Real Time Access

[ghost]

Can somebody explain why real time internet access is not planned? Some models only go to 2022, this data is too old for many current purposes. Lots of new data is available and needed but just cant be used because the models dont have the data. Very inconvenient.

[enema-combatant]

Can't we re-train on newer bodies of data? In effect, we're doing RAG on LocalDocs, right? Why can't we RAG in some stuff via API call, and freshen the various domains via vetted sources for training. Not sure, I'm no expert at all, but maybe this is the feature you're talking about?

[ghost]

i'm not a programmer so using API calls not possible. I mean i can just stick to using chatgpt or copilot, they have real time internet access. i just read somewhere else here that topics like mine are closed because its not planned.

[enema-combatant]

The use case (or principle design idea) here is 'offline' access. Think: disconnected from all internet. This is a 'preppers' or 'shit hits the fan' gold mine, honestly. Between Kiwix/ZIMs (which can also directly updated from live websites and save those periodically, on a schedule) and GPT4ALL, you can have an effective 'dead mans library of Alexandria'. Heck, you could also POINT GPT4ALL at those ZIMs and update it from there (ZIM->PDF conversion necessary). I mean, the world is your oyster, no?

[ghost]

i understand what you are saying, but if i must source data for gpt4all to use i am defeating the purpose of my use of ai to do my work in the first place. no, i think i will stick to the ones that can scrape data in real time. cheers.