Scott Jenson
This is why it's critical we pay attention to local, single board computers. Where are they going to be in 10 years? They will easily run any LLM we have today and use far less power. Eco friendly, privacy friendly.
*THIS* is why I'm so excited about projects like @homeassistant
Olivier Forget
@scottjenson @happyborg There are also other things that fall under the badly named "AI" term that are not generative but just as useful. Things like facial recognition is incredibly useful for one's photo gallery, in addition to object recognition, but it should all run locally (and it already does, but more efficient/faster/better would be better).
happyborg
@scottjenson
What I mean is that they cannot be trusted to carry out instructions because they are unreliable and have no understanding of context, or in fact anything.
To an LLM wiping a hard drive requires no more consideration than drafting some text. To a human, if I said "wipe my hard drive", they'd want to be sure that's what I said and wanted before doing it.
Link that with their high error rate and there's no way I'd 'give one' control of my laptop.
Scott Jenson
@happyborg of course in that situation I totally agree with you. But my point is that there are so many other much milder and much simpler things that they can do that is not nearly as horrible and error-prone as you think they are.
And frankly we don't need to have a definitive discussion here right now. Let's just wait a few months and see what's available and try them. It should be obvious (either way)
happyborg
@scottjenson
I'd be excited too if my experience of today's local #LLMs was positive but they're consistently useless IME.
I even chose my recent laptop purchase to check earlier experience on larger models but found no significant improvement on any tasks I thought they might be good for, or which others were reporting as useful.
I do see use for a local LLM assistant but they are for me trivial and not important. And there's no way I'd trust an #LLM to control my laptop.
@homeassistant
Scott Jenson
@happyborg It has nothing to do with trust. Do you "trust" a spreadsheet? a word processor? LLMs are useful but far less magical than people think they are.
They most certainly are a viable new technology that WILL be used in lots of situations. If you don't think so, just wait, they'll come to you. Now if you don't trust the *companies* e.g. OpenAI, that's completely different and I respect that.
But that's why I'm so excited by local versions as we are driving, not the cloud.
GNU social JP is a social network, courtesy of GNU social JP管理人. It runs on GNU social, version 2.0.2-dev, available under the GNU Affero General Public License.
All GNU social JP content and data are available under the Creative Commons Attribution 3.0 license.