The demo says knowledge bases can be created by anybody.
Yeah, if you’re an expert in X field you can probably do it alone just feeding relevant papers, expecting a good knowledgebase from 1 person is probably unrealistic tho, especially as in science there are competing theories and you should not blindly ‘trust science’ as was recently pushed by all mainstream media about one specific subject
1 Like
“f you’re an expert in X field you can probably do it alone”
I think that’s the idea. I have a lot of private knowledge, where I’m the expert and nobody else knows about it. I would expect to be able to use it for documenting privately developed software for example.
Imagine if you had only one knowledgebase about gravity, dark matter is a fact, but load mond enthusiasts kb and suddenly they have better predictions about galaxies? It’s not all black and white, nature is messy. Now will lobby groups try to censor kb’s from opposing side… Again details of all that are hidden for us
I don’t know what you’re talking about, to be honest. In my understanding knowledge bases are supposed to be able to be hosted locally. If that’s the case then it’s a matter of documentation.
I think the VCs will be the disappointed ones, stochastic parrot is as far from AI as it gets, just like Elisa was amazing, once people realise it just lies 90% of the time and PR hype dies down… If there are any useful applications, still yet to see
Yeah you have 1 KB bases on the bestest and greatest physicists of our time, it says dark matter real, then you have another KB based on MOND theorists (they can actually predict rotation curves better than DM guys that will just adjust numbers after the fact), explain how documentation fixes that, you can host both locally
It seems you have unreasonable expectation and in turn are of course disappointed.
I don’t expect very much. If I can develop applications which make use of TTS and STT and which can access an LLM with a vector db, it think it’s good. I looks like that should be possible, although at a slow speed with the mind 2 hardware.
Documentation for how to set up the local infrastructure fixes my dependency on remote APIs.
I don’t care too much about chat models with knowledge about the whole world.
You can already do that on device even with whisper, though without swapping only tiny/small models, maybe with 8gb ram medium comes into play. Not sure what you expect from KBs? Science is never settled, so if you get proponents of one theory to build KB of course it will contradict KB built by guys who support different one, there is no objective truth lol
Every knowledge base, if the knowledge is developing must be curated and every source can be biased - it’s not a technical problem.
Edit: What I ideally expect from a knowledge base is that I can ask questions by voice about facts I entered and get audio responses.
Is it developing when the current mainstream knowledge blacklists people trying to upend it, or is it stagnating and keeping the status quo, will you load the einstein KB when you have aether KB already and all MSM says aether real?
I’m not getting my news or relevant political/historical/societal facts from next-word-predictor AI (e.g. ChatGPT), which as we all should know can hallucinate and are also heavily biased towards western hegemony and in other ways IMO.
What you’re talking about is not a problem of AI in general. It’s a matter of media competence.
No, it’s a problem of just getting facts from pre-approved source, where ‘experts’ vow to be right, you will not get any new knowledge with aether KB, why even query it lol
1 Like
Yeah, that’s why i don’t. 
Edit: You mentioned whisper. If i’m not mistaken, this is only speech recognition and therefore only part of the heavy lifting an AI assistant needs to do. From mind 2 i expect an SDK with a more complete AI framework.
Whisper is stt (with translation to english for free) from like 50+ languages, I’m sure there are open tts’s with similar results? Not sure, check out speech note on openrepos it uses a lot of different backends, so some might be two way, and again for M2 details, sell your soul and phone number to discord
“I’m sure there are open tts’s with similar results?”
Last time I checked, around 3 or 4 years ago on Nvidia Xavier AGX (with >30 TOPS vs. 6 TOPS of Mind2 
), they were so slow i didn’t want to pursue it further. Quality was acceptable.
Discord only has my email? Do they force mobile accounts now?
No clue, keep us posted
I would assume language models were a lot less efficient those “3 or 4 years ago”, so you probably need a lot less computational power to achieve the same result nowadays.
And Discord doesn’t necessarily require a phone number, only when certain “anti-bot” measures are being triggered due to certain markers being deemed (too) suspicious, from my experience.
1 Like
I’m hoping for the same, but I would like to see some more hard facts before I decide to contribute (via pre-order) to the project.