I am running many different ones right now (local only, not connected to the internet) and have been for a few weeks. Most recently llama II 70B parameter model. I need to get an easy mechanism to fine tune and also ensure controls are solid in terms of “phoning home” I feel the latter part is pretty solid though.
Yes, I have been amazed, confused, worrying and pondering this since last November. As we get to launch I think we need to know more but the speed of dev in this area is mind boggling. What keeps me feeling better is personalised and local to you AI. I know that’s possible and we have some pretty radical ideas in that regard. The bigger and more important issue is global AI sucking up all our data and then using that to persuade and control us.
So the localised AI can give us the convenience and safety we will need.
It’s a large changing landscape and keeping up is not easy, but the runes look like we must do something to protect humanity with the advent of this incredibly powerful tech. If left to the global players then I have similar fears to many of the leading experts (not that I am one of those).
There is a path and it’s a path SAFE can offer many will not see, but I feel it is
worth looking at
likely transformational
Very much unknown as to it’s end
Most of all I think this is going to be humanities biggest challenge of this century - Do we embrace and control the machine or do we allow mass control. I don’t think we can ignore it.
There is an element of that for sure, the man never stops!
I’d like to get your minds going on:
How do we enable individuals to take advantage of all that AI has to offer in terms of productivity, learning, and enhancing the human condition, but within the bounds of controlling individual privacy and self sovereignty.
There is a new frontier in that regard. And what is either side of that line is simultaneously exciting and invigorating, and also terrifying and disturbingly dystopian (and not even considering AGI).
How do people, and then society, give meaningful consent to these things?
And then there is the role of public AI. How do we avoid it’s extreme centralisation and corporate capture (where it is already pretty much), and re-distribute power back to the commons?
Well… I’ll let you imagine, and dream too.
(But what I would say is David has been proven right more than once)
Those responses don’t demonstrate a route for the masses to be able to use LLMs (or whatever) as a S.A.F.E. interface to SN.
It doesn’t matter whether you’re talking about an LLM here or a locally hosted search engine, the path of leaving the masses grasping for that kind of help is known and it is convenience, capture, surveillance and exploitation.
SN surely needs a better answer than this if it is to be for everyone.
Let me clarify, if I can’t use it on my £50 Android phone, which just about manages this forum in a web browser, I’ll be very disappointed.
I do get that this not what we mean by launch, or even MVP, and I do get that much will be needed later.
I’m concerned that if that later takes very long after launch, the point of launch becomes moot wrt the fundamental: everyone. The E in Safe.
This, I think we all agree on. Where I am coming from is exactly this point. We all need protection and we all need equal access to knowledge and the SAFE network has to provide that in a way that is anonymous, secure and private.
So we are missing each other here and more will be revealed very soon, but powerful tech cannot be left in the hands of the few and withheld from the many, whether we like the tech or not, if it’s powerful then we need to make it equally available to every person on the planet and that means they need to be able to use it and do so securely without leaking data about themselves.
Just thinking about that and the current direction of tech (google/bing etc. are incorporating the LLM tech already) then the downsides of global control and access to our data will transform into global control of how we think. Much more than now.
A bit macabre perhaps but you know what I mean hopefully
In this interview Musk brings up some of those fears you talking about. When Musk confronted Google inventor Larry Page about the dangers of AI, Larry called Musk a speciest and wanted to build a AI god as fast as possible. Can be seen around 5:30
15:30 about AI bots and AI like Chatgpt corrected for politics or other interests to lie.
I’m saying that if the product relies on a local search engine or LLM because it is impractical for Joe Public to use xor addresses, the network will not be for Everyone.
And that this leaves open the trap for adversaries (such as a new Google like start up) to offer convenience (search or LLM services) and tempt users into giving up the privacy of Safe Network in order to access it.
To work for everyone SN needs to be useable out of the box without holes that adversaries can fill with convenience.
If it begins as a difficult to access but value filled data store, services will be built to make it convenient, and users will trade convenience for privacy all over again.
Thanks David. It sounds like we are of course agreed on the goals but that you believe you have a route which will work while not yet ready or able to reveal enough of that to convince me.
I look forward to putting something to my £50 Android phone test in a few months
It won’t use an LLM As local search, LLM data is just another data source to us. It may be the only one many folk use, but it has to be local to them too. At least for personal data.
The goal here to to not repeat those errors of the www. There is a growing concern that search as we know it may suffer here. Raw data may not be accessed so much etc. but that’s not our concern and not our call to make.
Where we need to be is equal access to Everyone to humanities knowledge/data or whatever ew want to call it. But the key is get everyone on the same playing field and leave none behind.
In terms of access to raw data, it does not need to be based on the old www either. We have options here with content addressable data itself. Thos options surely go well beyond somebody putting the data in container for us to access. We can all have our own containers, our own www, we can share that with folks we know and much more.
Social networks and mobile phones have largely made do without DNS (App Store/vid links etc. etc) and apps replace web sites more often that not now. There is a plethora of things to consider here.
AI just makes it all much more important that we do consider this now and ensure the network data is cryptographically correct and consistent. That is easy if data is self validating and DBCs are auditable. It get’s hard if data can be replaced and not proven to be correct,
Probably going in circles a wee bit here though sorry chap
There is a opening in the beginning for like Maidsafe foundation to make a service were trusted sources can get verified and get some form of green marking or similar. Even if the goal is to allow competition, from the start Maidsafe could help with verification so that trusted sources could apply for verification. It could be as simple as a public Safe site where the browser could have built in function to look up addresses and a built in function to mark address in green if it is verified and maybe also give sites names similar to DNS.
@dirvine
Original comment ended up in another wrong topic but been thinking about if it is possible with a system of pure supply/demand were node operators and clients meet and finds the current price for storing. Conclusion is that it is not possible based on that it would probably bring very skewed and concentrated storing of data, which would be bad for data replication for backup and liveness of the network. So random rewards as Maidsafe intends is probably the way to go for network liveness and even distribution of data over any given total current nodes.
Never heard anyone give this explanation but it feels logical correct when thinking about it
As probably many with me sometimes experiences internet down time a couple of times a year or more from example ISP maintenance or similar. Would it be possible to find solutions were nodes could recover their lost position in the network with a penalty or similar. Will a lost connection mean a node always needs to start over?
Yes, until there is something more concrete I don’t see how you envisage delivering what you describe (SAFE) using the elements you’ve mentioned (content addressable data plus a local LLM?) in devices that anyone can use.
My point is that using the fears of AI siphoning public data is not going to bring adoption, at the cost of convenience
I am not being dismissive about the dangers of AI and the need for security, I thought the rest of my post made that clear.
Of course it must be secure from day 1, and I get that at this stage NRS may not be essential, but I fail to see how this becomes a security dilemma, or an AI threat, I would like to understand that a bit better.
As long as the user clearly knows what is public and private, the AI would only be able to gather info from publicly available data, are we talking that even OSINT should be impossible to do in the Safe Network?
How it is marketed and to whom is a different—albeit related—question.
NRS isn’t an AI threat/dilema, and as I say it’s a bit of a tangent on this thread TBH, but this is the salient point:
it’s about giving effective controls and the infrastucture to allow people to control access to their data, personal info, thoughts, identity, and ‘self’. Yet still reap the benefits of having their own ‘cloud’ and utilitsing applications and AI on top of tht as they see fit: without middlemen.
It’s the same threat as before, but bigger stakes, and a more formidable adversary.
I would actually say that the damage is already done. Countless of leaked databases, thousands of leaked classified documents, billions of data gathered from social networks (when privacy controls were not a thing yet), billions of hours of media content, and all the OSINT sources… the training data is already out there.
In fact now they are talking about synthetic training data, they may not need us anymore.
The Safe Network will help to prevent further violations and personalized data mining. But remember, the people didn’t listen to us when we were warning about privacy risks from social networks and ad networks. People still click on accepting all cookies because they feel it doesn’t matter.
So clearly it is the wrong sales pitch to tell them that the SafeNetwork is a safe place from the mess that they themselves created by not giving a damn about privacy haha
Something as simple as an email AI that can look at all past email history and find contacts, matching emails, or answer questions in response to native language q’s.
… and for messaging/text of course.
Similar for one’s hard drive. essentially much better search. And if this can be extended to all of a person’s files on safe network, that’s cool.
Also shopping history. what was that thing I bought 3 years ago and shpped to my sister? I think maybe it was on amazon, ebay, or target…
Not at all! Whether stuff embraces consciousness or consciousness embraces stuff is essentially meaningless, because meaning is only relevant in consciousness. So beyond consciousness: who cares?
but you know what I mean hopefully