And: on which hardware? My XA2 does not have as nice AI-Engines on chip like my silicon apple…
I’ve said it again that first and foremost a device running SFOS needs to be a “reliable” device. Leaving outside issues like device availability, restoring a new one or issues i might face using a community port (gps, network issues, camera) as SFOS is right now you can’t rely on it.
Ie i’ve been bitten in the ass a few times by the browser not being able to handle certain stuff. Or the email arriving whenever it feels like -instead of always up to date-.
While AI is fine and can be helpful i’d preffer being able to do the basic stuff first. TBH i’d be more excited with a collaboration with a bank (for being able to pay etc) than with an AI assistant of sorts.
3 Likes
I love all this “artificial intelligence”. Ask it how to create mustard gas and it’ll say that it can’t tell you because it is forbidden. Ask it how to avoid creating mustard gas and it’ll tell you to never mix bleach and household cleaners. Intelligence, my ass 
4 Likes
The device is intended as a private-cloud-cum-AI-router that makes the user’s data accessible for running AI queries.
“We don’t believe it’s a good model to put AI run locally here in the phone. It’s not secure,” argues co-founder Antti Saarnio. “You can never make smartphones secure enough for that. We believe that in this AI era privacy becomes much more important.”
I genuinely fail to see how a static, always connected network device that is supposed to act as a basket for all your eggs (which, metaphorically speaking, means your data) is more secure than a smartphone that:
- You always carry around with you and cannot be taken away unless by force
- Has designated transistors to handle local models efficiently
- Does not necessarily require Internet traffic
To elaborate, Hexagon NPUs have been available since the Snapdragon 855 (2019) for Qualcomm devices. On top of that, being able to interact with both the unprocessed and AI-processed data locally due to the I/O offered by smartphones (which includes display, speakers, touchscreen, etc.) renders sending the data redundant and unsafe, due to the nature of the Internet.
I am glad Jolla is spending its already limited resources on the AI hype train and simply neglecting years long requests by SailfishOS users. In addition, the bad English and awfully written article (as expected by TechCrunch) really were the icing on the cake.
3 Likes
First of all, kudos to Jolla for getting in the prominent media again.
I think this can potentially work. It’s like running an LLM on your PC, being able to use it remotely and getting it to work with your Sailfish data – I’ve actually thought about attempting this but it’s way beyond my knowledge.
I don’t think anyone currently sells such a solution for any phone. OpenAI will lease you a custom database but that will be extortionate and you’d have to give them all your data.
Jolla will need to be able to prove it doesn’t phone home.
LLMs get regularly updated, which isn’t easy. If they can streamline that and get it working for Android too, I can see a small market there.
Why create something that already exists as software and can work with SailfishOS? It is called Nextcloud and has AI build in already (and is expanding).
As Jolla found out before, don’t focus on hardware (phone, tablet) and stick to your core purpose/ability. And how long will a hardware AI device last with the amount of improvement made each AI chip iteration delivers currently.
If you don’t have the amount of human resources (developers) and competitors have bigger teams and are already delivering, as Nextcloud does, rethink your idea of going this route.
I think they are jumping on a topic they are not able to compete in and will be the next rough sea they are getting themselves into.
As others said, deliver quality (less bugs) and stability first before starting with AI. Show us you actually can deliver on promises before jumping ship.
2 Likes
if the device costs the predicted 300€, I’ll get one. If its usable, why not to feed it sailfish OS documentation and code base, then use it to develop OS and apps with more ease?
Naturally if talk is the only way its able to provide information, its usefulness in coding is limited, but it can still assist. I’m willing to try it out. 300€ won’t buy me new phone even, so if I can use it for betterment of Sailfish OS, I’m all for it.
I feared that they would drop Sailfish OS completely, but not yet. So we might even get to see Mind² API on Sailfish.
3 Likes
You mean like this?
The sneak peak we’re getting is just that: A glimpse of what’s coming down the pipe.
There used to be a twitter bot that would call out that particular malapropism.
1 Like
I don’t want to sound negative, i love SFOS even through all the existing bugs, and i be using it as a daily driver, they are a David against Goliaths. I really like the idea they created now, of a physical AI device, and if i see it right, it would even be quite clever in the means of privacy, as for example in Germany to get a search warrant for a house or appartment, or to confiscate a physical harddrive in a property, rules are quite strict, in opposite to virtual storages or also the mobile device in your hands while on the way. But what i don’t see, or do not think is producing its on hardware. With big players like Samsung, LG, Xiaomi, or Apple. They have big money, so for them R&D are no problem. But in the case of Jolla unless they are doing some kind of cooperation, with universities, the state or perhaps a foundation specialised in the field, i just don’t see how Jolla could bring out an own physical AI assistant that is performant but als secure. As the hard and software would have to be perfectly adapted. For the moment there is no such incentive of Samsung, Xiaomi or Apple so how could David Jolla have the ressources, money and people to realize such a task. I wish them good luck from my heart. And would consider to buy a such a device if it ever comes out, but i would see other priorities for Jolla for the moment. But maybe it succeeds, and Jolla can generate money to further improve SFOS and also Android support 
4 Likes
Also like they are going separate device way instead of hamfisting AI onto device (battery issues alone are a dealbreaker, and all those AI chips like tensor on google pixels (hell even MTK boasted having AI chips on the soc of cosmo lol) still are dormant and unused as were not designed for llms, AI on device is couple years away). Who cares nextcloud has some initial AI apis for LLMs? Maybe jolla can even just slam a nextcloud instance into that black box, who cares, as long as they provide what they seem to promise - privacy oriented way to use your own data by you alone, sounds good. Biggest hurdles will be in secure apis + secure connection to your black AI box, communication needs to be two way if we expect AI to come up with helpful tips (gotta have access to the latest updates be it from emails/sms/even audio calls) so something like Zero Tier One, but without the third party as most people don’t have external IPs (and v6 is still bit away as always). Whatever brings in investor cash (they gotta hedge their bets on the next BIG thing) to keep sfos alive, and that’s the current hype, so be it
1 Like
Because it can do sweet FA with SailfishOS? Or Android or Apple for that matter.
Firstly, their AI is OpenAI. Even if you’re willing to give them all your personal data, it wouldn’t know what to do with it. OpenAI isn’t at all open so you’d be better off running Llama on your own server. You then have to train it to do remote access of your phone.
Now, LLMs aren’t people, they’re neural networks based off text input. Training such a thing is easier said than done.
Ideally, you’d put the thing on custom low-power hardware with an Nvidia laptop GPU. This will of course cost you around $200-300.
It’s doable but not by me and probably not by you either.
So sad, to me that sounds like SFOS will continue to loose much needed focus, funding and motivated people to work on it.
Last breaths?
2 Likes
When AI can make that
-
on my X 10 II or any upcoming device apps like puremaps or osmscout running without crashing when getting a GPS fix,
-
the native mail app is able to handle mail aliases,
-
the native browser is able to browse any website without crashing,
-
android apps are able to interact via bluetooth, so that my Garmin sportswatch can talk with my smartphone,
-
car bluetooth is always working well to receive and make calls,
-
I can make phone calls via loudspeaker with good sound quality on both sides,
-
and I get my double tap to wake up from JP1 back as the icing on the cake
-
…
then, yes then I am happy with AI.
3 Likes
Instead of runung on a separate device, the model could also run on a mobile device. So it might be part of SFOS one day …
3 Likes
I never used Siri on my old ipad mini. I switched everything off that I do not need. No predictive text, no airdrop, handoff and I don’t like Apple meddling with my pictures. And I hate pop ups that want me to do more with them. On my Sailfish device and Volla device I don’t use spelling correction and predictive text either. On my laptop I get a pop up in Firefox when I write or search in e.g. German, English or French, asking if I fancy translation. So each time I click that away.
One of the advantages of Sailfish is that the user interface is simple so I can have more control over it.The file e.g. app is great.
AI is fantastic for medical science, surgery, radiology etc. AI is ideal for disabled persons. But AI makes non-disabled persons disabled and lazy.
Pupils get no reading skills, no skill in math, their brains do not need to store knowledge because they can ‘google’, their memory is not trained so they have less associations.
Of course it is easy in certain circumstances ( a digital secretary) but if AI becomes standard and widely used we will see more people losing their independence.
About Jolla: 40 people worked on the project, the article says. Couldn’t they better have worked on the stabilising and improvement of Sailfish?
4 Likes
It’s hot and popular and at the same time the best AI experts are warning for the disadvantages.
2 Likes
Yes, it was called ‘internetofshit’. Mostly satirical about ‘the internet of things’.
But lets do an assumption. What if the creation of the AI device is more or less just to get into the headlines. Because they really created something that is not on the market, and this seems to create great interest in the medias. With the bigger attention, it will attract private consumers but also investors. If now the upcoming hardware sells good, Jolla will get new money for financing. So maybe now could be the moment, to bring forward a new SFOS device with a new port and also the possibility to flash the devices, because with the publicity a greater public could be attracted. And a user base big enough, to live from it, and attract App developpers
4 Likes
I agree the publicity and ability to attract new community members is the most interesting part. New hardware is also risky, as Jolla has found out in the past.
I am personally not attracted by the AI features, but if it is a custom hardware then Jolla can probably bring full hardware support. There are some missing features that annoy me with SFOS and with the 10iii, but I can live without them (user @silta above quoted some). However missing communication protocols that are becoming compulsory is an absolute deal breaker for a phone, so it should be a priority.
If ideally the new device has flawless VoLTE support, also 5G and US bands, it’s a good phone and I’ll get one, independently of the case for AI.
3 Likes
Is it only me or do we have a lot of posts here saying something like «Blabla, but can you fix $mypapercutbug»?
8 Likes