I find it odd that any LLM could be considered open source. Sure the weights are available to download and use, but you can't reasonably reconstruct the output model as it's impractical for an individual to gather a useful dataset or spend $5,000,000+ of GPU time training.
Distillation can extract the knowledge from an existing model into a newly trained one. That doesn't solve the cost problem, but costs are steadily coming down.
That's still a crude repurposement of an inscrutable artifact. Open source requires you to share the source data from which that artifact (the model parameters) was created.
I'm working on a dumb tool to picture-in-picture show rendered GPS data as a mini-map with a forward- and backward-looking trail on top of GoPro footage (to memorialize cool bike rides).
Do keep in mind that it is relative to new aluminium. It might still be far better for the environment to use other materials.
For example, a plastic container might result in X emissions, a recycled aluminium one in 2X emissions, and a virgin aluminium one in 150X emissions. It just means we should absolutely recycle aluminium when used, but it doesn't by definition mean we should be using aluminium in the first place.
I much prefer Fabric server to Paper. Paper has a couple game-breaking differences in advanced redstone, and Fabric in general has more mod compatibility. There are options to disable the redstone differences, but it's a bit annoying.
Basically just enable the unsupported settings. My players were happy with the end result of each one I flipped on so presumably the behavior was as expected.
The kind that runs locally, or that is run by my trusted (by me, by my employer, ...) email provider. Email is just too valuable to be exposing to extra third parties.
Phones have made this more inconvenient because it's hard to make 'the app that runs locally' run whenever 'the app that views my email is running'. That means running locally is a big constraint on the UX, for example it probably can't reliably sort emails before my phone picks them up. It's simply the case however that UXes that don't work with local apps aren't viable products to sell to me.
Open source not running locally is closed source, for all intents and purposes.
A GNU/Linux box supposedly loaded with nothing but software libre is closed source, if it's remote to you and you don't own it. If you send your data to it to be processed, who knows what happens to it.
I hear what you’re saying, and I agree that running something locally is appreciably different than using something hosted elsewhere.
But I think it’s critical to be precise about this. Open source is still open source, and this designation is still meaningful even if an open source product provides a hosted version. The downsides of letting someone else host something are independent of the the open/closed source status.
Yeah, and I briefly considered that, but in truth I can't be bothered to run software stack of this complexity for my personal use for this. It's designed as a mutli-user web app, and that shows in the number of different services used in the backend. I like the idea of an app that sorts my email better, but not that much.
I would say I don't consider this a criticism of the product - they're clearly selling a hosted product that they were kind enough to open source, not making a product for end users to run that they happen to sell a hosted version of. It being inconvenient to self host doesn't reflect poorly on them.
It just needs to be proven and matured a bit more. It doesn't give the impression that much time has been spent building it. There is no indication that a third party has reviewed it for potential problems.
I don't think that's too high a standard. There is a Gmail email client, Mimestream, that you have to fully trust to use the Gmail API responsibly. It's gone through an extensive security review from Google's side to be allowed to use that API at a high volume. It was also in beta for a couple years and fixed a ton of bugs and edge cases during that period. There is evidence that a small team was hired which put more eyes and specialized minds on issues.
Mimestream has well exceeded the threshold to be trusted. I think this app could get to the threshold sometime next year if they work and communicate the right way. I kind of doubt they will spend that effort.
That's my opinion as an actual potential user (unlike those wanting it to be made of entirely locally run services, who weren't going to use it anyway.)
"It's gone through an extensive security review from Google's side to be allowed to use that API at a high volume."
Every app that has access to the Gmail API has been through this process including Inbox Zero. But I agree that you can have more trust in products that have been on the market for longer.
The safest way to run the app is to self-host but I agree that's a hassle for most to do.
I'm not the OP, but I would only trust an app like this if it didn't require any external services whatsoever. If every component can be self hosted + they add support for a local maildir or maybe IMAP, then it would be fun to try self hosting an instance of this app, but without that I wouldn't have any interest.
I mean it says you can self host it yourself... and its on github and opensource, and if it's using openai api, you can always use a self hosted api that's compatible with openai api to run mistral or llama to run it all locally.
Yes, you're right, but I don't know if it's possible to replace the tinybird component with anything local, perhaps there's a convenient alternative I'm not aware of (I'd never heard of tinybird before seeing it in the README's list of external services). Either way, the documentation doesn't have a lot of information about a local-only use case, so I can only imagine it would be inconvenient. If they add support for self hosted mail sources in the future, it may still be worth trying though.
Yes, atm it relies on Tinybird. But the level of trust you need for that is different. And the data in Tinybird is encrypted so even if you don't trust them to run a Clickhouse database, if you set an encryption key they won't be able to do much with your data.
We are planning to support a version that doesn't rely on Tinybird at all but stores all data in your browser using indexeddb.
