If there was a standardized way to save this information, and tie it up to each commits, it would be insanely useful to amass a very valuable training dataset.
HN user
Marlinski
Hacker and researcher in computer science. I am particularly interested in AI, and decentralization technologies (Blockchain, P2P and Mesh Networks). I have great passion for new and innovative technologies that enables communication between people, devices and places.
This is the actual standardized approach with the dtn protocol:
https://www.rfc-editor.org/rfc/rfc4838
It makes no sense to use a protocol where the source is the one responsible for taking care of retransmission when the network spans accross minutes/hours of delay.
I used to be an android developer and they disable my account because I took too long to reply to their mail. Since then I have been unable to recover it, they never reply to email and process your request to oblivion. Their bureaucracy is even worse than our french administration and that is saying something! At this point google is basically digital sovietism.
that was basically Rumble an app I developped 10 years ago: https://github.com/Marlinski/Rumble
I worked the field both academic and startup, I even made one of the first implementation of the Bundle protocol for store carry and forward (IETF transport protocol for the deep space network RFC9171).
Turns out the Mobile OS are making this kind of communication nearly impossible. To work well, it basically needs background job (automatic scan of nearby ble/wifi/radio) and automatic connection without user interaction (imagine being prompted to accept a connection every time you pass by someone), both have been basically made impossible (especially after covid).
There's so much to wonder and discover, this is a great post. I have been many times to China and it's always a delight for my tastebuds.
Mistral is great, I love their image generation and speed at which it replies. They really don't benefit as much hype from the others contenders but it feels like they are the silent undertaker.
That bright yellow background imprinted my eyes for 5 minutes, there's other way to leave an impression than fucking up with my senses, like writing a clear concise article instead of a never ending list of examples interrupted with distracting fake mouse move.
An horrible web experience through and through.
Wish this was done over Bundle Protocol, would love to code that for them!
I'd suggest setting up a raspberry pi and build your own service locally hosted at home!
I suppose it can be quantizised
it's been criminalized to hell by IP holders and hollywood. Such a shame they killed the best tech of the previous decade. Could have revolutionized how we distribute content, approach CDN and even streaming.
Their main advantage for now is their super clean API. Open source alternative are already on par with GPT-3.5 and 4 capabilities, they just don't have as good a package but that could change rather quickly too.
I am the developer of this program and I just noticed this thread. If you have any question about how it works I will be happy to answer.