Do you think there will still be an incentive to release weights in that scenario? Everyone will have models only if there continue to be companies releasing weights.
Companies won't but I suspect this is a role that something else open source-y will fill that niche. Maybe orgs like wikimedia or internet archive, maybe some hackers just making things, maybe nation states that want to disrupt other players. Also model training will get better and better both on the algo and the hardware side. You can easily see a world where you might be able to train a good enough model on a home lab in a few days.
But you will need training data. Like a whole Internet search engine or massive data scraping. That‘s a thing that will not change with better algorithms, hardware or cheaper energy.
Data is the only moat but they'll be starting in the same place the current set of players statyed out just a few years ago. I suspect that the delta between what is publicly available (if not legally publicly available! see scihub) and what open ai and anthropic have is relatively small.
I remember trying an iPhone 12 in 2020 and feeling it was so fast that no phone task would ever be able to use all that power. Definitely not my current experience on my now old iPhone 12. A lot of it can be attributed to ever increasing ram usage by web pages, but that doesn’t seem to be all.
This setup feels cumbersome, since you also have to manually track which items you have read. Kobo seems to offer better features in this sense (better than a jail broken kindle), however I like the build of my Kindle Oasis 2 too much.
As an Italian living in another EU country, I always thought that the amount of (broken) bureaucracy of Italy was not particularly worse. However this story comes after a couple more I heard this week, in a line of absurd practice possibly due to absurd regulations.
I actually contributed to retrowin32 to get Solitaire running there. Back then the only AI tool available was Copilot, and it took me several days just to get the main window showing, without menus or dialogs.
The current state of RetroTick was achieved in less than one hour using Claude Code.
Funnily enough, when programming with agents in statically typed languages I always find myself in need of reminding the agent to check for type errors from the LSP. Seems like it's something they're not so fond of.
This is a viewpoint commonly held by students who were exposed to imperative programming before having any class in maths. However it shouldn't survive long after that.
I just moved to macOS for the first time, and my only way to adapt to its multi-tasking has been keeping exactly one window per open application, never zero or more than one. The fact that Finder can't be treated like that is a real pain. I will focus on it essentially randomly, and it will disrupt my intended interaction.
I don't get the reasoning behind the zero-window cmd-tab interaction, but if it is there I guess that there's a reason behind it?
On macOS you can have an app that is running without any windows open and you use the menu bar to invoke different commands in that app. This is why cmd+tab allows you to switch to an app that doesn't have any window open, essentially cmd+tab is an app switcher and not a window switcher. If you want a window switcher you can use AltTab an open source window switcher for macOS.
reply