So now the average Joe has to make 3 vids instead of 1 every time - as everyone else is now min maxing their engagement?

Just sounds rubbish both for viewers (look how terrible titles and thumbnails are after a/b tests) and uploaders alike

Same token burn / cost though right?

I think OP's point remains, if you generate 140k pairs, your local model would need to run that many to offset having just used the generator (SOTA or not) model to begin with.

I wonder if another approach if latency is a concern is just to do a two shot pass with Jev (perhaps given small context you'd want one to match command, then one to match args of given command) would be an extremely fast, and cheap way to do it - rather than training your own.

> Same token burn / cost though right?

Not quite. One, because you save on the initial query being sent multiple times. Two, because the reasoning will be very similar at the beginning; "user asked me to", "let's check what's in this project already", etc. you'll get similar actual output cost, but input and reasoning will be shorter.

Initial personal use case (on my side project https://mealplannr.io/lists) is lists have a "smart categorise" button to group items into 20 or so preset lists.

Previously this took up to ~30-60 seconds using deepseek v4 flash (even with a medium list size) - Jev is <1 second @ same cost with typesafe guarantee

As a bonus I can also instantly categorise new items - rather than sending them into an "unknown" category (and waiting for user to have to click "categorise" again)

They’re a believer, but they won’t be paying third parties for that once it becomes feasible

Why can't it make a judgement grounded in real data points - then the human can accept / deny that proof much quicker?

Surely there's a middle ground / this is faster than reading all papers front to back.

Openrouter and huggingface both sold to multi billion dollar entities in the same few weeks - perhaps this is great for the founders, but as a consumer I hate this acquisition loop that happens with every company of adequate size these days.

Or maybe they know that this AI bubble will not last.

If that truly is the main selling point - it seems like a shallow moat versus skills + rest..

What's worse than an AI generated comment... a comment about having generated an AI comment that you can't read!

If it's useful to the conversation (and objective + with sources) post the key findings - IMO I think it's AI generated prose / subjective posts that will get blasted.