Taskaside @Taskaside1

@hmorneau @lydiahallie @theo well seems it was dictated by the low weekly limits of fable, that particular theo's choice, but yeah if they don't speed up inference for higher amt of reasoning tokens, then anything over high is gonna be slow, i mean, yeah

0 0 0 31 0

View Details

Taskaside @Taskaside1

3 weeks ago

@jdjb78 @bitbased @theo well, making a true router model/ at least keeping an eye out on couple metrics of cost/performance all these benchmarks people make would be great, but since even launching an orchestrator model drives the cost of cache and in/out tokens by extension... it do be tough

0 0 1 60 0

View Details

Taskaside @Taskaside1

3 weeks ago

@meine___meinung @weswinder @coffaccelerate @thsottiaux what difference does it make when OpenAI needs to be cooking up some serious response to Opus for frontend work, otherwise people will just switch over from astra, good thing OAI is known to update and iterate on .1 increment - so here is to gpt 6.1

0 0 0 55 0

View Details

Taskaside @Taskaside1

4 weeks ago

@bitbased @jdjb78 @theo harness that does handles both should route intelligently enough to take care of your choice dilemma of what to choose where. and that is clearly missing. Despite theo's benchmarks etc ( I mean clearly sometimes it's a preference choice but sometimes it isn't. )

1 0 1 141 0

View Details

Taskaside @Taskaside1

a month ago

@ChrisGPT @gailcweiner i mean any recent/decent model can, the cua is crazy, closest thing to codex app computer use feature by a mile

0 0 1 55 0

View Details

Taskaside @Taskaside1

2 months ago

@EvanOtero @pedrodias host some smaller audio gen models, or release some, akin to gemma model line up, add good quality tts playback of generated chat response in ai studio available, aware of context like full files so it doesn't read them blindly in full when unnecessary