badatnames

I've worked in a few places that had ultra-privileged analysis tools, would it really be a surprise to learn their security team has an MCP that queries a full text index of all user inputs. It would almost be more surprising to learn they don't, seems the kind of thing that would be invaluable for model research too

> this can be a matter of scaling at 50% capacity

At which point we're paying double just to work around cold start latency, which was the original point.

That's only how it works when misapplying mean() to characterise application performance. In your scenario, every request is hitting either an overloaded instance, or when scaled down, a 20-30 second wait for spin up, better hope they're neither interactive nor some jobs configured with a short timeout. The mechanisms are so completely different they're hard to even make meaningful comparisons between

So long as you don't count >2 orders of magnitude difference in scaleup time as particularly special..

These one-sided arguments come up every time there's a Lambda thread. They're useful and easily cheaper almost anywhere you have sparse compute or RAM requirements, even for heavyweight tasks like transcoding. A t3.nano meanwhile can simultaneously be overpriced ($3.50 for an idle task) and a bottleneck (30Mbit/s sustained network) in just as many places as a Lambda could be applied incorrectly.

There is a collection of these someplace. Rust is better but it'll never be perfect, news at 10

> Even formats like H.264 are being dropped from some hardware decode engines because it’s so easy to do on CPUs now, as can be seen even on the Raspberry Pi 5

CPU decode efficiency most certainly was not the reason h.264 was dropped on the Pi

I'm simply not seeing this

DS4 (not 4.1) crossed my dont-care threshold and I genuinely stopped paying attention to new models. I'd love to try GLM 5.3 but I just don't see any point in spending the effort any more. I can get passable intelligence for a bargain price either direct from China or from a ZDR EU provider for a small markup. Paying 10x more will not make me 10x happier, it's unlikely to make me even 1.1x happier now I've got some intuition for the natural limits of these models.

I don't even bother checking how much I spent on API any more, its well under $30 over the past 2 months despite daily constant use. Who even needs a subscription at these numbers?

This interminable argument has been raging at least since the days of the East India Company, I'm sure even Jesus had some thoughts on the preferential relationship the money changers had with the temple. It makes the world go round. All the wealth used to train these models comes from compatible relationships between power and creativity, and we rely on that compatibility. That a regime to lock out US vendors from European infrastructure contracts was even necessary is too fresh a story to reduce it to a complaint as boring as regulatory capture, go ask the folk in Greenland for their thoughts.

Their paper shows this comes with huge quality loss, but that doesn't make it a negative result by any means

Yes, and image editors absolutely can. If JXL lets me correct over/underexposed images by default everywhere I find them then it's worth it even just for that reason alone