aae42

They were also really slow on heartbleed

beekeeping

i think this guy's just trying to gain future favor from the AI overlords

Disagree, I rather not subsidize that. I also don't really get why it has to be an EV, can't a self driving car just as easily be gas powered?

In my admittedly little experience running a tiny LLM service on a single 5090 for friends, I would guess a fair bit. It depends on how many tok/sec you're targeting and how much memory you're allowing be used for context (and their context). Definitely more than 1, MAYBE an order of magnitude.

Yes

I gather this is the whole point of the original article. It seems to have been pulled, so I can't confirm.

At what point do the AI companies start building their own hardware, I wonder.

Preparing them for the future when all of their work will be laid out this way to maximize productivity

A dull knife

$5 USB mic?

anyone could run this themselves with open web UI and litellm, in fact that's the stack I've been using

i find these discussions comparing the "vision language models" to the old computer vision tech pretty interesting

since there are still strengths the computer vision has, i wonder why someone hasn't made an "über vision language service" that just exposes the old CV APIs as MCP or something, and have both systems work in conjunction to increase accuracy and understanding

I've noticed this on my org as well, people refer to "putting things in git" and 75% of the time they're referring to the source forge.