Viewing profile — jszymborski

This is indeed how it works in some countries, but depending on how strong a nations social net that isn't necessarily true.

Beyond all the other more salient objections one might have with this pitch, there is just the very practical fact that there really isn't enough signal from "noninvasive" sensors …

It's not easy to disentangle and it varies by country and institution, but those positions are not normally directly funded by public research grants.

> Scientific discovery is bottlenecked. Yah, by funding and how we award it, not by an imaginary lack of undergrad and grad students. Scientific funding requires a shotgun approach…

Super impressive results, and an awesome write up. Thanks :)

From the article > In the days of generative image models, it's still fun to see what can be achieved using traditional image transformation techniques. > But the really fun part o…

margin:0 auto is the (probably incorrect) incantation I have always reached for

If the error bars are so large that the old and new analyses aren't statistically different, then the margins are too uncertain to make any claim in the original work... which is s…

It's annoying they didn't plot the new and old "perceived" curves on the same figure, but if you pay attention to the y-axis, there is a very big difference. In the old plot, the b…

I've switched over to unsloth studio and am hitting those speeds now, thanks for the tip.

bahaha, LMStudio's badge looks a lot like a B but the tooltip says it is indeed an 8 (I just assumed it was B for "byte", but that's a poor initialism considering "bit"). https://d…

No, it's what I get repeatedly. Recall that it's QB_0 though. I'll get back to you when the Q4_K_M quant finishes downloading. EDIT: I get 259 tok/s with the Q4_K_M quant

I think sometimes the quantization is the source of the conflicting reports since those are often not included in discussions. Honestly, 27B is too slow for me to use very often, s…

Thanks for flagging, this is on my local rig and it's driving my display too. I'm curious now, will take a closer look. These are the tok/s as reported by LMStudio. EDIT: Updating …

On my RTX 5090, Qwen3.6-35B-A3B is my go to for coding, but Gemma-4-26b-a4b (Q4_0) is blazing fast and super high quality. I sometimes use the dense cousins of these models (Qwen3.…

I don't even know where to begin with this comment...

First and foremost, if DHH meant "junkies" or "homeless" by "gypsies", a term widely regarded as derogatory used to describe Romani people [0], then already there is plenty to take…

K2.6 is cheaper than GLM5.2 (at least on DeepInfra) and I've found it works as good as Sonnet for my purposes. Both tend to think themselves into circles a bit and aren't super tok…

Datacenters that have not and might not be built to accommodate for future AI demand that may or may not grow to the extent or as fast as the companies anticipate.

In fairness, the original crop of detectors were based on perplexity scores and were entirely useless (famously, the declaration of independence often came back as 100% AI generate…

This comment is for like three people on HN, but anybody remember when Paul Neave had a website/blog/showcase (c.2007-2010ish?) where the borders of the page had similar jelly effe…

For open weight models, you can choose from a few providers. Each have their own caveats, none of ToS'/Privacy Policies I entirely trust, nor do many make renewable energy claims.

Big if true! One of the caveats here is that the file size seems to balloon 341 times.

I'm just saying they should just drop the dubious claims and just say "I made a font that I think looks cool".