rasmus1610.bsky.social
Applied AI, Neuroradiology, Web Development
68 posts
86 followers
134 following
Prolific Poster
Active Commenter
comment in response to
post
huggingface.co/dslim/bert-b...
comment in response to
post
Sentence classification, information retrieval, named entity recognition are among the top use cases.
Everything that isn’t concerned with generating text but doing „discriminative“ tasks on text.
comment in response to
post
I experimented a little with Preact and HTM and it’s kind of nice.
comment in response to
post
Absolutely not :)
comment in response to
post
Your productivity is incredible 😀 if someone knows how much time you put into the current course, it’s even more incredible
comment in response to
post
Is Jurassic nbdev but for Deno? 😀 great name tbh
comment in response to
post
Haha, no apparently they don’t have a www to non-www redirect. Says all: www.editorialmanager.com
comment in response to
post
I bet even random forests or support vector machines will still get you very far for many problems in the real world
comment in response to
post
But what do people use for discriminative tasks like image classification nowadays? I mean the use cases haven’t gone away, or did they?
Using a VLM for image classification just feels wrong …
comment in response to
post
100%!!!
I love `uvx`
bsky.app/profile/rasm...
comment in response to
post
TIL: you have to post gifs as a video, otherwise it won't run.
comment in response to
post
I know that e.g. Pandas support such a programming style out of the box, but unfortunately it's not as widespread as in javascript land :(
comment in response to
post
Yes that’s the idea… will report back :) ultra efficient visual RAG sounds exciting
comment in response to
post
I know, but you need a VLM to generate the answer when you retrieved the correct document, or am I mistaken?
comment in response to
post
Actually the post was because I use it in a Astro.build project. But I like to use in FastHTML for stuff like showing and hiding dropdowns, mobile navigation etc.
All of this can be done in pure js but alpine makes it so much more concise
comment in response to
post
Oh I’m so excited to see how it performs with ColPali et al :)
comment in response to
post
Phew, would be awkward otherwise I guess 😂
comment in response to
post
Maybe an internal peer review (aka showing your work to colleagues and ask them of their opinion) would have the same effect tho
comment in response to
post
The peer review of my recent paper (using fast.ai btw ☺️) is surprinsgly helpful. But this is more of an exception than the rule
comment in response to
post
Okay, a quick google search didn't turn up what I was thinking about so I think that's something I need to investigate myself 😁