Profile avatar
slckl.bsky.social
I work on software, mostly in Rust, mostly computer vision. Love weird music, science and pompous writing. I buy many books, but read few. Will mostly post/repost about AI stuff.
197 posts 87 followers 244 following
Regular Contributor
Active Commenter
Posts 15 Comments 35

This looks interesting! "Together these results support the idea that the principles of self-attention used in LLMs overlap with the mechanisms of language processing within the human hippocampus, possibly due to similar prediction-oriented computational goals."

submitted 2 days ago • 1 comment

Music can only be consumed sequentially, and unlike podcasts/videos and other forms of audio expression which can be conveniently sped up, music... can not be sped up. This is a great tragedy, an underappreciated ceiling of human ability.

submitted 2 days ago • 0 comments

Writing a report for a project about VLMs, and the "theory" section feels like trying to catch a speeding train with a fly catcher. Most sources are outdated the moment they've been published. With multi-modal it feels like all variations of it work with enough elbow grease (and data and compute..).

submitted 2 days ago • 1 comment

I feel baited.

submitted 5 days ago • 0 comments

Wow

submitted 18 days ago • 7 comments

Honestly, this photographer was wasted on corporate stock photos.

submitted 36 days ago • 11 comments

I didn’t want to learn about Platonism, but I finally caved

submitted 32 days ago • 9 comments

The sigil was drawn in salt and ash, the candles lit at the pentagram points, the incantation declaimed. There was a shimmer - a demon appeared. "Curious. What ritual is this?" "I got it from ChatGPT. I included all protections in my prompt!" "I see," the demon said and stepped out of the sigil.

submitted 35 days ago • 39 comments

zapping a box of rocks until it summons the omni-comedian to deliver a mediocre joke derived from the thousands of tortured souls it has enslaved

submitted 36 days ago • 3 comments

Yeah we finally have a model report with an actual data section. Thanks Qwen 3! github.com/QwenLM/Qwen3...

submitted 44 days ago • 1 comment

New blogpost: "Training as we know it might end". It was originally a panorama of the new methods of synthetic generation but the stakes are now much higher and I openly wonder if model training is not soon going to change forever. vintagedata.org/blog/posts/t...

submitted 49 days ago • 4 comments

In 1999 philosopher Douglas Lackey polled philosophy professors asking them to rank the 5 most important books and the 5 most important articles in philosophy from the 20th century. Here are the results from the 414 replies:

submitted 51 days ago • 8 comments

honestly these are quite helpful. i feel like i’m finally understanding the benchmarks

submitted 51 days ago • 0 comments

isn't this just any complex python software, towers of inscrutable spaghetti layers, with miraculously working multiprocessing, sharded across the data center to infinity, complex write-only pipelines that do illegible holy work, the old ones slumber across the ffi boundary, just add one more if bro

submitted 53 days ago • 2 comments

Highly recommend the old video presentations as well: www.youtube.com/watch?v=yBL7... Shows how much science remains to be done with training relatively small models (100Ms of params).

submitted 53 days ago • 0 comments