Davide CIttaro

I just knew that Apple is supporting a project for CUDA bindings on Silicon

I wish I had resources to port it to Metal so that I can run on my phone.
Also, that would make sense of all these Mac laptops so popular among scientists.

“Ballmer was right” hurts.

Is it CUDA accelerated? NVIDIA specific?

4-panel SMBC comic update. Two women are talking as they walk down the street. The first woman, dressed in yellow, asks which modern democracy is the most like Athenians. The second woman - a bespectacled redhead, answers without hesitation that it is America. She continues to explain that some countries are as intellectually vibrant as the United States, and some countries are just as violent as the United States. She concludes that only one modern nation could both produce and execute Socrates. The woman in yellow responds that it's not *exactly* the same, as Athens didn't let women vote, prompting the bespectacled woman to say "we'll get there! Don't give up!"

On the last Socrates joke, like a dozen people told me I should've called his nutritional supplements Himlock, and it just kills me that I didn't.

COMIC ◆ www.smbc-comics.com/comic/athenian
PATREON ◆ www.patreon.com/ZachWeinersm...
STORE ◆ smbc-store.myshopify.com

A bit frustrating when a PI presents data and results but doesn’t go into details because “I’m not a bioinformatician”.
You may not be able to do the same calculations, but I expect you to be able to understand and present data.

The 978 genes in L1000 assay can be used to infer the expression of many other genes, and yet don’t make a commercial panel in any assay (that I’m aware of). Why?

Great work!

You can call it orthogonal

Actually that’s a true request I heard 😖

It’s weirder when NGS is used to validate qRT-PCR

The "publish or perish" culture must perish. Scientists need time to think.

We just published our Slow Science Manifesto, where we argue that huge changes are needed in the way we fund, publish, and evaluate science.

Read more and sign here: www.slow-science.com

I see now in discussion. I couldn’t read the paper before bioRxiv rendered it in HTML 😅

Next are hypestgraphs

You forgot the 5th point: the prefix “hyper” makes things much more interesting 😅

Yep, we do that kind of blunders

Great! But what was missing in inmoose?

inmoose.readthedocs.io/en/stable/

How long does it take to update profile data?

Every time I need to test an R package I spend hours compiling, resolving dependencies, fix inconsistencies.
I used to work with FreeBSD and Gentoo Linux, I swear I never spent so much time in compiling and complaining like with R

Same

Exactly!

Yep!
Before uploading my source I explained the context, I naively thought it would have ported code seamlessly :-)
I wonder how much would have taken to debug the error without prior knowledge (or worse: not even noticing the problem)

*20000

I’m simply glad that the value of my +20y experience was worth something.

I prompted Claude about this and it modified the code to allow parsing of block compressed gz files.
Now it works and it is 10x faster than Python.
Good for me, maybe it could be faster, but that’s not what I’m interested into, right now [6/n]

I complained once more and Claude started thinking deeper.
Then I thought fastq.gz file are often compressed with BGZF. Quick test compressing 2000 with regular gzip and the software worked
[5/n]

I reported to Claude, it started working on fastq parser, adding some more checks.
It didn’t work.
[4/n]

No problem, Claude corrected everything and then the code worked. I thought it was blazing fast (few seconds to complete 100M seqs), but I realized it ended after 15000 seqs were processed [3/n]

I wanted to test Rust performance but I cannot write complete code (yet?), so I asked Claude to port the code for me.
Little time and I had my .rs and .toml files. Didn’t compile [2/n]

I wanted to share today’s experience with AI writing code for me. Interesting (ish).

I had to process some large fastq files to count specific sequences (barcode/guide, CRISPR library), I had a Python script that works but is slow (~200 items/s). [1/n]