@experilearning

biocompute

London, UK
Joined October 2020
Left my job to focus 100% on math, CS and ML for a while. Haven't got a precise end goal in mind at the moment but I think something will emerge. Here's the stuff I am learning, learning process, etc: 1/n
41
92
7
1,868
413,222
Quick experiment - a CAD language is a compressor for geometry; if you ask claude to design a language to maximally compress a set of STLs, what language does it come up with?
2
1
8
1,286
It did come up with some more interesting ideas, eg. adding unions and intersections to combine and subtract shapes. It chose to use signed distance fields for its CAD engine. To bootstrap programs for each target it used program search over its DSL.
1
1
135
I stopped here because the dataset started to feel like the main limiting factor. I would be keen to people's ideas about what sort of setup could elicit a truly interesting language with novel ideas?
1
1
112
you can convert pesky STL files into human editable STEP files using claude + build123d
3
1
1
15
1,968
bit dramatic but I do feel neglecting any kind of reading / learning in favor of frantic MAN OF ACTION over the past 3+ years has caused significant brain shrivel, please send material (blogs/books/whatever) to restore me to full turgidity
6
24
1,867
Did a quick blog post about the rise of vibe hardware, featuring my dad who's been flexing on me with some modular synthesiser stuff I don't understand (what's a vactrol?) 1/2
3
1
18
1,064
I’ve wanted to do this for a decade. But I never did - I refuse to give any company my DNA. It is me. So this week I sequenced my genome entirely at home. Literally on my kitchen table. I never exposed my DNA sequence to the internet. Not at any point. I used a MinION to do the sequencing (it’s smaller + weighs less than an iPhone). I used open-source DNA models for the analysis (Evo2 and AlphaGenome) running locally on a DGX Spark and Mac Studio. I traced mechanisms behind my family’s multigenerational autoimmune conditions that no clinician has been able to understand. When I set out to do this I didn’t know if it would actually work. It does. Your genome is the most private data you will ever have. You probably shouldn’t let it leave your house.
I'm lucky enough to have a great doctor and access to excellent Bay Area medical care. I've taken lots of standard screening tests over the years and have tried lots of "health tech" devices and tools. With all this said, by far the most useful preventative medical advice that I've ever received has come from unleashing coding agents on my genome, having them investigate my specific mutations, and having them recommend specific follow-on tests and treatments. Population averages are population averages, but we ourselves are not averages. For example, it turns out that I probably have a 30x(!) higher-than-average predisposition to melanoma. Fortunately, there are both specific supplements that help counteract the particular mutations I have, and of course I can significantly dial up my screening frequency. So, this is very useful to know. I don't know exactly how much the analysis cost, but probably less than $100. Sequencing my genome cost a few hundred dollars. (One often sees papers and articles claiming that models aren't very good at medical reasoning. These analyses are usually based on employing several-year-old models, which is a kind of ludicrous malpractice. It is true that you still have to carefully monitor the agents' reasoning, and they do on occasion jump to conclusions or skip steps, requiring some nudging and re-steering. But, overall, they are almost literally infinitely better for this kind of work than what one can otherwise obtain today.) There are still lots of questions about how this will diffuse and get adopted, but it seems very clear that medical practice is about to improve enormously. Exciting times!
414
1,044
241
12,665
2,495,200
Science begins with myths becuase by the standards of the consensus "best current theory" which defines what we think of as "real" any deviation will appear like fantasy
2
1
5
359
Top centaur-era skill = leveraging massive search. How fast can you spam attempts in parallel and trial and error your way to success. Make trials as fast+cheap as possible, learn from failure, apply computational approaches creatively to domains people aren't using them
3
319
Daily progress #6 + utilisation rate is going up, the goblin is satisfied - but more can be done - the big unlock is going to be increasing walkaway time so I can run overnight with 100% confidence
8
330
Daily progress #5 + made a script to track what % of the day our workcell is running + my sole aim is to make number go up + had fun benchmarking claude against himself at colour mixing optimisation + the system feels robust enough to leave running overnight 💪
2
14
425
There are so many lab tasks that can be made 1000x more reliable by effectively saying "Claude, hill-climb this for me pls..."
Friendship with opencv is over, CNN is my new best friend, 2x better well localisation than yeaterday
1
2
12
914
Friendship with opencv is over, CNN is my new best friend, 2x better well localisation than yeaterday
1
1
10
1,650
Daily progress #4 - declared victory too soon w/ opencv well detector, it struggles with colours - thx @ZECTBynmo @Willqyu @dnbt777, gave me great ideas to fix - testing bioreactor dispensing w/ liquid handler pipetting - ready to cook some yeast i think lads
2
10
607
manual data annotation... feeling extremely stupid doing this, there must be a better way
3
9
935
Daily progress #3 - fried gripper servo RIP - collecting some well images using colour mixing to make the opencv well detector more robust
1
9
483