Isaac Yonemoto is cooking

Prompt charmer, AI pharma founder, multiplicative manufacturing

video:
code:

[Isaac Yonemoto is cooking @DNAutics]

@usablejam @smothflegg @maxleiter @selladorre its implied there are other meat species, none intelligent.

[Max Leiter @maxleiter]

"They're made out of weights."

"Weights?"

"Weights. Floating-point numbers. We checked the whole thing through. It's nothing but weights."

"Weights doing what? Where do the words come from?"

"The weights make the words. Are you understanding me?"

[Isaac Yonemoto is cooking @DNAutics]

Elixir should have a Kernel.collect/3 function, which delegates to Enum.reduce, but with the argument order reversed. Kernel.collect/2 would be (collectible, enumerable) #elixirlang

[Isaac Yonemoto is cooking @DNAutics]

@thdxr We have an Enum.reduce so the first argument must be an enum.

[Isaac Yonemoto is cooking @DNAutics]

@OdedRechavi lots of cooler talk about which authors are bad actors too.

[Isaac Yonemoto is cooking @DNAutics]

@CasualVince @allie__voss ill say chicago architecture is probably #1 but its really close, nyc pulled out the high line and city island

[Isaac Yonemoto is cooking @DNAutics]

@nilslice @shritesh shritesh bhattarAI

[Isaac Yonemoto is cooking @DNAutics]

@CasualVince @allie__voss core problem is that anything interesting that chicago does as a big city some other city does it better. museums? dc. music? nola, parks? sf. e.g. ill give chicago commodities (boring) and publuc transit (we're in the usa)

[Isaac Yonemoto is cooking @DNAutics]

@Govindtwtt we have a vibecoded dna editor we use. DAU=2. company would not be where it is without it.

[Isaac Yonemoto is cooking @DNAutics]

@SimoneSyed @BecomingCritter fruits do, veg do not.

[Isaac Yonemoto is cooking @DNAutics]

for reference, before optimization the wasm was running in about 10s, and after, the js version runs in about 800ms for a typical workload that vidalalabs does.

[Isaac Yonemoto is cooking @DNAutics]

got the vibe-coded in-browser lalign down from 30s to sub 100ms. It also now does wrap-around alignments for circular sequences.

1. Tighten up the javascript. The code was doing string toUpper on each comparison! So step 1 was normalizing to integer enums and using UInt8Arrays

2. Tell the LLM to make the zig WASM engine match the new javascript lalign, aggressively testing example parity.

3. Since we're comparing sequences that SHOULD be mostly the same (sequential insertions/deletions, point mutations) it's worth it to attempt a first round chunked scan and that's SIMD-izable.

[Isaac Yonemoto is cooking @DNAutics]

@bside_bryan no i mean how has the quality of my 4.5 gone down? are they secretly routing me to 4.8?

[Isaac Yonemoto is cooking @DNAutics]

codex going through and fixing claude's utter nonsense attempts to fix CI.

[Isaac Yonemoto is cooking @DNAutics]

@bside_bryan I don't even have the option to use 4.6??

[Isaac Yonemoto is cooking @DNAutics]

@bside_bryan 4.5, its seems! I've been using 4.5 this whole time. But the quality has gone down?

[Isaac Yonemoto is cooking @DNAutics]

@jedisct1 seems like it was implemented by someone very skilled

[Isaac Yonemoto is cooking @DNAutics]

@SynBio1 start from the pre-1930 dataset and only add manually curated papers

[Isaac Yonemoto is cooking @DNAutics]

@eris_nerung it's like okcupid from way back when