ashu

20. doing gov stuff

[ashu @_devott]

@inferredbylisa @theakshsharma This is difficult, because each individual has different motives.
Scientists have immense curiosity in their field, engineers like to build, and entrepreneurs like to lead and solve problems. It’s very hard to be all three since you cannot truly subscribe to all 3 mindsets.

[ashu @_devott]

@alz_zyd_ you could say that for any sector actually. in general if people work very well together, things get done fast without many issues

its just that finance functions dont run well most times

[ashu @_devott]

@watling_samuel SCM uses various utility functions, they are heavily involved in math/ml

[ashu @_devott]

Pursuing true causality in a set might not be worth if your correlation matrix is already incredibly accurate. most times mimicking P(x|y) is significantly better using attention compared to P(x|do(y)) since its incredibly difficult to get right via dag based scms.

[ashu @_devott]

@4TristanS this is super fascinating, how do you make them so fast? different polymer?

[ashu @_devott]

@JimDMiller @ai_sentience im stealing this, this is a great statement

[ashu @_devott]

@mal_shaik just wait till actual sim results

[ashu @_devott]

at my current ingestion rate for gov data, i will run out storage. (1tb). 800 of those gigs are just data.

may have to scale to fast storage online, any recs?

[ashu @_devott]

@compileandpush @shcallaway which ones?

[ashu @_devott]

@justinskycak I’d agree to an extent, having foundational knowledge in something very niche can allow you to pattern recognise in a different way

building is great, but learning unorthodoxically with future application is arguably better

[ashu @_devott]

@N_and_ni they are

[ashu @_devott]

@ivanburazin what are you building?

[ashu @_devott]

@alameenpd thats amazing, very excited to see what you build

[ashu @_devott]

update: just got 40 more gb of training data over 4 hours

gonna start analysis soon

[ashu @_devott]

@mal_shaik things are moving

[ashu @_devott]

@AdrianDittmann u cannot compare the avg math guy to Ramunujan

[ashu @_devott]

@zekramu this very true, the skill of understanding where to save money and if u need to build, how to build effectively is heavily under-appreciated.

its execution and workflows that changed not rlly system design, which is demand

[ashu @_devott]

@juleslogs mechanistic interpretability is so fascinating, especially understanding emergent behaviour - reverse engineering grokking was so cool

[ashu @_devott]

@0xglitchbyte idk, rust is a very good language for perf, memory usage, and structure

not saying it should be used everywhere but for high perf systems.. id argue it should