Barak Bercovitz

[Barak Bercovitz @BarakBercovitz]

@ID_AA_Carmack IMO it's not the algorithms or compute (NNs/FLOPS) as much as it is the datasets. Current LLMs are downstream of >20 years of Internet and >10 years of social networks - all curating amazingly valuable data

[Barak Bercovitz @BarakBercovitz]

@ChShersh ADTs & pattern matching 😉

[Barak Bercovitz @BarakBercovitz]

@MatthewCroughan Unsupported by most standard security tools - therefore hard to use in production at big companies. Sec tools aren't designed to scan for CVEs in multiple copies of the same libs under /nix/store

[Barak Bercovitz @BarakBercovitz]

@DanAloni No, it works bc that's a thought experiment.

[Haz @diegohaz]

TIL you can render diagrams on GitHub using markdown 😱

Markdown / Preview

[Barak Bercovitz @BarakBercovitz]

@DanAloni https://t.co/qidab8ykaf

[Barak Bercovitz @BarakBercovitz]

@buggymcbugfix To me it means they'd recognize Math not specific implementations of Math ideas (C, Haskell). So Math is more universally understood which I guess is a good thing for problem solving.
So which one?
`class Animal` and `Cow` that "moo"s
Or
Monoid, Group, Functor, Monadn etc

[Barak Bercovitz @BarakBercovitz]

@christopherdone I disagree. Idioms in Java are invented and all about `class Animal` with a `noise` and `Cow` that "moo"s. Haskell idioms and APIs were (arguably) discovered decades ago in Math - Monoid, Group, Functor, Monad.

This "math thinking" is better at compositionality, correctness etc

[Barak Bercovitz @BarakBercovitz]

@DanAloni saying something's "always decades away" ignores progress. it's like saying neural nets are impractical in the 80s-90s-20s.
batteries are great, but there's a good reason why photosynthesis on earth doesn't depend on rare earth metals

[Barak Bercovitz @BarakBercovitz]

@DanAloni imo this is unlikely to be how things unfold in a few decades from now. batteries will always be important, but on a larger scale, other solutions as well such as - hydrogen, artificial photosynthesis, fusion (https://t.co/OYNR0Cpox5), etc.

[Barak Bercovitz @BarakBercovitz]

@TechnoEmpress IMO it depends on your use-case. Sometimes decoupling "language" from "implementation" in a narrow domain, is more important than "low-level" performance/complexity. I view valuable projects like https://t.co/up2uFPtGiu as (conceptually) close to

[Barak Bercovitz @BarakBercovitz]

@dooygoy @tidalcycles @nixos_org imo this is still a good option to package @tidalcycles and all of its deps. reproducibly. I remember having some difficulties installing/configuring SuperCollider, etc.

[Barak Bercovitz @BarakBercovitz]

Started playing with @tidalcycles in #VIM in @nixos_org (my config: https://t.co/XXhqsTv7gu, https://t.co/VPatWhyxNY)

[Barak Bercovitz @BarakBercovitz]

@istathar @tikhonjelvis yes for `pyfi` it was an issue, but we managed to use it with `pandas`/`sklearn`. for @tweagio 's HaskellR I haven't used it enough to say

[Barak Bercovitz @BarakBercovitz]

@hmemcpy With $$$ you'd want to be correct. But it's not life&death (hard real time, aviation, etc). Big fintech institutions have resources&time to pay for learning curve and invest long term.
Ask why there's so much ad-hoc, unreproducible, Python Data Science in notebooks everywhere

[Barak Bercovitz @BarakBercovitz]

@tukett @smdiehl Are you familiar with @unisonweb ?

[Barak Bercovitz @BarakBercovitz]

@DanAloni To me it's more like "Machine Learning" > "Systems prog." during COVID.. unsurprisingly when every junior DS gold-rushes to "cure cancer"

[Barak Bercovitz @BarakBercovitz]

@kerckhove_ts @angerman_io @brick_haskell Do you know of graphing support like https://t.co/3CgeKfWMzJ or https://t.co/Q41ErKHoXb?

[Barak Bercovitz @BarakBercovitz]

https://t.co/gaW3KH2lKi